Search

Alzheimer's disease (AD) is a prevalent neurodegenerative disease with no effective treatment. Efficient and rapid detection plays a crucial role in mitigating and managing AD progression. Deep learning-assisted smartphone-based microfluidic paper analysis devices (μPADs) offer the advantages of low cost, good sensitivity, and rapid detection, providing a strategic pathway to address large-scale disease screening in resource-limited areas. However, existing smartphone-based detection platforms usually rely on large devices or cloud servers for data transfer and processing. Additionally, the implementation of automated colorimetric enzyme-linked immunoassay (c-ELISA) on μPADs can further facilitate the realization of smartphone μPADs platforms for efficient disease detection. This paper introduces a new deep learning-assisted offline smartphone platform for early AD screening, offering rapid disease detection in low-resource areas. The proposed platform features a simple mechanical rotating structure controlled by a smartphone, enabling fully automated c-ELISA on μPADs. Our platform successfully applied sandwich c-ELISA for detecting the β-amyloid peptide 1–42 (Aβ 1–42, a crucial AD biomarker) and demonstrated its efficacy in 38 artificial plasma samples (healthy: 19, unhealthy: 19, N = 6). Moreover, we employed the YOLOv5 deep learning model and achieved an impressive 97 % accuracy on a dataset of 1824 images, which is 10.16 % higher than the traditional method of curve-fitting results. The trained YOLOv5 model was seamlessly integrated into the smartphone using the NCNN (Tencent's Neural Network Inference Framework), enabling deep learning-assisted offline detection. A user-friendly smartphone application was developed to control the entire process, realizing a streamlined "samples in, answers out" approach. This deep learning-assisted, low-cost, user-friendly, highly stable, and rapid-response automated offline smartphone-based detection platform represents a good advancement in point-of-care testing (POCT). Moreover, our platform provides a feasible approach for efficient AD detection by examining the level of Aβ 1–42, particularly in areas with low resources and limited communication infrastructure. The schematic diagram of the overview of an offline deep learning-assisted smartphone-based paper-based microfluidic platform for screening of Alzheimer's disease. [Display omitted] • Deep learning-assisted smartphone-based platform enables offline detection. • Fully automated microfluidic paper analysis devices (μPADs). • Smartphone-based platform for detection of biomarker of Alzheimer's disease. [ABSTRACT FROM AUTHOR]

In the production of paper cups in industrial factories, it is tried to print high quality cups with less waste loss with the help of sensors and heating resistances mounted on the paper cup machine. In this study, a system that detects faulty products based on image processing and removes it by controlling the machine with servo motors, asynchronous motors and programmable logic controller (PLC) is designed. For fault product detection, classification has been performed using real-time Haarcascade algorithm and You Only Look Once (YOLO) algorithm which is a deep learning methods, and real-time object detection has been carried out using the OpenCv library. With this study, an effective faulty product detection and removing hardware system was realized by adapting artificial intelligence algorithms to a machine used in industry. Based on the results, a whole system can be applied to systems that involve removing a faulty product from a band in any production, packaging etc. facility is proposed. A hardware consisting of servo motors, asynchronous motors and PLC was designed to separate faulty cups from the existing paper cup production machine in this study. Then, a data set composed of 1068 images was created with images taken from the camera for faulty and faultless paper cups. Using this dataset, the effect of different deep learning methods on performance in the real-time system has been examined and successful results have been obtained. The optimal outcome was achieved, yielding a real-time application accuracy rate of 90.8% through the utilization of the Yolov5x architecture. [ABSTRACT FROM AUTHOR]

• Five Machine Learning (ML) models compared for wave height time series prediction. • Complex ML models do not outperform simple AR in wave height time series prediction. • Comment to related papers: signal decomposition in test set series is WRONG. Significant Wave Height (SWH) is crucial in many aspect of ocean engineering. The accurate prediction of SWH has therefore been of immense practical value. Recently, Artificial Intelligence (AI) time series prediction methods have been widely used for single-point short-term SWH time-series forecasting, resulting in many AI-based models claiming to achieve good results. However, the extent to which these complex AI models can outperform traditional methods has largely been overlooked. This study compared five different models - AutoRegressive (AR), eXtreme Gradient Boosting (XGB), Artificial Neural Network (ANN), Long Short-Term Memory (LSTM), and WaveNet - for their performance on SWH time series prediction at 16 buoy locations. Surprisingly, the results suggest that the differences of performance among different models are negligible, indicating that all these AI models have only "learned" the linear auto-regression from the data. Additionally, we noticed that many recent studies used signal decomposition method for such time series prediction, and most of them decomposed the test sets, which is WRONG. [ABSTRACT FROM AUTHOR]

• Qualitative and quantitative detection of amine gases could be achieved by CSA. • A visible detection of beef freshness using the amine-responsive CSA was proposed. • ResNet34 had the best performance for beef freshness detection based on CSA. • T-SNE could further visualize and understand the classification process of DL. This study developed an innovative approach that combines a colourimetric sensor array (CSA) composed of twelve pH-response dyes with advanced algorithms, aiming to detect amine gases and assess the freshness of chilled beef. With the assistance of multivariate statistical analysis, the sensor array can effectively distinguish five amine gases and enable rapid quantification of trimethylamine vapour with a limit of detection (LOD) of 8.02 ppb and visually monitor the fresh levels of chilled beef. Moreover, the utilization of deep learning models (ResNet34, VGG16, and GoogleNet) for chilled beef freshness evaluation achieved an overall accuracy of 98.0 %. Furthermore, t -distributed stochastic neighbour embedding (t -SNE) visualized the feature extraction process and provided explanations to understand the classification process of deep learning. The results demonstrated that applying deep learning techniques in the process of pattern recognition of CSA can help in realizing the rapid, robust, and accurate assessment of chilled beef freshness. [ABSTRACT FROM AUTHOR]

Offshore wind-hydrogen systems operate in harsh marine environments for extended periods, posing risks of low accessibility and high failure rates. This paper proposes a data-driven fault detection framework for offshore wind-hydrogen systems, aiming to promptly detect potential faults based on the system's physical parameters. The research establishes a comprehensive model based on the operating principles of offshore wind-hydrogen systems. The study uses fault injection techniques to simulate common partial faults in offshore wind-hydrogen systems and collects relevant physical parameters (such as wind turbine power generation, energy storage battery charging/discharging current, and electrolyzer hydrogen production rate). A fault detection model was constructed using a combination of Convolutional Neural Networks (CNN) and Bidirectional Long Short-Term Memory (BiLSTM) to enable precise detection of potential faults in the system. The established model provides theoretical support for designing, optimizing, and operating offshore wind-hydrogen systems, reducing reliance on actual experiments and lowering project risks and costs. Introducing artificial faults can evaluate the system's performance in different fault scenarios. This paper proposes a CNN-BiLSTM fault detection method that achieves automatic feature extraction and high-precision classification of time-series fault data, confirming the practicality and feasibility of offshore wind-hydrogen systems. This investigation realizes the integration of the offshore wind-hydrogen system model and deep learning, as well as a fully automated learning process from data to fault detection. By calculating and comparing results, it is verified that the framework has a high accuracy of 96.9%. This innovative research offers new methods and perspectives for enhancing the reliability and efficiency of fault detection in offshore wind-hydrogen systems. • A comprehensive multi-system simulation model for offshore wind-hydrogen systems is established. • Fault injection is employed to acquire data samples for offshore wind-hydrogen systems. • Utilizing the CNN-BiLSTM method, a high-accuracy deep learning fault detection model is crafted. • A simulation model-based data-driven deep learning framework for fault detection is devised in this study. [ABSTRACT FROM AUTHOR]

3D printed ceramic parts often have defects due to their inherent brittleness. These defects are of various scale sizes. Detecting these defects, especially with multiple sizes, is a challenging task in the field of detection. Furthermore, there is a lack of surface defect real-time detection systems suitable for industrial applications in this field. Based on this, this paper makes two contributions: the design and development of a real-time defect detection system, and the proposal of a multiscale defect detection method for 3D printed ceramic surfaces within this system. Firstly, the paper establishes the overall structure of the real-time detection system for surface defects on 3D printed ceramic components, and describes the hardware and software components of the system. On this basis, a dataset of ceramic surface defect images is collected and constructed. Then, experimental analyses point out the shortcomings of the You Only Look Once version-5 (YOLOv5) model for multiscale defect detection. To address the shortcomings, the YOLOv5 model is optimized from three aspects, resulting in the Deep separable convolution + residual network-SKNetwork-Efficient Channel Attention Network-YOLOv5 (DepRes-SK-ECA-YOLOv5) model for multiscale defect detection. This model improves the ability to extract and fuse features of defects at different scales. The experimental results show that the DepRes-SK-ECA-YOLOv5 model can achieve 93.5 %, 91.6 %, 94.3 %, and 0.198 s for Precision, Recall, mAP, and Speed for the test set, respectively. Finally, the paper designs the workflow for the system software. The system hardware and system software are integrated to form a real-time detection system. The performance of the detection system is verified through experiments. [ABSTRACT FROM AUTHOR]

Advances in artificial intelligence (AI) and automated robotics will profoundly influence space operations. By utilising machine learning and deep learning approaches, AI-enabled systems may accomplish tasks as well as improve their own performance. These capabilities are useful in the often-remote settings of outer space and will grow in value as automated space operations become more widespread. As AI extends throughout the space domain, automated algorithms will take on many of the roles that have historically been handled by humans. Artificial intelligence is progressing from theory to implementation in the space environment by exposing new satellites and orbital autonomous vehicles to new data. Even though all initial computational parameters are provided, such systems' outputs can be very unpredictable, putting people, property, and the environment at risk. This paper investigates the application of United Nations space treaties, selected regional AI regulations, and various 'soft-law' instruments and industry initiatives focusing on responsible AI system development to space-based AI systems. Following that, reforms are proposed to clarify the practical relationship between AI systems and the international legal regime that governs space, as well as a 'bottom-up' regulatory approach to better facilitate the future development of regulation governing the use of AI by the global space sector. While this work does not purport to provide a conclusive resolution to these multifaceted matters, its objective is to underscore significant obstacles that arise at the convergence of space law and AI, serving as a preliminary foundation for subsequent discussions on this issue. • Advances in AI and automated robotics will have a profound impact on space operations. Automated algorithms will take on roles traditionally handled by humans as AI becomes more widespread in space. • However, the unpredictable outputs of AI systems can put people, property, and the environment at risk, raising questions about liability. • The paper investigates the application of UN space treaties, regional AI regulations, and industry initiatives to space-based AI systems. • Reforms are proposed to clarify the relationship between AI systems and the international legal regime governing space. • A 'bottom-up' regulatory approach is suggested to facilitate future regulation of AI in the global space sector. [ABSTRACT FROM AUTHOR]

Improving generalization ability in multi-robot formation can reduce repetitive training and calculation. In this paper, we study the multi-robot formation problem with the ability to generalize the target position. Since the generalization ability of neural network is directly proportional to spatial dimension, we adopt the strategy of using different networks to solve different objectives, so that the network learning can focus on the learning of one objective to obtain better performance. In addition, this paper presents a distributed deep reinforcement learning method based on soft actor–critic algorithm for solving multi-robot formation problem. At the same time, the formation evaluation assignment function is designed to adapt to distributed training. Compared with the original algorithm, the improved algorithm can get higher reward cumulative values. The experimental results show that the proposed algorithm can better maintain the desired formation in the moving process, and the rotation design in the reward function makes the multi-robot system have better flexibility in formation. The comparison of control signal curve shows that the proposed algorithm is more stable. At the end of the experiments, the universality of the proposed algorithm in formation maintenance and formation variations is demonstrated. [ABSTRACT FROM AUTHOR]

Emotional intelligence (EI) involves recognizing, regulating, and understanding emotions in oneself and others, enabling effective relationship management. Emotional Intelligence acts as a differentiator in both personal and professional lives. Nevertheless, many people lack the ability of conveying and interpreting human emotions efficiently due to which they fall short in their social and professional lives. However, this can be improved if people know their Emotional Quotient (EQ) score and work towards improving it by implementing certain skills. Although traditional methods for measuring EQ exist, majority of them are manual and are built on self-efficacy based questionnaire which are often susceptible to faking. In order to overcome these lacuna, this paper proposes a digital approach that can be used for an automatic evaluation of EQ of a person via audio visual signal input. The paper devises four major factors self-awareness, self-regulation, self-motivation and social relationships with their weightages that can be utilised for automatic evaluation through the use of technology. The paper also proposes quantifiable parameters for measuring these factors. The proposed criteria were developed with the emergence of deep learning and its applicability in assessing a person's EQ in mind. [ABSTRACT FROM AUTHOR]

The integration of deep learning techniques from the field of image generation into digital rock analysis has resulted in substantial advancements. However, generating high-quality and highly controllable heterogeneous digital rocks still presents challenges and necessities. This paper reports a digital rock reconstruction method using a multi-condition denoising diffusion probabilistic model (MCDDPM). This method has the diffusion and denoising processes inherent in the diffusion model, ensuring high-quality reconstructed samples. Meanwhile, various physical property parameters are embedded as conditions in the network layers of the model at multiple scales, enhancing the conditional control ability of this method. We compared the porosity, pore size distribution, and two-point correlation coefficient under the same conditions, as well as the evaluation indicators of different methods. The comparative results provide convincing evidence for the accuracy of our proposed digital rock reconstruction method. In addition, studies on the control effect of different conditions on reconstructed samples also demonstrate the effectiveness of our strategy. Therefore, MCDDPM can reconstruct reasonable heterogeneous digital rocks based on combinations of physical property conditions, which provides effective technical support for expanding the library of heterogeneous digital rock samples and even achieving zero-shot learning of digital rocks. • The real numerical rock possesses strong heterogeneity, and the method proposed in this paper can effectively reconstruct heterogeneous digital rock. • Compared to other deep learning network models, the reconstruction effect of the method proposed in this paper is at a high level. • The method we propose supports multiple physical property conditions as input. [ABSTRACT FROM AUTHOR]

Accurate cloud and shadow detection is a crucial prerequisite for optical remote sensing image analysis and application. Multi-temporal-based cloud and shadow detection methods are a preferable choice to detect clouds in complex scenes (e.g., thin clouds, broken clouds and clouds with interference from artificial surfaces with high reflectivity). However, such methods commonly require cloud-free reference images, and this may be difficult to achieve in time-series data since clouds are often prevalent and of varying spatial distribution in optical remote sensing images. Furthermore, current multi-temporal-based methods have limited feature extraction capability and rely heavily on prior assumptions. To address these issues, this paper proposes a Siamese network (Siamnet) for cloud and shadow detection based on Time-Series cloudy Images, namely TSI-Siamnet, which consists of two steps: 1) low-rank and sparse component decomposition of time-series cloudy images is conducted to construct a composite reference image to cope with dynamic changes in the cloud distribution in time-series images; 2) an extended Siamnet with optimal difference calculation module (DM) and multi-scale difference features fusion module (MDFM) is constructed to extract reliable disparity features and alleviate semantic information feature dilution during the decoder part. TSI-Siamnet was tested extensively on seven land cover types in the well-known Landsat 8 Biome dataset. Compared to six state-of-the-art methods (including four deep learning-based methods and two classical non-deep learning-based methods), TSI-Siamnet produced the best performance with an overall accuracy of 95.05% and MIoU of 84.37%. In three more challenging experiments, TSI-Siamnet showed enhanced detection of thin and broken clouds and greater anti-interference to highly reflective surfaces. TSI-Siamnet provides a novel strategy to explore comprehensively the valid information in time-series cloudy images and integrate the extracted spectral-spatial–temporal features for reliable cloud and shadow detection. [ABSTRACT FROM AUTHOR]

Cameras are rapidly becoming the choice for on-board sensors towards space rendezvous due to their small form factor and inexpensive power, mass, and volume costs. When it comes to docking, however, they typically serve a secondary role, whereas the main work is done by active sensors such as lidar. This paper documents the development of a proposed AI-based (artificial intelligence) navigation algorithm intending to mature the use of on-board visible wavelength cameras as a main sensor for docking and on-orbit servicing (OOS), reducing the dependency on lidar and greatly reducing costs. Specifically, the use of AI enables the expansion of the relative navigation solution towards multiple classes of scenarios, e.g., in terms of targets or illumination conditions, which would otherwise have to be crafted on a case-by-case manner using classical image processing methods. Multiple convolutional neural network (CNN) backbone architectures are benchmarked on synthetically generated data of docking manoeuvres with the International Space Station (ISS), achieving position and attitude estimates close to 1 % range-normalised and 1 deg, respectively, an established rule of thumb for the navigation measurement accuracy during final approach. The integration of the solution with a physical prototype of the refuelling mechanism is validated in laboratory using a robotic arm to simulate a berthing procedure. • A novel AI-based solution for spacecraft docking and refuelling is presented. • First demonstration of deep learning for vision-based pose estimation in docking. • Various CNN backbones are benchmarked on photorealistic ISS refuelling datasets. • ResNet-101 reports average 1.19% error in position, 0.29 ∘ in attitude per trajectory. • Prototypes of the developed mechanism are validated in lab with 7-DOF manipulator. [ABSTRACT FROM AUTHOR]

Recently, computer vision and artificial intelligence are being used as enabling technologies for plant phenotyping studies, since they allow the analysis of large amounts of data gathered by the sensors. Plant phenotyping studies can be devoted to the evaluation of complex plant traits either on the aerial part of the plant as well as on the underground part, to extract meaningful information about the growth, development, tolerance, or resistance of the plant itself. All plant traits should be evaluated automatically and quantitatively measured in a non-destructive way. This paper describes a novel approach for identifying plant roots from images of the root system architecture using a convolutional neural network (CNN) that operates on small image patches calculating the probability that the center point of the patch is a root pixel. The underlying idea is that the CNN model should embed as much information as possible about the variability of the patches that can show chaotic and heterogeneous backgrounds. Results on a real dataset demonstrate the feasibility of the proposed approach, as it overcomes the current state of the art. • Root systems must be monitored to assess the growth and well-being of a plant. • State-of-the-art approaches mainly use classic ML or U-networks for segmentation. • CNNs can be used for monitoring considering patch-based information. • These models are simpler and faster, and provide better segmentation performance. [ABSTRACT FROM AUTHOR]

The accidental leakage of gaseous hydrogen from hydrogen refueling station facilities has the potential to result in a large fire and explosion catastrophe. The ability to forecast the spatiotemporal evolution of hydrogen plumes in real-time is crucial for monitoring the distribution of plume concentrations and mitigating the risk of fire and explosion. The utilization of deep learning has been extensively employed in a diverse range of spatiotemporal forecasting tasks. However, these approaches encounter limitations in accurately quantifying uncertainty when predicting spatiotemporal concentrations and plume boundaries. This research paper introduces a novel model called DPSTNN_H 2 Evolution, which is a deep probabilistic spatial-temporal neural network designed for forecasting the spatiotemporal evolution of hydrogen plumes. The benchmark dataset is constructed by the implementation of numerical simulations pertaining to the inadvertent leak of hydrogen within a hydrogen refueling station. The findings of the study indicate that incorporating quantified uncertainty information might enhance the precision of plume boundaries and the resilience of plume concentrations when predicting the spatiotemporal development of hydrogen. By employing Monte Carlo sampling with a sample size of m = 100 and a dropout rate of p = 0.1 , the model demonstrates the ability to provide real-time inference for 10 instances of plumes, while maintaining a high level of accuracy with R2 = 0.9829. In comparison to the current state-of-the-art model, the proposed model demonstrates superior accuracy and robustness in forecasting the spatiotemporal evolution of hydrogen plumes. In general, our proposed model has the potential to serve as a dependable option for the development of a digital twin for emergency management of hydrogen refueling stations. • Probabilistic spatial-temporal neural network model for hydrogen plume evolutions is proposed. • Normalized uncertainty contours and uncertainty intervals to improve accuracy and robustness. • Numerical experiment of hydrogen leakage and dispersion in refueling station are conducted. • Optimal hyper-parameters of the proposed model are determined • Comparison between the proposed model with point-estimation based model is conducted [ABSTRACT FROM AUTHOR]

Accurate and swift estimation of the future performance degradation of Proton Exchange Membrane Fuel Cells (PEMFCs) is crucial for the health management of PEMFC. However, most existing studies primarily concentrate on uncovering the inherent patterns of performance degradation parameters, neglecting the fundamental relationships and pattern variations between performance degradation and other characteristic parameters (such as flow rate, temperature, and pressure). This paper seeks to address this research gap. To that end, we delve into feature parameters, extracting underlying patterns of feature evolution to construct a feature exploration-based framework for predicting PEMFC performance degradation. Initially, Time-Varying Filtered Empirical Mode Decomposition (TVF-EMD) is employed to decompose each feature variable, isolating the core information of the features. Furthermore, to reduce the computational burden caused by redundant features, Uniform Manifold Approximation and Projection (UMAP) is used to perform dimensionality reduction on the decomposed feature subsequences, ensuring the efficiency and representativeness of the extracted features. Finally, to overcome the shortcomings of long short-term memory networks (LSTM) in terms of convergence and the effective use of time series information, we introduce an innovative Mogrifier LSTM (M-LSTM) model to enhance the accuracy of PEMFC performance degradation predictions. The experimental and validation results demonstrate that the extraction of feature parameters significantly enhances the predictive precision of the baseline model. Relative to the compared models, there was an improvement of at least 16.859% in Root Mean Square Error (RMSE), a minimum increase of 23.333% in Mean Absolute Error (MAE), and a boost of no less than 0.913% in the coefficient of determination (R 2 ). Simultaneously, the proposed predictive framework demonstrates good stability and accuracy in predicting PEMFC performance degradation. • Developed a novel feature processing framework for predicting the performance degradation of PEMFC. • Employed TVF-EMD in conjunction with UMAP to extract evolving trends and cyclical patterns of feature parameters. • Introduced an innovative M-LSTM network to enhance the interaction between historical information and current inputs. • Feature processing framework improves single-model performance degradation predictions [ABSTRACT FROM AUTHOR]

Prior research has associated spoken language use with depression, yet studies often involve small or non-clinical samples and face challenges in the manual transcription of speech. This paper aimed to automatically identify depression-related topics in speech recordings collected from clinical samples. The data included 3919 English free-response speech recordings collected via smartphones from 265 participants with a depression history. We transcribed speech recordings via automatic speech recognition (Whisper tool, OpenAI) and identified principal topics from transcriptions using a deep learning topic model (BERTopic). To identify depression risk topics and understand the context, we compared participants' depression severity and behavioral (extracted from wearable devices) and linguistic (extracted from transcribed texts) characteristics across identified topics. From the 29 topics identified, we identified 6 risk topics for depression: 'No Expectations', 'Sleep', 'Mental Therapy', 'Haircut', 'Studying', and 'Coursework'. Participants mentioning depression risk topics exhibited higher sleep variability, later sleep onset, and fewer daily steps and used fewer words, more negative language, and fewer leisure-related words in their speech recordings. Our findings were derived from a depressed cohort with a specific speech task, potentially limiting the generalizability to non-clinical populations or other speech tasks. Additionally, some topics had small sample sizes, necessitating further validation in larger datasets. This study demonstrates that specific speech topics can indicate depression severity. The employed data-driven workflow provides a practical approach for analyzing large-scale speech data collected from real-world settings. • Transcribed smartphone-collected speech recordings via automatic speech recognition • Utilized a deep learning topic model for identifying topics from transcriptions • Identified six risk topics associated with high depression symptom severity • Explored the contexts of topics using behavioral and linguistic characteristics [ABSTRACT FROM AUTHOR]

Feature extraction and matching play integral roles in autonomous vision navigation. As a typical morphological feature, craters can be used as navigation landmarks. However, the spin of small bodies introduces substantial variations in illumination and viewing angles within acquired images, thereby posing challenges to feature extraction and matching algorithms. Addressing these challenges, a deep learning-based approach emerges as pivotal in mitigating the limitations of existing navigation algorithms. This paper presents an algorithm designed to segment noteworthy craters and irregular regions on such small bodies, thereby furnishing crucial autonomous navigation data for missions involving uncharted terrains. By seamlessly integrating deep learning techniques with conventional computer vision-based feature description methods, the algorithm achieves matching segmentation results of noteworthy precision. The experimental results show that the algorithm has better results compared to the existing mainstream segmentation networks. And it can achieve the accurate matching of segmentation results. • Accurate segmentation of small-body craters via deep-learning Algorithms. • Gradient descriptor construction method based on segmentation results. • The proposed method has viewpoint, illumination, scale, and rotation invariance. [ABSTRACT FROM AUTHOR]

Although current semantic segmentation approaches have achieved impressive performance, their ability to incrementally learn new classes is limited. Moreover, pixel-by-pixel annotations are costly and time-consuming. Therefore, a new field called Weakly-supervised Incremental Learning for Semantic Segmentation (WILSS) has emerged, which learns new classes using image-level labels. However, image-level labels do not provide sufficient detail, and we discover that the state-of-the-art of WILSS suffers from confusion between old knowledge and new knowledge. To address this issue, we propose W eakly-supervised I ncremental learning for S emantic segmentation with Class H ierarchy (WISH), a method that considers the hierarchical structure of each class when determining which knowledge to trust in cases of confusion between old and new knowledge. Our method has achieved new state-of-the-art performances in all settings compared to the previous methods on the Pascal VOC and MS COCO datasets. • Our paper introduces WISH, utilizing hierarchy and image labels for class incremental learning in semantic segmentation. • We efficiently utilize hierarchy, boosting segmentation performances without added costs. • Our method outperforms existing methods in all configurations, showcasing hierarchical integration's effectiveness. [ABSTRACT FROM AUTHOR]

In this paper, we address the image denoising problem in presence of speckle degradation typically arising in ultra-sound images. Variational methods and Convolutional Neural Networks (CNNs) are considered well-established methods for specific noise types, such as Gaussian and Poisson noise. Nonetheless, the advances achieved by these two classes of strategies are limited when tackling the de-speckle problem. In fact, variational methods for speckle removal typically amounts to solve a non-convex functional with the related issues from the convergence viewpoint; on the other hand, the lack of large datasets of noise-free ultra-sound images has not allowed the extension of the state-of-the-art CNN denoiser methods to the case of speckle degradation. Here, we aim at combining the classical variational methods with the predictive properties of CNNs by considering a weighted total variation regularized model; the local weights are obtained as the output of a statistically inspired neural network that is trained on a small and composite dataset of natural and synthetic images. The resulting non-convex variational model, which is minimized by means of the Alternating Direction Method of Multipliers (ADMM) is proven to converge to a stationary point. Numerical tests show the effectiveness of our approach for the denoising of natural and satellite images. • An hybrid strategy for the despeckling problem combining convolutional neural networks and variational methods is proposed. • The proposed neural architecture is statistically-inspired and not requires the training on a large data-set of images. • A proof of convergence to a stationary point for the alternating direction method is given. [ABSTRACT FROM AUTHOR]

Timely and effective fault diagnosis is crucial to guarantee the reliability and durability of proton exchange membrane fuel cell (PEMFC). Segmented cell technology has been proven to be a powerful tool for exploring the internal characteristics of fuel cells, yet it is seldom applied for fault diagnosis. This paper proposes a data-driven fault diagnosis method that combines segmented cell and deep learning technology for PEMFC water management fault diagnosis. The dynamic characteristics of the current distribution of the fuel cell under different levels of faults are explored with segmented cell technology. A dual-input convolutional neural network is applied to combine current distribution data and sensor data for a more comprehensive system status for fault diagnosis. The results show that the model can efficiently identify flooding and drying faults with more than 98.5% accuracy, faster and more accurately than other algorithms. This work highlights the significance of segmented cell technology for the comprehension of water management inside PEMFC and provides a new concept for PEMFC fault diagnosis. [Display omitted] • The segmented cell technology is applied to investigate the current distribution during flooding and drying process. • A dual-input CNN is adopted to establish a fault diagnosis model for PEMFC water management. • The proposed method exhibits excellent diagnostic performance, with an accuracy rate of over 98.5%. [ABSTRACT FROM AUTHOR]

The large number of near-Earth asteroids (NEAs) has greatly impacted human space activities and Earth security. However, detecting NEAs in astronomical images with complex, varying backgrounds is still extremely challenging. In this paper, we propose a deep segmentation assisted asteroid tracking algorithm, termed DSAT, to construct a possible pipeline for faint NEA tracking in astronomical images. First, the single-frame object detection problem is converted to a segmentation problem, enabling robust extraction of faint potential moving objects. Then, a multiframe motion prior-based moving object tracking algorithm is proposed to find real NEAs. We further propose a distance tolerance criterion to help DSAT achieve effective tracking in practical situations when detection has partially failed. Finally, the pipeline is tested with both simulated and real astronomical images at different SNRs and in crowded fields. The results showed that our pipeline has the potential to detect and track faint NEAs in complex backgrounds. Our code is publicly available at https://github.com/zhenhongdu/DeepSegAsteroidTracker. [ABSTRACT FROM AUTHOR]

The paper explores the tensile behaviour of plain weave ceramic matrix composites (CMCs) by employing a nonlinear constitutive model for fibre tow composites, driven by in-situ X-ray computed tomography data. Through the application of a deep-learning-based image segmentation method, the study successfully identified material damage modes, including matrix cracks, fibre pull-out, and interface debonding. Three-dimensional characterisation and quantitative analyses of damage have been conducted. At the mesoscopic level, a theoretically derived nonlinear constitutive model for fibre tow composites, accounting for continuous damage to both fibre and matrix, is established by using damage data. At the macroscopic level, a unit cell model is constructed for the numerical simulation of plain weave CMCs. Material's initial defects are prone to induce crack initiation. External matrix cracks can propagate to tows, and crack deflection may occur during this process. Furthermore, adjacent microcracks have the potential to connect, forming a larger one under high stress. Theoretical results predicted by the nonlinear constitutive model for fibre tow composites under tension exhibit good agreement with test data. The built unit cell model accurately predicts the tensile mechanical behaviour of plaine weave CMCs. [ABSTRACT FROM AUTHOR]

• Introduced the Siamese 4D-Alznet, which combines five distinct blocks of CNNs. • We proposed three custom TL models incorporating frozen and replacing layers. • Our methods prove 95.07% binary classification performance in terms of accuracy. • All techniques performed best for NC, EMCI, LMCI, and AD classification. Alzheimer's disease (AD) is the general form of dementia, leading to a progressive neurological disorder characterized by memory loss due to brain cell damage. Artificial Intelligence (AI) assists in the early identification and prediction of AD patients, determining future risks and benefits for radiologists and doctors to save time and cost. Since deep learning (DL) approaches work well with massive datasets and have recently become helpful for AD detection, there remains an area for improvement in automating detection performance. Present approaches somehow addressed the challenges of limited annotated data samples for binary classification. This contrasts with prior state-of-the-art techniques, which were constrained by their incapacity to capture abstract-level information. In this paper, we proposed a Siamese 4D-AlzNet model comprised of four parallel convolutional neural network (CNN) streams (Five CNN layer blocks) and customized transfer learning models (Frozen VGG-19, Frozen VGG-16, and customized AlexNet). Siamese 4D-AlzNet was vertically and horizontally stored, and the spatial features were passed to the final layer for classification. For experiments, T1-weighted MRI images comprised of four distinct subject classes, normal control (NC), mild cognitive impairment (MCI), late mild cognitive impairment (LMCI), and AD, have been employed. Our proposed models achieved outstanding accuracy, with a remarkable 95.05% accuracy distinguishing between normal and AD subjects. The performance across remaining binary class pairs consistently exceeded 90%. We thoroughly compared our model with the latest methods using the same dataset as our reference. Our proposed model improved NC-AD and MCI-AD classification accuracy by 2% 7%. [ABSTRACT FROM AUTHOR]

Unmanned Aerial Vehicle (UAV) has seen a dramatic rise in popularity for remote-sensing image acquisition and analysis in recent years. It has brought promising results in low-altitude monitoring tasks that require detailed visual inspections. Semantic segmentation is one of the hot topics in UAV remote sensing image analysis, as its capability to mine contextual semantic information from UAV images is crucial for achieving a fine-grained understanding of scenes. However, in the remote sensing field, recent reviews have not focused on combining "UAV remote sensing" and "semantic segmentation" to summarize the advanced works and future trends. In this study, we focus primarily on describing various recent semantic segmentation methods applied in UAV remote sensing images and summarizing their advantages and limitations. According to the distinction in modeling contextual semantic information, we have categorized and outlined the methods based on graph-based contextual models and deep-learning-based models. Publicly available UAV-based image datasets are also gathered to encourage systematic research on advanced semantic segmentation methods. We provide quantitative results of representative methods on two high-resolution UAV-based image datasets for fair comparisons and discussions in terms of semantic segmentation accuracy and model inference efficiency. Besides, this paper concludes some remaining challenges and future directions in semantic segmentation for UAV remote sensing images and points out that methods based on deep learning will become the future research trend. • Summarize semantic segmentation methods and datasets for UAV remote sensing images. • Provide fair comparisons of representative methods in accuracy and efficiency. • Identify challenges and future directions for UAV image semantic segmentation. • Semantic segmentation methods based on deep learning will be the mainstream. [ABSTRACT FROM AUTHOR]

In PROPELLER MRI, obtaining sufficient high-quality blade data remains a challenge, so the efficiency and generalization of deep learning-based reconstruction models are deteriorated. Due to narrow rotated and translated blades acquired in PROPELLER, the technique of data augmentation that is used for deep learning-based Cartesian MRI reconstruction cannot be directly applied. To address the issue, this paper introduces a novel approach for the generation of synthetic PROPELLER blades, and it is subsequently employed in data augmentation for undersampled blades reconstruction. The principal aim of this study is to address the challenges of reconstructing undersampled blades to enhance both image quality and computational efficiency. Evaluation metrics including PSNR, NMSE, and SSIM indicate superior performance of the model trained with augmented data compared to non-augmented counterparts. The synthetic blade augmentation significantly enhances the model's generalization capability and enables robust performance across varying imaging conditions. Furthermore, the study demonstrates the feasibility of utilizing synthetic blades exclusively in the training phase, suggesting a reduced dependency on real PROPELLER blades. This innovation in synthetic blade generation and data augmentation technique contributes to enhanced image quality and improved generalization capability of the associated deep learning model for PROPELLER MRI reconstruction. [ABSTRACT FROM AUTHOR]

While deep neural networks have shown impressive results in automatic speaker recognition and related tasks, it is dissatisfactory how little is understood about what exactly is responsible for these results. Part of the success has been attributed in prior work to their capability to model supra-segmental temporal information (SST), i.e., learn rhythmic-prosodic characteristics of speech in addition to spectral features. In this paper, we (i) present and apply a novel test to quantify to what extent the performance of state-of-the-art neural networks for speaker recognition can be explained by modeling SST; and (ii) present several means to force respective nets to focus more on SST and evaluate their merits. We find that a variety of CNN- and RNN-based neural network architectures for speaker recognition do not model SST to any sufficient degree, even when forced. The results provide a highly relevant basis for impactful future research into better exploitation of the full speech signal and give insights into the inner workings of such networks, enhancing explainability of deep learning for speech technologies. • Literature explains speaker recognition in neural nets by modeling of voice dynmaics. • Diagnostic: We quantify how well deep learning models actually capture dynamics. • Observation: State-of-the-art deep nets do not model speaker prosody but ignore it. • Interpretation as "cheating": Achieving high without putting in due effort. • Outlook: Increasing task difficulty biases models towards prosody, but not enough. [ABSTRACT FROM AUTHOR]

Cancer is a condition in which abnormal cells uncontrollably split and damage the body tissues. Hence, detecting cancer at an early stage is highly essential. Currently, medical images play an indispensable role in detecting various cancers; however, manual interpretation of these images by radiologists is observer-dependent, time-consuming, and tedious. An automatic decision-making process is thus an essential need for cancer detection and diagnosis. This paper presents a comprehensive survey on automated cancer detection in various human body organs, namely, the breast, lung, liver, prostate, brain, skin, and colon, using convolutional neural networks (CNN) and medical imaging techniques. It also includes a brief discussion about deep learning based on state-of-the-art cancer detection methods, their outcomes, and the possible medical imaging data used. Eventually, the description of the dataset used for cancer detection, the limitations of the existing solutions, future trends, and challenges in this domain are discussed. The utmost goal of this paper is to provide a piece of comprehensive and insightful information to researchers who have a keen interest in developing CNN-based models for cancer detection. [ABSTRACT FROM AUTHOR]

Large deep learning models are impressive, but they struggle when real-time data is not available. Few-shot class-incremental learning (FSCIL) poses a significant challenge for deep neural networks to learn new tasks from just a few labeled samples without forgetting the previously learned ones. This setup can easily leads to catastrophic forgetting and overfitting problems, severely affecting model performance. Studying FSCIL helps overcome deep learning model limitations on data volume and acquisition time, while improving practicality and adaptability of machine learning models. This paper provides a comprehensive survey on FSCIL. Unlike previous surveys, we aim to synthesize few-shot learning and incremental learning, focusing on introducing FSCIL from two perspectives, while reviewing over 30 theoretical research studies and more than 20 applied research studies. From the theoretical perspective, we provide a novel categorization approach that divides the field into five subcategories, including traditional machine learning methods, meta learning-based methods, feature and feature space-based methods, replay-based methods, and dynamic network structure-based methods. We also evaluate the performance of recent theoretical research on benchmark datasets of FSCIL. From the application perspective, FSCIL has achieved impressive achievements in various fields of computer vision such as image classification, object detection, and image segmentation, as well as in natural language processing and graph. We summarize the important applications. Finally, we point out potential future research directions, including applications, problem setups, and theory development. Overall, this paper offers a comprehensive analysis of the latest advances in FSCIL from a methodological, performance, and application perspective. • In-depth survey on FSCIL methods, applications, performance. • Categorizing FSCIL into five subcategories for clear analysis. • Evaluating FSCIL research on benchmarks for strengths, weaknesses. • FSCIL applications in computer vision, NLP, and graph. [ABSTRACT FROM AUTHOR]

Deep neural networks have achieved outstanding performance in computer vision tasks. Convolutional Neural Networks (CNNs) typically operate in the spatial domain with raw images, but in practice, images are usually stored and transmitted in their compressed representation where JPEG is one of the most widely used encoder. Also, these networks are computationally intensive and slow. This paper proposes performing the learning and inference processes in the compressed domain in order to reduce the computational complexity and improve the speed of popular CNNs. For this purpose, a novel graph-based frequency channel selection method is proposed to identify and select the most important frequency channels. The computational complexity is reduced by retaining the important frequency components and discarding the insignificant ones as well as eliminating the unnecessary layers of the network. Experimental results show that the modified ResNet-50 operating in the compressed domain is up to 70% faster than the spatial-based traditional ResNet-50 while resulting in similar classification accuracy. Moreover, this paper proposes a preprocessing step with partial encoding to improve the resilience to distortions caused by low-quality encoded images. Finally, we show that training a network with highly compressed data can achieve a good classification accuracy with up to 93% reduction in the storage requirements of the training data. [ABSTRACT FROM AUTHOR]

Despite the wide variety of methods developed for synthetic image attribution, most of them can only attribute images generated by models or architectures included in the training set and do not work with unknown architectures, hindering their applicability in real-world scenarios. In this paper, we propose a verification framework that relies on a Siamese Network to address the problem of open-set attribution of synthetic images to the architecture that generated them. We consider two different settings. In the first setting, the system determines whether two images have been produced by the same generative architecture or not. In the second setting, the system verifies a claim about the architecture used to generate a synthetic image, utilizing one or multiple reference images generated by the claimed architecture. The main strength of the proposed system is its ability to operate in both closed and open-set scenarios so that the input images, either the query and reference images, can belong to the architectures considered during training or not. Experimental evaluations encompassing various generative architectures such as GANs, diffusion models, and transformers, focusing on synthetic face image generation, confirm the excellent performance of our method in both closed and open-set settings, as well as its strong generalization capabilities. • New verification framework for open-set architecture attribution of synthetic images. • Tested with several types of generative architectures in closed and open set scenarios. • Generalization tests prove that the system can verify unknown models of the architecture. • Outperforms state-of-the-art methods for open-set architecture attribution with rejection. [ABSTRACT FROM AUTHOR]

Artificial Intelligence has found wide application, but also poses risks due to unintentional or malicious tampering during deployment. Regular checks are therefore necessary to detect and prevent such risks. Fragile watermarking is a technique used to identify tampering in AI models. However, previous methods have faced challenges including risks of omission, additional information transmission, and inability to locate tampering precisely. In this paper, we propose a method for detecting tampered parameters and bits, which can be used to detect, locate, and restore parameters that have been tampered with. We also propose an adaptive embedding method that maximizes information capacity while maintaining model accuracy. Our approach was tested on multiple neural networks subjected to attacks that modified weight parameters, and our results demonstrate that our method achieved great recovery performance when the modification rate was below 20%. Furthermore, for models where watermarking significantly affected accuracy, we utilized an adaptive bit technique to recover more than 15% of the accuracy loss of the model. • Fragile watermarking for rigorous neural network integrity detection. • Watermarking ensures 100% tampering detection, precise location, and recovery. • Watermarking fragility by combining Mutual- and Self-neural net parameters. • Incorporates watermarking with neural net feature, maintaining watermarked models' performance via gradient descent. • Bit-level training controls step size with carefully set hyperparameters. [ABSTRACT FROM AUTHOR]

• A solar image captioning method that applies a transformer-based deep learning natural language processing is proposed. • A DeepSDO description dataset for training solar image captioning models including nine solar events is provided. • The experimental results show that the proposed method outperforms other benchmark methods in terms of four evaluation metrics. Solar images are essential for identifying and predicting solar phenomena, and have been used as key information for analyzing space weather. In this paper, we propose a solar image captioning method that applies a transformer-based deep learning (DL) natural language processing method. In addition, we provide a new DeepSDO description dataset for training solar image captioning models. First, we develop the DeepSDO description dataset using solar image data from Korean Data Center for solar dynamics observatory (SDO) and scripts from the National Aeronautics and Space Administration (NASA) SDO gallery website. The DeepSDO description dataset includes nine solar events: sunspots, flares, prominences, prominent eruptions, coronal holes, coronal loops, filaments, active regions, and eclipses. Second, we train the DL-based image captioning model, the meshed-memory transformer, using the DeepSDO description dataset. The experimental results show that the proposed method outperforms other benchmark methods in terms of four evaluation metrics. This study demonstrates that DL-based image captioning can successfully generate solar image captions for multiple solar features, and could potentially be used in other themes of solar physics and space weather. [ABSTRACT FROM AUTHOR]

• In clinical needle EMG interpretation, conventional machine learning methods were more often studied than deep learning methods. • Most studies developed models for classification of contraction signals and a few addressed the classification of rest signals. • Current research has methodological shortcomings, which prevent clinical implementation, but are surmountable. This scoping review provides an overview of artificial intelligence (AI), including machine and deep learning techniques, in the interpretation of clinical needle electromyography (nEMG) signals. A comprehensive search of Medline, Embase and Web of Science was conducted to find peer-reviewed journal articles. All papers published after 2010 were included. The methodological quality of the included studies was assessed with CLAIM (checklist for artificial intelligence in medical imaging). 51 studies were identified that fulfilled the inclusion criteria. 61% used open-source EMGlab data set to develop models to classify nEMG signal in healthy, amyotrophic lateral sclerosis (ALS) and myopathy (25 subjects). Only two articles developed models to classify signals recorded at rest. Most articles reported high performance accuracies, but many were subject to bias and overtraining. Current AI-models of nEMG signals are not sufficient for clinical implementation. Suggestions for future research include emphasizing the need for an optimal training and validation approach using large datasets of clinical nEMG data from a diverse patient population. The outcomes of this study and the suggestions made aim to contribute to developing AI-models that can effectively handle signal quality variability and are suitable for daily clinical practice in interpreting nEMG signals. [ABSTRACT FROM AUTHOR]

Geographic spaces is a vast and complex system involving multiple elements and nonlinear interactions of these elements, and rich in geographical phenomena, processes and patterns. Artificial intelligence methods (AI) are increasingly utilized to extract information of interest, patterns and insights from massive remote sensing (RS) images. Among them, two representative paradigms for RS interpretation are knowledge-driven symbolism and data-driven connectionism. Knowledge-driven approaches are certain and theoretically sound, which consider the regularities, formulae with accompanying theorems, and expert knowledge. However, it is difficult to address huge-scale RS data and complex nonlinear problems. The data-driven paradigm has stronger data learning and representation ability, yet poor interpretability, low trustworthiness of interpretation results, and reliance on massive labeled data. To address these limitations, we argue that current RS intelligent interpretation requires the guidance of geographic ideas and a unified theoretical framework rather than the independent use of remotely sensed data, duplicative development of interpretation models, or isolated applications of geographic information extraction. Thus, we introduce the concepts of trustworthy RS interpretation (TRSI) with multi-granularity of space, time, and attribute to facilitate the understanding of the morphology, types, and indicators (visual perception) and the perspective of the structure, evolution, and trends (geographical cognition) of complex geographical space. Second, integrated technologies of the TRSI with multiple scales and stages are developed, including a pixel-level visual perception, a geo-parcel-level quantitative inversion, and a scene-level geographical cognition. Finally, a RS big data interpretation system with quantitative semantic parsing is designed to support precise applications with cloud–edge-end collaboration. Rather than a simple overview and accumulation of traditional RS intelligent interpretation methods, this paper analyzes current challenges in RS interpretation and proposes new concepts, technologies, and applications of TRSI. It will empower us to perceive the complex geographical space comprehensibly and reveal its structural composition and evolutionary mechanism, serving geospatial decision applications such as evaluation, planning, and forecasting. [ABSTRACT FROM AUTHOR]

Land-cover information reflects basic Earth's surface environments and is critical to human settlements. As a well-established deep learning architecture, the fully convolutional network has achieved impressive progress in various land-cover mapping tasks. However, most research has focused on designing powerful encoders, ignoring the exploration of decoders. The existing handcrafted decoders are relatively simple and lack flexibility, limiting the generalizability for complex remote sensing scenes. In this paper, we propose a Land-cOVEr mapping Neural Architecture Search framework (LoveNAS) to automatically find efficient decoders that are compatible with the encoders and tasks. Specifically, LoveNAS introduces a hierarchical dense search space, including densely connected layer-level and multi-scale operation-level search spaces. The search spaces contain independent connection and operation fusion strategies, facilitating sufficient interaction of multi-scale features. After searching based on large-scale datasets, a series of pre-trained encoders and adaptive decoders are obtained. These can be smoothly applied to multi-scene tasks using weight-transfer network training. Experimental results on normal and disaster scenes shows that LoveNAS outperforms 16 handcrafted architectures and NAS methods. Some searched structures coincide with the existing advanced artificial designs, revealing the potential value of LoveNAS in network design and guidance. Group's website: http://rsidea.whu.edu.cn/resource%5fsharing.htm. GitHub page: https://github.com/Junjue-Wang/LoveNAS. [ABSTRACT FROM AUTHOR]

Building footprint information is crucial for comprehending global urban development processes. Deep learning algorithms have shown significant potential in building extraction from high spatial resolution imagery. However, the requirement for large-scale annotated data has been a limitation for applying deep learning methods to city-level or national-level building mapping. The dynamic change and distinct landscape variation of cities in different geographic locations further emphasizes the need for automatic building footprint extraction. In this paper, we propose a Self-supervised Knowledge Transfer (SKTrans) framework for extracting building footprints from unlabeled remote sensing images over large areas. To address building tone differences cross regions and time periods, a tone shift mechanism is introduced to reduce the contrast, brightness, and saturation differences between buildings. Furthermore, considering building style diversity across different cities, a knowledge integration module is proposed to develop a comprehensive depiction including tonal differences, multi-scale variance, structural attributes and semantic confusion between roads and buildings. The effectiveness of SKTrans is demonstrated on three public datasets of WHU, Massachusetts, and Inria. To further evaluate the feasibility of large-scale mapping for automatic building footprint, global generalization experiments cross regions and time periods, which is based on imagery from five countries of different continents were implemented. The comprehensive results demonstrate that SKTrans surpasses existing state-of-the-art methods, supporting the time-series city-level building mapping without labeled dataset. Additionally, the architectural landscape pattern changes can be quantitatively analyzed based on the generalization results to facilitate sustainable urban development. Overall, SKTrans provides a new insight to meet the needs of high-precision label-free building mapping and update from large-scale multi-style remote sensing images. [ABSTRACT FROM AUTHOR]

• Extract ancient city wall sites at pixel level from LiDAR remote sensing data based on deep learning. • Improvement strategies in the dataset generation, model training, and post-processing steps. • A significant source of inspiration for the inspection, conservation, and monitoring of cultural heritage. Ancient city walls, one of the most notable distinguishing features of Chinese ancient cities, are military defenses constructed of rammed earth. The ancient city walls have considerable study value because they served as the city's boundary and a symbol of power. However, as a result of natural erosion and human activities, many sites have been ruined. Existing optical remote sensing technologies, LiDAR point cloud processing algorithms, and deep learning methods are inadequate for the extraction and segmentation of ancient city wall sites. The novel semantic segmentation method for ancient city wall sites is described in this paper that extracts sites at the pixel level from LiDAR remote sensing data based on deep learning. To begin, the point cloud data collected by airborne laser scanning is processed into DEM data, and the distribution of ancient city walls in the study area is obtained through archaeological survey and expert interpretation. The dataset for deep learning semantic segmentation is then generated using image cropping and data augmentation techniques. Third, implement a U-Net semantic segmentation framework for microtopographic archaeological sites, and predict ancient city wall sites in the testing region after model training. Finally, the deep learning results are optimized using the connected component analysis method, and prediction mistakes such as holes and noise are removed. Taking Jinancheng, the capital city of the Chu kingdom, as an example, the proposed method process can identify and extract the ancient city wall sites at the pixel level, where the evaluation metrics reach 94.12% (Precision) and 81.38% (IoU). The experiment results are excellent due to improvement strategies in the dataset generation, model training, and post-processing steps. Thus, this study is significant for the current survey and protection of ancient city wall sites. The source code will be freely available at https://github.com/wshunli/Open-CHAI-CityWalls. [ABSTRACT FROM AUTHOR]

• A Digital Cultural Heritage dataset of synthetic vaults exported as point cloud, based on shape grammar discussion. • A vaults synthetic dataset creation method for automatically generating training data when real 3D objects are insufficient. • A Deep Learning system able to segment and distinguish types of architectural vaults according to standard classifications. Historical heritage is demanding robust pipelines for preserving, enhancing, and disseminating its prominent value. Semantic segmentation of 3D Point Clouds has gained increasing attention over the years, since it might assist in automati-cally recognizing historical architectural elements, thus facilitating large dataset management. Nonetheless, semantic segmentation is particularly challenging in Cultural Heritage (CH) domain, due to the shapes complexity and the limited repeatability of elements across different architectures, which strengthens the difficulty to define common patterns within the same class of elements. Besides, as Deep Neural Networks demand an appreciably amount of labelled data to be trained, the lack of available annotated heritage point clouds prevent the research in this direction. To tackle these issues, in this paper it is proposed a Deep Learning system able to recognize historical building elements by lever-aging synthetic point cloud. The generation of the 3D models, vaults, is based on a procedural modeling approach that follows the ideal shapes, according to the rules of descriptive geometry for the main types of vaults. The approach has been applied to a newly synthetic dataset which is publicly available. This dataset comprises 6 labelled points clouds, derived from a comprehensive on-tological taxonomy in order to describe an univocal and robust architectural hierarchy: barrel vaults, groined vaults, mirror vaults, barrel vaults with clois-ter heads and lunettes, barrel vaults with lunettes, sail vaults. The experiments yield high accuracy, demonstrating the effectiveness and suitability of the pro-posed approach. [ABSTRACT FROM AUTHOR]

Data biases such as class imbalance and label noise always exist in large-scale datasets in real-world. These problems bring huge challenges to deep learning methods. Some previous works adopted loss re-weighting, sample re-weighting, or data-dependent regularization to mitigate the influence of these training biases. But these methods usually pay more attention to class imbalance problem when both the class imbalance and label noise exist in training set simultaneously. These methods may overfit noisy labels, which leads to a great degradation in performance. In this paper, we propose a gradient-aware learning method for the combination of the two biases. During the training process, we update only a part of crucial parameters regularly and rectify the update direction of the rest redundant parameters. This update rule is conducted both in the encoder and classifier of the deep network to decouple label noise and class imbalance implicitly. The experimental results verify the effectiveness of the proposed method on synthetic and real-world data biases. [ABSTRACT FROM AUTHOR]

Deep learning models have been widely used during the last decade due to their outstanding learning and abstraction capacities. However, one of the main challenges any scientist has to face using deep learning models is to establish the network's architecture. Due to this difficulty, data scientists usually build over complex models and, as a result, most of them result computationally intensive and impose a large memory footprint, generating huge costs, contributing to climate change and hindering their use in computational-limited devices. In this paper, we propose a novel dense feed-forward neural network constructing method based on pruning and transfer learning. Its performance has been thoroughly assessed in classification and regression problems. Without any accuracy loss, our approach can compress the number of parameters by more than 70%. Even further, choosing the pruning parameter carefully, most of the refined models outperform original ones. Furthermore, we have verified that our method not only identifies a better network architecture but also facilitates knowledge transfer between the original and refined models. The results obtained show that our constructing method not only helps in the design of more efficient models but also more effective ones. [ABSTRACT FROM AUTHOR]

Traffic flow prediction plays an instrumental role in modern intelligent transportation systems. Numerous existing studies utilize inter-embedded fusion routes to extract the intrinsic patterns of traffic flow with a single temporal learning approach, which relies heavily on constructing graphs and has low training efficiency. Different from existing studies, this paper proposes a spatio-temporal ensemble network that aims to leverage the strengths of different sequential capturing approaches to obtain the intrinsic dependencies of traffic flow. Specifically, we propose a novel model named graph temporal convolutional long short-term memory network (GT-LSTM), which mainly consists of features splicing and patterns capturing. In features splicing, the spatial dependencies of traffic flow are captured by employing self-adaptive graph convolutional network (GCN), and a non-inter-embedded approach is designed to integrate the spatial and temporal states. Further, the aggregated spatio-temporal states are fed into patterns capturing, which can effectively exploit the advantages of temporal convolutional network (TCN) and bidirectional long short-term memory network (Bi-LSTM) to extract the intrinsic patterns of traffic flow. Extensive experiments conducted on four real-world datasets demonstrate that the proposed network obtains excellent performance in both forecasting accuracy and training efficiency. [ABSTRACT FROM AUTHOR]

Recent years have witnessed increasing interest in adversarial attacks on images, while adversarial video attacks have seldom been explored. In this paper, we propose a sparse adversarial attack strategy on videos (DeepSAVA). Our model aims to add a small human-imperceptible perturbation to the key frame of the input video to fool the classifiers. To carry out an effective attack that mirrors real-world scenarios, our algorithm integrates spatial transformation perturbations into the frame. Instead of using the l p norm to gauge the disparity between the perturbed frame and the original frame, we employ the structural similarity index (SSIM), which has been established as a more suitable metric for quantifying image alterations resulting from spatial perturbations. We employ a unified optimisation framework to combine spatial transformation with additive perturbation, thereby attaining a more potent attack. We design an effective and novel optimisation scheme that alternatively utilises Bayesian Optimisation (BO) to identify the most critical frame in a video and stochastic gradient descent (SGD) based optimisation to produce both additive and spatial-transformed perturbations. Doing so enables DeepSAVA to perform a very sparse attack on videos for maintaining human imperceptibility while still achieving state-of-the-art performance in terms of both attack success rate and adversarial transferability. Furthermore, built upon the strong perturbations produced by DeepSAVA, we design a novel adversarial training framework to improve the robustness of video classification models. Our intensive experiments on various types of deep neural networks and video datasets confirm the superiority of DeepSAVA in terms of attacking performance and efficiency. When compared to the baseline techniques, DeepSAVA exhibits the highest level of performance in generating adversarial videos for three distinct video classifiers. Remarkably, it achieves an impressive fooling rate ranging from 99.5% to 100% for the I3D model, with the perturbation of just a single frame. Additionally, DeepSAVA demonstrates favourable transferability across various time series models. The proposed adversarial training strategy is also empirically demonstrated with better performance on training robust video classifiers compared with the state-of-the-art adversarial training with projected gradient descent (PGD) adversary. • Sparse attacks on video models: perturb fewer frames to gain high fooling rate. • Combining additive and spatial perturbations to enhance attacking performance. • Using SSIM instead of l p -norm to maintain the human perception. • Applying Bayesian Optimisation to identify the most critical frame to perturb. • A new adversarial training method based on combination of diverse perturbations. [ABSTRACT FROM AUTHOR]

Decentralized deep learning algorithms leverage peer-to-peer communication of model parameters and/or gradients over communication graphs among the learning agents with access to their private data sets. The majority of the studies in this area focus on achieving high accuracy, with many at the expense of increased communication overhead among the agents. However, large peer-to-peer communication overhead often becomes a practical challenge, especially in harsh environments such as for an underwater sensor network. In this paper, we aim to reduce communication overhead while achieving similar performance as the state-of-the-art algorithms. To achieve this, we use the concept of Minimum Connected Dominating Set from graph theory that is applied in ad hoc wireless networks to address communication overhead issues. Specifically, we propose a new decentralized deep learning algorithm called minimum connected Dominating Set Model Aggregation (DSMA). We investigate the efficacy of our method for different communication graph topologies with a small to large number of agents using varied neural network model architectures. Empirical results on benchmark data sets show a significant (up to 100X) reduction in communication time while preserving the accuracy or in some cases, increasing it compared to the state-of-the-art methods. We also present an analysis to show the convergence of our proposed algorithm. • Proposing a communication-efficient protocol for decentralized learning systems. • Finding valuable connections in the network to reduce communication overhead. • Hybrid of federated learning and gossiping methods. • Verifying the method performance for different networks, datasets, and models. • Preserving accuracy while reducing communication, even in large-scale networks. [ABSTRACT FROM AUTHOR]

Hyper-realistic images that are not differentiable from authentic images to regular viewers have become extremely easy to generate and highly accessible. Furthermore, the increasing pervasiveness of social media networks in our daily lives has facilitated the easy dissemination of fake news accompanied by such synthetic images. Hyper-realistic artificial face images are often illicitly used as profile pictures on social media sites, further using such profiles to spread fabricated information, resulting in social perils. Most available synthetic image detectors are challenging to implement in practical scenarios due to their high complexity and performance degradation for images from Online Social Networks (OSNs). In this work, we develop a deep learning-based lightweight synthetic image detector called Relative Chrominance Distance Network (RCD-Net). In this paper, we introduce the RCD image feature set for the first time, which gives a pair-wise chrominance component-based distance measure. To show its effectiveness, we explore multiple luminance-chrominance spaces. Compared to the state-of-the-art (SOTA), our model hugely reduces the network parameter requirements, making it incredibly lightweight. We also study the robustness of the proposed solution against common post-processing operations in the context of online social media networks. Experimental results prove that the proposed solution achieves SOTA performance at a much lower complexity than available solutions. • GAN-generated synthetic images, visually indistinguishable from authentic photos, pose a serious threat to society. • Synthetic images exhibit different statistical properties in the colour domain than authentic images. • Our proposed feature RCD (Relative Chrominance Distance) performs excellently to differentiate real and synthetic images. • Proposed deep RCD-Net feeds on RCD, high-frequency residual, luminance features; achieves extreme parameter optimization. • RCD-Net achieves high robustness against common image post-processing operations and compression. [ABSTRACT FROM AUTHOR]

We study how the topology of feature embedding space changes as it passes through the layers of a well-trained deep neural network (DNN) through Betti numbers. Motivated by existing studies using simplicial complexes on shallow fully connected networks (FCN), we present an extended analysis using Cubical homology instead, with a variety of popular deep architectures and real image datasets. We demonstrate that as depth increases, a topologically complicated dataset is transformed into a simple one, resulting in Betti numbers attaining their lowest possible value. The rate of decay in topological complexity (as a metric) helps quantify the impact of architectural choices on the generalization ability. Interestingly from a representation learning perspective, we highlight several invariances such as topological invariance of (1) an architecture on similar datasets; (2) embedding space of a dataset for architectures of variable depth; (3) embedding space to input resolution/size, and (4) data sub-sampling. In order to further demonstrate the link between expressivity & the generalization capability of a network, we consider the task of ranking pre-trained models for downstream classification task (transfer learning). Compared to existing approaches, the proposed metric has a better correlation to the actually achievable accuracy via fine-tuning the pre-trained model. • This paper categorizes the change in topological complexity through Betti-Numbers. • Attempts to characterize the learning ability of the DNNs through topological transformation of feature space. • Attempts to use the topological complexity of feature space for the task of ranking pre-trained models. [ABSTRACT FROM AUTHOR]

The impact crater is the most dominant geomorphic structural units over the planetary surface and its accurate extraction is of great significance to investigate the evolution of the Moon, the impact history of the solar system and the space environment. Through the impact crater's identification, especially small-scale, and their spatial distribution characteristics, the impact flux, morphological characteristics, age of impact craters and surface degradation can be revealed. Therefore, this paper presents an automatic method for the impact crater detection, especially small-scale ones, using anchor-free deep learning. Due to the limitation of traditional object proposals for detecting different types of impact craters, we retrain the anchor-free CenterNet model using a transfer learning strategy, where the detection task of impact crater is formulated into the detection of its center point and regression of its property (i.e., size) without non-maximal suppression. We select the stacked Hourglass network as backbone to aggregate different levels of feature for enhancing the capability of estimating the impact crater centers. Moreover, we find the center points of impact craters on the image features' heatmap only based on their locations, instead of box overlap, which allows us to detect different types of impact craters, even the impact craters that contain other impact craters. The model is trained in an end-to-end manner and applied to detect the impact craters on the lunar images between ±50° of latitude, with the spatial resolution of 100 m/pixel, from Wide Angle Camera (WAC) onboard Lunar Reconnaissance Orbiter (LRO) mission. The obtained results are compared relative to an existing crater database both qualitatively and quantitatively that suggests the reliability and robustness of the developed method in the automatic detection of small-scale impact craters, where the smallest one is 500 m. Moreover, the developed method is capable to detect different types of impact craters, including dispersal and connective ones, with the recall of 73.66% and precision of 78.27% compared with an existing crater database. The code is publicly available at https://github.com/ShuoweiZhang/crater_detection. [ABSTRACT FROM AUTHOR]

Hydrogen attack is the major failure cause of hydrogen equipment breakdown maintenance. Especially, common problems like cracks frequently occur but are challenging to find while operating, presenting an issue for safety and production. In the process of hydrogen damage evolution, a method for defect state recognition is proposed in this paper. Acoustic emission (AE) technology is used for inspecting the entire hydrogen charging process. The characteristic parameters including the counts and duration of the AE signals are first preprocessed, and the current damage states such as the dislocation propagation, and the occurrence of cracks are identified. Then, a deep learning convolutional neural network is used to create a hydrogen defect recognition (HDR) model with the input of a short-time Fourier transform for the feature vector extraction of various damage status AE signals. Finally, the hydrogen defect recognition experiment revealed that HDR is better in classification accuracy at 98.37% due to dislocation propagation and cracks identification. The study can provide an online damage recognition approach for damage state early warning and evaluation to guarantee hydrogen equipment safety operation. • Monitor the hydrogen damage process using acoustic emission detection technology. • Characteristics of the acoustic sources in hydrogen damage process are studied. • Neural network is used for hydrogen damage signal recognition and diagnosis. • STFT transform image is used as the neural network model input. [ABSTRACT FROM AUTHOR]

In recent times, there has been notable and swift advancement in the field of image dehazing. Several deep learning techniques have demonstrated remarkable proficiency in resolving homogeneous dehazing issues. Nonetheless, the current dehazing approaches are generally formulated to deal with homogeneous haze, which is often undermined in real-world scenarios due to the uncertain haze dispersion. In this paper, we propose a dehazing model named RepDehazeNet by combining a structurally Reparameterization Encoder-Decoder subnet and a Full-Resolution Attention subnet. To be specific, the structural reparameterization idea is introduced into the encoder–decoder subnet to strengthen the feature extraction of dehazed images and improve the feature extraction speed. RepDehazeNet is compared with seven SOTA models on different datasets in terms of PSNR, SSIM, parameter quantity, and inference time. Compared to the DW-GAN model, the proposed RepDehazeNet model reduces the number of parameters by 2.7 million, and improves the inference speed by 90.3%, while achieving a higher PSNR of 0.5 dB on the NH-Haze2021 dataset. The experimental results demonstrate that the proposed RepDehazeNet model can effectively improve the real-time performance, accuracy of dehazing synthesized and nonhomogeneous haze images. [Display omitted] • Structural reparameterization dehazenet: outstanding performance, faster speed. • Replacing Tanh with ReLU leads to better results. • Transfer learning addresses the problem of insufficient samples. • Dual subnets method proves highly effective in datasets of different scales. [ABSTRACT FROM AUTHOR]

Recently, ensemble multiple deep learning (DL) classifiers has been reported to be an effective method for improving remote sensing classification accuracy. Although these approaches still follow the conventional pattern of inputting instance features and outputting corresponding classes, they often overlook the intrinsic relationships between pixels beyond their spatial features. As a result, the diversity in the ensemble classification results primarily relies on different DL models. However, training the DL models consumes a significant amount of time, and training multiple networks not only incurs additional time costs but also affects the overall efficiency. To address this, a new approach has been proposed in this paper, which takes advantage of the relationships between pixels and their combinations to generate diverse classification results. It's a novel ensemble classification framework, termed as the Doublet-Based Ensemble Classification Framework (DBECF), which eliminates the need for multiple classifiers. The DBECF starts by utilizing the training set to combine different samples to generate doublets. Then, features are assigned to these doublets through an exponentiation operation, resulting in a doublet training set. Using both the original training set and the derived doublet datasets, the DBECF is trained. For each input pixel, the DBECF produces multiple classification results, which are then integrated to obtain a more accurate output. To validate the proposed approach, experiments were conducted on three datasets, including multispectral images, hyperspectral images, and time series images. The maximum accuracies achieved by DBECF on the three datasets are 87.80 %, 97.71 %, and 83.51 %, respectively. In comparison to the contrastive methods, the incremental improvements in accuracy are 3.73 %, 7.66 %, and 9.16 %, respectively. The experimental results indicate that no matter using DL or non-deep learning for training, our proposed framework achieves progress on accuracy improvement outperforming classifications using comparative approach that based on single instance. This research provides a new perspective on the combination of DL and ensemble learning, highlighting its important implications and practical value in enhancing classification accuracy and efficiency. [ABSTRACT FROM AUTHOR]

Nowadays, meteorological data plays a crucial role in various fields such as remote sensing, weather forecasting, climate change, and agriculture. The regional and local studies call for high spatial resolution gridded meteorological data to identify refined details, which however is generally limited due to the models, platforms, sensors, etc. Downscaling has been a significant and practical way to improve spatial resolution. In recent years, with superior feature extraction and expression abilities, deep learning (DL) has outperformed traditional methods in various areas, and exhibits huge potential to establish a complicated mapping between large-scale and local-scale meteorological data. Therefore, this paper provides a systematic review of DL in statistical downscaling for deriving high spatial resolution gridded meteorological data. This review first presents the background, including a taxonomy of downscaling methods, the role of DL in statistical downscaling, and the analogy between downscaling and image super-resolution. It shows evidence of how downscaling can benefit from DL, particularly super-resolution networks. Subsequently, this review focuses on the recent development of the DL-based statistical downscaling of gridded meteorological data, especially the deep architectures, including convolutional neural networks to capture the spatial dependencies of meteorological variables, recurrent neural networks to reveal the temporal states from time series, and generative adversarial networks to facilitate the reconstruction of high-frequency details, as well as the major structure residual learning and attention mechanism. In addition, this review demonstrates the specific issues towards downscaling, including scaling factors, spatial–temporal and variable correlations, and paired datasets construction, and then gives a comprehensive summary of the status of datasets, toolsets and metrics. The future challenges from the perspective of unsupervised models, transformer architecture, data fusion, physical-informed learning, generalization capacity, and uncertainty quantification for downscaling are finally discussed. • Deep learning maps large-scale predictors to local-scale predictand for downscaling. • The major deep structures for downscaling networks and enhancement are reviewed. • The specific issues on downscaling factor, correlations, and datasets are addressed. • Unsupervised, physical-informed learning, uncertainty etc. pose great challenges. [ABSTRACT FROM AUTHOR]