Search

Defect detection is crucial in ensuring the quality of steel products. This paper proposes a novel deep neural network, cascaded adaptive global location network (CAGLNet), for detecting steel surface defects. The main objective of this study is to address the challenges associated with the irregular shape and dense spatial distribution of defects on steel. To achieve this goal, CAGLNet integrates a feature extraction network that combines residual and feature pyramid networks, a cascade adaptive tree-structure region proposal network (CAT-RPN) that eliminates the need for prior knowledge, and a global localisation regression for steel defect detection. This paper evaluates the effectiveness of CAGLNet on the NEU-DET dataset and demonstrates that the proposed model achieves an average accuracy of 85.40% with a fast frames per second of 10.06, outperforming those state-of-the-art methods. These results suggest that CAGLNet has the potential to significantly improve the effectiveness of defect detection in industrial production processes, leading to increased production yield and cost savings. Abbreviations: AT-RPN, adaptive tree-structure region proposal network; CAGLNet, cascaded adaptive global location network; CAT-RPN, cascade adaptive tree-structure region proposal network; CNN, convolutional neural network; DNN, deep neural network; EPNet, edge proposal network; FPN, feature pyramid network; FCOS, fully convolutional one-stage detector; FPS, frames per second; GMM, Gaussian mixture model; IoU, intersection-over-union; ROIAlign, region of interest align; RPN, region proposal network; ResNet, residual network; ResNet50_FPN, residual network and feature pyramid network; SABL, side aware boundary localisation; SSD, single-shot multiBox detector; TPE, Tree-structured Parzen estimator [ABSTRACT FROM AUTHOR]

A force plate is mainly used in biomechanics; it aims to measure the ground reaction force in a persons walking or standing position. In this study, a large-area force mat of the piezoresistance sensing type was developed, and a deep-learning-based weight measurement calibration method was applied to solve the problem in which measurements are not normalized because of physical limitations in hardware and signal processing. The test set was composed of the values measured at each point by weight and the value of the center of the pressure variable, and the measured value was predicted using a deep neural network (DNN) regression model. The calibration verification results show that the average weight errors range from a minimum of 0.06% to a maximum of 3.334%. This is simpler than the previous method, which directly measures the ratio of the resistance value to the measured weight of each sensor and derives an equation.

IoT-based healthcare (HC) systems face security and efficiency challenges. Existing solutions, such as secure transmission models, enhanced security protocols, and secure frameworks, neglect patient authentication and rely on resource-intensive cryptography, leading to vulnerabilities and increased energy consumption. The use of Exponential Key-based Elliptical Curve Cryptography (EKECC) in previous work raises concerns about its long-term viability against quantum computing threats. Additionally, the Path-Weighted Q Reinforcement Learning (PWQRL) technique is limited to discrete action spaces, hindering its applicability in IoT-based HC systems with continuous action spaces. To address these issues, Enhanced Mayfly with Active elite approach Clustering based Deep Q Learner Routing with Enhanced Binary ring-learning-with-errors (EMACDQLEB) protocol is proposed in this paper. EMACDQLEB incorporates a quantum-resistant cryptographic scheme based on Enhanced Binary Ring-Learning-with-Errors (EBRLWE) and a routing algorithm using Deep Q-Networks (DQN). EBRLWE employs an additional encryption key to enhance data security against quantum threats. DQN enables optimal path selection for data transmission by using a Deep Neural Network (DNN) to approximate the Q-value function, improving routing efficiency. Experimental results show that EMACDQLEB outperforms previous methods in average energy consumption, reliability, and communication overhead. This paper aims to mitigate vulnerabilities and improve the HC infrastructure in the IoT era. [ABSTRACT FROM AUTHOR]

Emergency communication is a key link to ensure the stable operation of communication networks. And large-scale multiple-input multiple-output (MIMO) technology can significantly improve coverage of communication systems. This work aims to optimize the backhaul process through deep neural network to improve the response speed and data transmission efficiency of base stations. Specifically, the proposed method uses deep neural network to represent the backhaul process of base stations, and explores data transmission under high efficiency and low delay. Simulation experiments are designed to verify performance advantages of the proposed method in large-scale MIMO wireless ad hoc networking scenarios, by using three comparison metrics. The experimental results show that the proposed method has a significant performance improvement in data transmission rate, system stability and resource utilization compared with traditional methods. Besides, the feasibility and effectiveness of the algorithm in practical application are verified by simulation and actual environment test. To sum up, the proposed algorithm can provide a new direction for the technical research in the field of emergency communication. [ABSTRACT FROM AUTHOR]

Sedimentary basins, integral to Earth's geological history and energy resource exploration, undergo complex changes driven by sedimentation, subsidence and geological processes. Gravity anomaly inversion is a crucial technique offering insights into subsurface structures and density variations. Our study addresses the challenge of complex subsurface structure assessment by leveraging deep neural networks to invert observed gravity anomalies. Optimization approaches traditionally incorporate known density distributions obtained from borehole data or geological logging for inverting basement depth in sedimentary basins using observed gravity anomalies. Our study explores the application of deep neural networks in accurate architectural assessment of sedimentary basins and demonstrates their significance in mineral and hydrocarbon exploration. Recent years have witnessed a surge in the use of machine learning in geophysics, with deep learning models playing a pivotal role. Integrating deep neural networks, such as the feedforward neural networks, has revolutionized subsurface density distribution and basement depth estimation. This study introduces a deep neural network specifically tailored for inverting observed gravity anomalies to estimate two‐dimensional basement relief topographies in sedimentary basins. To enhance computational efficiency, a one‐dimensional discrete cosine transform based discretization approach is employed. Synthetic data, generated using non‐Gaussian fractals, compensates for the scarcity of true datasets for training the deep neural network model. The algorithm's robustness is validated through noise introduction with comparisons against an efficient and traditional global optimization‐based approach. Gravity anomalies of real sedimentary basins further validate the algorithm's efficacy, establishing it as a promising methodology for accurate and efficient subsurface imaging in geological exploration. [ABSTRACT FROM AUTHOR]

Most of the existing research works on stock price prediction ignore the synergistic effect of multi-source factors. Therefore, an enterprise stock price prediction method based on multi-source information fusion based on deep neural network (DNN) is proposed. Based on the condition of multi-source information fusion, the original data related to enterprise stock price are collected from multiple sources, and the structure level and model framework of the method are constructed. The feature extraction technology of DNN is used to extract the features useful for stock price prediction from the pre-processed data, and the DNN structure considering multi-source information fusion is established. Finally, feature extraction and evaluation settings are completed based on data variables. LOSS, ACCURACY and other indicators were used for analysis. The results show that compared with typical prediction methods, this method can make use of multiple information sources more comprehensively and improve the prediction accuracy. In addition, the proposed method has certain flexibility and can adapt to the characteristics and changes of different stock markets [ABSTRACT FROM AUTHOR]

Coronary Artery Bypass Graft (CABG) is a common cardiac surgery, but it continues to have many associated risks, including the need for blood transfusions. Previous research has shown that blood transfusion during CABG surgery is associated with an increased risk for infection and mortality. The current study aims to use modern techniques, such as deep neural networks and data synthesis, to develop models that can best predict the need for blood transfusion among CABG patients. Results show that neural networks with synthetic data generated by DataSynthesizer have the best performance. Implications of results and future directions are discussed. [ABSTRACT FROM AUTHOR]

Can brain structure predict human intelligence? T1-weighted structural brain magnetic resonance images (sMRI) have been correlated with intelligence. However, the population-level association does not fully account for individual variability in intelligence. To address this, studies have emerged recently to predict individual subject's intelligence or neurocognitive scores. However, they are mostly on predicting fluid intelligence (the ability to solve new problems). Studies are lacking to predict crystallized intelligence (the ability to accumulate knowledge) or general intelligence (fluid and crystallized intelligence combined). This study tests whether deep learning of sMRI can predict an individual subject's verbal, comprehensive, and full-scale intelligence quotients (VIQ, PIQ, and FSIQ), which reflect fluid and crystallized intelligence. We performed a comprehensive set of 432 experiments, using different input image channels, six deep learning models, and two outcome settings, in 850 healthy and autistic subjects 6–64 years of age. Our findings indicate a statistically significant potential of T1-weighted sMRI in predicting intelligence, with a Pearson correlation exceeding 0.21 (p < 0.001). Interestingly, we observed that an increase in the complexity of deep learning models does not necessarily translate to higher accuracy in intelligence prediction. The interpretations of our 2D and 3D CNNs, based on GradCAM, align well with the Parieto-Frontal Integration Theory (P-FIT), reinforcing the theory's suggestion that human intelligence is a result of interactions among various brain regions, including the occipital, temporal, parietal, and frontal lobes. These promising results invite further studies and open new questions in the field. [ABSTRACT FROM AUTHOR]

The geo-position spoofing detection in the images enables the tamper-proof protection of the images to ensure the authenticity of the images. Currently, available techniques for geo-position spoofing detection primarily depend on geo-position data stored in the database. In addition, these techniques are limited in performance due to the shortage of prior information, require more training samples and achieve poor generalization. To address the aforementioned limitations in this research, Eques Chimp Optimization-Deep Convolutional Neural Network (ECO-DCNN) is proposed to enhance the ability of the model to adapt to the evolving spoofing methods, handle noisy images and maintain computational efficiency. The significance of the research relies upon effectively tracing the source cameras by extracting the Photo Response NonUniformity noise (PRNU) from the images, which is then subjected to spoofing detection using Eques Chimp Optimized Deep Neural Network (ECO-DCNN). The ECO is designed by hybridizing the decision-making characters and social aspects of Eques and Chimp in the adaptive parameter tuning of the detection model to guarantee image integrity. Furthermore, for the Training Percentage (TP) analysis, the model exhibits a remarkable performance of 97.00% accuracy, 97.10% sensitivity, 97.19% specificity and 0.97 MCC. Similarly, the K-fold analysis achieves the values with accuracy, sensitivity, specificity and MCC for the model are 96.00%, 96.89%, 95.96% and 0.96%, respectively. [ABSTRACT FROM AUTHOR]

This concept will incorporate unmanned aerial vehicles (UAV) into the upcoming sixth-generation (6G) cellular network. UAVs have gained increasing interest for their ability to serve as a relay for efficiently boosting the range and communication rate of wireless networks. In the contexts of 6G and the Internet of Things (IoT), mm-Wave spectrum and non-orthogonal multiple access (NOMA) are expected to offer ubiquitous connectivity for many users. However, there is a chance of blocking severe conditions on the wireless terrestrial link between end users and the base station. Instead, because UAVs fly at a high altitude, it is much more likely that they will establish a line of sight (LoS) link. The multi-UAV network that enables an uplink (UL) NOMA cellular system is the main topic of this article. In addition, hybrid beamforming is used and operates in the mm-wave band. During hybrid beam forming, the Deep neural network (DNN) approach is used to classify each UAV's various directions of arrival. We develop a new design for the sum-rate maximization problem for the uplink mmWave transmission of the UAV-aided NOMA 6G network. The performance in terms of achievable sum rate compared to UL NOMA interconnections without the aid of a UAV network is better. [ABSTRACT FROM AUTHOR]

Social relationship prediction has garnered significant attention in the development of artificial intelligence technology due to its potential to promote socioeconomic development. Meantime, the increasing popularity and adoption of intelligent devices located in the Internet of Things (IoT) environment have generated abundant data, which can serve as a basis for social relationship prediction. Nevertheless, the highly fragmented distribution of user data and the implementation of data protection policies can lead to the dilemma of "data scarcity" during the process of social relationship prediction. The application of federated learning can effectively address data segregation and fragmentation. Additionally, existing studies on social relationship prediction fail to consider users' temporal sequence information. To this end, we present a novel Federated Deep Relationship Prediction framework, named FDRP, which adopts the principle of vertical federated learning. Specifically, under the IoT environment, the offline operation initially assigns virtual items and ratings to users. Subsequently, the online operation executes social relationship prediction, where the client converts the sparse data into dense vectors and extracts the overall temporal sequence features, as well as the server performs the model parameter aggregation. Through extensive experiments conducted on the Epinions dataset, FDRP demonstrates excellent prediction accuracy and convergence speed, effectively mitigating the risk of inference attacks. [ABSTRACT FROM AUTHOR]

The advent of voice assistance technology and its integration into smart devices has facilitated many useful services, such as texting and application execution. However, most assistive technologies lack the capability to enable the system to act as a human who can localize the speaker and selectively spot meaningful keywords. Because keyword spotting (KWS) and sound source localization (SSL) are essential and must operate in real time, the efficiency of a neural network model is crucial for memory and computation. In this paper, a single neural network model for KWS and SSL is proposed to overcome the limitations of sequential KWS and SSL, which require more memory and inference time. The proposed model uses multi-task learning to utilize the limited resources of the device efficiently. A shared encoder is used as the initial layer to extract common features from the multichannel audio data. Subsequently, the task-specific parallel layers utilize these features for KWS and SSL. The proposed model was evaluated on a synthetic dataset with multiple speakers, and a 7-module shared encoder structure was identified as optimal in terms of accuracy, direction of arrival (DOA) accuracy, DOA error, and latency. It achieved a KWS accuracy of 94.51%, DOA error of 12.397°, and DOA accuracy of 89.86%. Consequently, the proposed model requires significantly less memory owing to the shared network architecture, which enhances the inference time without compromising KWS accuracy, DOA error, and DOA accuracy. [ABSTRACT FROM AUTHOR]

Frequency-hopping (FH) communication adversarial research is a key area in modern electronic countermeasures. To address the challenge posed by interfering parties that use deep neural networks (DNNs) to classify and identify multiple intercepted FH signals—enabling targeted interference and degrading communication performance—this paper presents a batch feature point targetless adversarial sample generation method based on the Jacobi saliency map (BPNT-JSMA). This method builds on the traditional JSMA to generate feature saliency maps, selects the top 8% of salient feature points in batches for perturbation, and increases the perturbation limit to restrict the extreme values of single-point perturbations. Experimental results in a white-box environment show that, compared with the traditional JSMA method, BPNT-JSMA not only maintains a high attack success rate but also enhances attack efficiency and improves the stealthiness of the adversarial samples. [ABSTRACT FROM AUTHOR]

The Norwegian Environment Agency is responsible for updating a map of undisturbed nature, which is performed every five years based on aerial photos. Some of the aerial photos are already up to five years old when a new version of the map of undisturbed nature is published. Thus, several new nature interventions may have been missed. To address this issue, the timeliness and mapping accuracy were improved by integrating Sentinel-2 satellite imagery for the detection of new roads across Norway. The focus on new roads was due to the fact that most new nature interventions include the construction of new roads. The proposed methodology is based on applying U-Net on all the available summer images with less than 10% cloud cover over a five-year period, with an aggregation step to summarize the predictions. The observed detection rate was 98%. Post-processing steps reduced the false positive rate to 46%. However, as the false positive rate was still substantial, the manual verification of the predicted new roads was needed. The false negative rate was low, except in areas without vegetation. [ABSTRACT FROM AUTHOR]

Under the present Indian government initiative, allweather roads are being taken up for four pilgrimage locations in the Uttarakhand state of India. The Rishikesh to Gangotri road axis is a major road used by local citizens and tourists. Rainfall and numerous anthropogenic activities become the primary reasons for landslide hazards in the area. An accurate Landslide Susceptibility Map (LSM) for any area is of paramount importance for the decision makers of land-use planning. The present study gives a comparative analysis of recent advanced algorithms, i.e. CatBoost, LightGBM and deep neural network topology for generating the LSM by following pixel-based. Fourteen causative factors along with landslide inventory of 154 locations are used for the study. LSM are generated based on JENKS natural break criteria using all the algorithms and their performance comparison is evaluated. Overall accuracy for train and test data, prediction accuracy, area under receiver operating characteristics (AUROC) score for test data, and computational time for model fit on train data; are the criteria used for performance evaluation of each algorithm. In this study, it is observed that LSM can be generated at considerably fast pace if CatBoost or LightGBM is used while deep neural network-based topology gives marginally better results on all other performance measure. [ABSTRACT FROM AUTHOR]

Flotation is a physical and chemical process that uses the difference between mineral surface properties to separate valuable minerals from worthless minerals through the attachment of bubbles. The visual information of the flotation foam surface can reflect the flotation effect and will be closely related to the flotation conditions, directly indicating the degree of mineralization of the foam layer. Considering the problem that coal slurry foam images are difficult to segment due to the mutual adhesion of bubbles and fuzzy boundaries, in this work, we proposed a new and effective segmentation algorithm for extracting coal slurry flotation foam edge information. Based on the encoder and decoder structure, a selective multi-branch input segmentation deep neural network model was designed. We evaluated the proposed method and model using a self-built coal slurry foam image segmentation dataset. The intersection over union, mean intersection over union, and PA evaluation indicators of this method increased by 12.8376%, 7.2716%, and 1.457%, respectively, compared with those of the U-Net model. The experimental results showed that the selective branch segmentation network model had high accuracy and robustness in extracting coal slurry bubble edge information, and it could effectively solve the problem of difficult segmentation of coal slurry foam images. This study provided a foundation for optimizing flotation process parameters and realizing intelligent flotation site parameters. [ABSTRACT FROM AUTHOR]

Diabetic Retinopathy (DR) is a diabetic issue that influences the eyes. It harms the veins of the light-delicate tissue behind the eye (retina). DR is an issue of diabetes and the main source of visual deficiency. Since there are trained doctors to diagnose the disease, but to make their work easier, ensemble architectures were proposed for the prediction of grading of DR using modern technology known as deep learning. Deep neural network results in a better diagnostic system in comparison to machine learning networks. ResNET 101 and Ensemble CNN architectures were proposed using Adam Optimization for the detection of DR. From the results, it has been observed that for any constant size Network architecture, the loss cannot be further reduced after a certain number known as Bayes error. With the help of a learning curve, it can easily understand that the proposed architecture costs are decreasing after every iteration. The Ensemble and ResNET 101 model attain 87.67% and 81.28% accuracy, respectively, and 0.5046 and 0.5328 cross-entropy loss, respectively. The proposed Ensemble model results in an 18.4% improvement over the Inception V3 model and 7.8% over the proposed ResNet 101 model. [ABSTRACT FROM AUTHOR]

Clinical sleep diagnosis traditionally relies on polysomnography (PSG) and expert manual classification of sleep stages. Recent advancements in deep learning have shown promise in automating sleep stage classification using a single PSG channel. However, variations in PSG acquisition devices and environments mean that the number of PSG channels can differ across sleep centers. To integrate a sleep staging method into clinical practice effectively, it must accommodate a flexible number of PSG channels. In this paper, we proposed FlexSleepTransformer, a transformer-based model designed to handle varying number of input channels, making it adaptable to diverse sleep staging datasets. We evaluated FlexSleepTransformer using two distinct datasets: the public SleepEDF-78 dataset and the local SleepUHS dataset. Notably, FlexSleepTransformer is the first model capable of simultaneously training on datasets with differing number of PSG channels. Our experiments showed that FlexSleepTransformer trained on both datasets together achieved 98% of the accuracy compared to models trained on each dataset individually. Furthermore, it outperformed models trained exclusively on one dataset when tested on the other dataset. Additionally, FlexSleepTransformer surpassed state-of-the-art CNN and RNN-based models on both datasets. Due to its adaptability with varying channels numbers, FlexSleepTransformer holds significant potential for clinical adoption, especially when trained with data from a wide range of sleep centers. [ABSTRACT FROM AUTHOR]

Accurately predicting drug-target affinity is crucial in expediting the discovery and development of new drugs, which is a complex and risky process. Identifying these interactions not only aids in screening potential compounds but also guides further optimization. To address this, we propose a multi-perspective feature fusion model, MFF-DTA, which integrates chemical structure, biological sequence, and other data to comprehensively capture drug-target affinity features. The MFF-DTA model incorporates multiple feature learning components, each of which is capable of extracting drug molecular features and protein target information, respectively. These components are able to obtain key information from both global and local perspectives. Then, these features from different perspectives are efficiently combined using specific splicing strategies to create a comprehensive representation. Finally, the model uses the fused features to predict drug-target affinity. Comparative experiments show that MFF-DTA performs optimally on the Davis and KIBA data sets. Ablation experiments demonstrate that removing specific components results in the loss of unique information, thus confirming the effectiveness of the MFF-DTA design. Improvements in DTA prediction methods will decrease costs and time in drug development, enhancing industry efficiency and ultimately benefiting patients. • The MFF-DTA model is proposed to accurately predict the affinity relationship between drugs and target proteins. • The model includes multiple feature extractors to capture global and local key features. • We create a comprehensive and information-rich representation. • The model has demonstrated strong performance on public datasets. [ABSTRACT FROM AUTHOR]

Similar waveforms and overlapping frequency ranges make distinguishing blasting vibrations and microseismic signals challenging, causing interference with coal mine microseismic monitoring systems. To address this problem, we propose a spectrum dataset (MSData) reflecting the spectrum features of both signal types and present a signal classification network (SCNet) combining CNNs and Transformers for signal classification. The network can learn multi-dimensional features of both signals from MSData and automatically and efficiently identify the two signal types. Experimental results yield F1-scores of 0.991 for microseismic signals and 0.993 for blasting vibration signals, meeting engineering application requirements. [ABSTRACT FROM AUTHOR]

The intelligibility of speech is a primary component to assess the severity level of Dysarthria, a speech disorder, which is caused not only due to weakness in vocal motor muscles but also difficulty in controlling its movements. Prior information about the severity of Dysarthria, aids the therapist during the rehabilitation process. This paper introduces a novel hybrid architecture comprising Gaussian Mixture Model and Neural Network (GMM-NN) for categorizing Dysarthric severity into four classes based on speech intelligibility. Mel Frequency Cepstral Coefficients (MFCC) extracted from the segmented speech signals are used to train the classifier. The proposed model produced a 1.9% improvement in accuracy when compared to the baseline Gaussian Mixture Model (GMM). The Gaussian Mixture Model Deep Neural Network (GMM-DNN) and Gaussian Mixture Model Feed Forward Neural Network (GMM-FFNN) architectures showed an accuracy of 96.7% and 96.42% with F1 scores of 0.9649, 0.9604 respectively. [ABSTRACT FROM AUTHOR]

A deep neural network (DNN) was trained to generate a multiparametric ultrasound (mpUS) volume from four input ultrasound-based modalities (acoustic radiation force impulse [ARFI] imaging, shear wave elasticity imaging [SWEI], quantitative ultrasound-midband fit [QUS-MF], and B-mode) for the detection of prostate cancer. A DNN was trained using co-registered ARFI, SWEI, MF, and B-mode data obtained in men with biopsy-confirmed prostate cancer prior to radical prostatectomy (15 subjects, comprising 980,620 voxels). Data were obtained using a commercial scanner that was modified to allow user control of the acoustic beam sequences and provide access to the raw image data. For each subject, the index lesion and a non-cancerous region were manually segmented using visual confirmation based on whole-mount histopathology data. In a prostate phantom, the DNN increased lesion contrast-to-noise ratio (CNR) compared to a previous approach that used a linear support vector machine (SVM). In the in vivo test datasets (n = 15), the DNN-based mpUS volumes clearly portrayed histopathology-confirmed prostate cancer and significantly improved CNR compared to the linear SVM (2.79 ± 0.88 vs. 1.98 ± 0.73, paired-sample t -test p < 0.001). In a sub-analysis in which the input modalities to the DNN were selectively omitted, the CNR decreased with fewer inputs; both stiffness- and echogenicity-based modalities were important contributors to the multiparametric model. The findings from this study indicate that a DNN can be optimized to generate mpUS prostate volumes with high CNR from ARFI, SWEI, MF, and B-mode and that this approach outperforms a linear SVM approach. [ABSTRACT FROM AUTHOR]

The progressive loss of motor function in the brain is a hallmark of Parkinson's disease (PD). Electroencephalogram (EEG) signals are commonly used for early diagnosis since they are associated with a brain disorder. This work aims to find a better way to represent electroencephalography (EEG) signals and enhance the classification accuracy of individuals with Parkinson's disease using EEG signals. In this paper, we present two hybrid deep neural networks (DNN) that combine convolutional neural networks with long short-term memory to diagnose Parkinson's disease using EEG signals, that is, through the establishment of parallel and series combined models. The deep CNN network is utilized to acquire the structural features of ECG signals and extract meaningful information from them, after which the signals are sent via a long short-term memory network to extract the features' context dependency. The proposed architecture was able to achieve 97.6% specificity, 97.1% sensitivity, and 98.6% accuracy for a parallel model and 99.1% specificity, 98.5% sensitivity, and 99.7% accuracy for a series model, both in 3-class classification (PD patients with medication, PD patients without medication and healthy). [ABSTRACT FROM AUTHOR]

Image Aesthetic Assessment (IAA) is an emerging paradigm that predicts aesthetic score as the popular aesthetic taste for an image. Previous IAA approaches take a single image as input to predict the aesthetic score of the image. However, we discover that most existing IAA methods fail dramatically to predict the images with a large variance of aesthetic voting distribution. Motivated by the practice that people consider similar experiences to improve the consistence of the voting result, we present a novel Multiple Image joint Learning Network (MILNet) to mimic this natural process. Our novelty is mainly three-fold: (a) Semantic-based retrieval method that constructs aesthetic similarity (the similarity of aesthetic attribution) to select reference images; (b) Graph network reasoning that initializes and updates the weight of intrinsic relationships among multiple images; (c) Adaptive Earth Mover's Distance (AdaEMD) loss function that adjusts weight for easy and hard instances to mitigate unbalanced distribution of aesthetic datasets. Our evaluation with the benchmark AVA and TAD datasets demonstrates that the proposed MILNet outperforms state-of-the-art IAA methods. The code is available at https://github.com/flyingbird93/MILNet. [ABSTRACT FROM AUTHOR]

Neural networks are widely used in machine learning and data mining. Typically, these networks need to be trained, implying the adjustment of weights (parameters) within the network based on the input data. In this work, we propose a novel approach, RandomNet, that employs untrained deep neural networks to cluster time series. RandomNet uses different sets of random weights to extract diverse representations of time series and then ensembles the clustering relationships derived from these different representations to build the final clustering results. By extracting diverse representations, our model can effectively handle time series with different characteristics. Since all parameters are randomly generated, no training is required during the process. We provide a theoretical analysis of the effectiveness of the method. To validate its performance, we conduct extensive experiments on all of the 128 datasets in the well-known UCR time series archive and perform statistical analysis of the results. These datasets have different sizes, sequence lengths, and they are from diverse fields. The experimental results show that the proposed method is competitive compared with existing state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Purpose: Multiparametric arterial spin labeling (MP‐ASL) can quantify cerebral blood flow (CBF) and arterial cerebral blood volume (CBVa). However, its accuracy is compromised owing to its intrinsically low SNR, necessitating complex and time‐consuming parameter estimation. Deep neural networks (DNNs) offer a solution to these limitations. Therefore, we aimed to develop simulation‐based DNNs for MP‐ASL and compared the performance of a supervised DNN (DNNSup), physics‐informed unsupervised DNN (DNNUns), and the conventional lookup table method (LUT) using simulation and in vivo data. Methods: MP‐ASL was performed twice during resting state and once during the breath‐holding task. First, the accuracy and noise immunity were evaluated in the first resting state. Second, CBF and CBVa values were statistically compared between the first resting state and the breath‐holding task using the Wilcoxon signed‐rank test and Cliff's delta. Finally, reproducibility of the two resting states was assessed. Results: Simulation and first resting‐state analyses demonstrated that DNNSup had higher accuracy, noise immunity, and a six‐fold faster computation time than LUT. Furthermore, all methods detected task‐induced CBF and CBVa elevations, with the effect size being larger with the DNNSup (CBF, p = 0.055, Δ = 0.286; CBVa, p = 0.008, Δ = 0.964) and DNNUns (CBF, p = 0.039, Δ = 0.286; CBVa, p = 0.008, Δ = 1.000) than that with LUT (CBF, p = 0.109, Δ = 0.214; CBVa, p = 0.008, Δ = 0.929). Moreover, all the methods exhibited comparable and satisfactory reproducibility. Conclusion: DNNSup outperforms DNNUns and LUT with respect to estimation performance and computation time. [ABSTRACT FROM AUTHOR]

Detecting violence in images has emerged as an important area of research and application as digital imaging and computer vision advance. Detecting and assessing violent content automatically in visual data holds great promise for enhancing public safety, security, and content moderation. Violent incidents are difficult to detect because of disruptions in optical flow, complications caused by camera aperture variations, and difficulties with illumination and feature tracking. In this work, a novel F3DNN-Net has been proposed for violence detection in images. The input videos from the surveillance camera are initially converted into frame sequences and pre-processed. The Yolov8 (You Only Look Once) algorithm is employed to detect objects within pre-processed frames. Simultaneously, the ego algorithm is applied to the pre-processed images for motion estimation, ensuring comprehensive coverage of information. Afterwards, the features are extracted using the fined feature extraction phase as independent component analysis (ICA), and the selected features are selected using the Dingo algorithm. Then, the fine-tuned fused (FF) features are implemented in the deep neural network (DNN) approach for the categorization of violent and non-violent cases. According to the findings, the proposed F3DNN-Net detects violence in images with an accuracy rate of 99.33%. The proposed DNN enhances an overall F1-score of 10.46%, 6.91%, and 4.44%, better than Alex Net, ResNet, and Dense Net. The proposed F3DNN-Net enhances the overall accuracy by 6%, 0.3%, and 29.33% better than Inception-Resnet-V2, CNN, LSTM, and ResNet50, respectively. [ABSTRACT FROM AUTHOR]

\nPlain Language SummaryBrain tumors present a formidable diagnostic challenge due to their aberrant cell growth. Accurate determination of tumor location and size is paramount for effective diagnosis. Magnetic Resonance Imaging (MRI) and Positron Emission Tomography (PET) are pivotal tools in clinical diagnosis, yet tumor segmentation within their images remains challenging, particularly at boundary pixels, owing to limited sensitivity. Recent endeavors have introduced fusion-based strategies to refine segmentation accuracy, yet these methods often prove inadequate. In response, we introduce the Parallel-Way framework to surmount these obstacles. Our approach integrates MRI and PET data for a holistic analysis. Initially, we enhance image quality by employing noise reduction, bias field correction, and adaptive thresholding, leveraging Improved Kalman Filter (IKF), Expectation Maximization (EM), and Improved Vibe Algorithm (IVib), respectively. Subsequently, we conduct multi-modality image fusion through the Dual-Tree Complex Wavelet Transform (DTWCT) to amalgamate data from both modalities. Following fusion, we extract pertinent features using the Advanced Capsule Network (ACN) and reduce feature dimensionality via Multi-objective Diverse Evolution-based selection. Tumor segmentation is then executed utilizing the Twin Vision Transformer with dual attention mechanism. Implemented our Parallel-Way framework which exhibits heightened model performance. Evaluation across multiple metrics, including accuracy, sensitivity, specificity, F1-Score, and AUC, underscores its superiority over existing methodologies.Diagnosing brain tumors is challenging due to their abnormal growth. Accurately determining the tumor’s location and size is crucial for effective treatment. Magnetic Resonance Imaging (MRI) and Positron Emission Tomography (PET) are important tools in clinical diagnosis, but accurately segmenting tumors in their images, especially at the edges, is difficult due to limited sensitivity. To address this, we propose the Parallel-Way framework, which integrates MRI and PET data for a comprehensive analysis. First, we enhance image quality using noise reduction, bias field correction, and adaptive thresholding techniques like the Improved Kalman Filter (IKF), Expectation Maximization (EM), and Improved Vibe Algorithm (IVib). Next, we merge the MRI and PET data using the Dual-Tree Complex Wavelet Transform (DTWCT) to combine information from both modalities. After fusion, we extract relevant features using the Advanced Capsule Network (ACN) and reduce their complexity with Multi-objective Diverse Evolution-based selection. Tumor segmentation is then performed using the Twin Vision Transformer with a dual attention mechanism. Our Parallel-Way framework aims to improve the accuracy of brain tumor segmentation, overcoming limitations of existing methods. By integrating data from MRI and PET and applying advanced image processing and machine learning techniques, we believe our approach will provide a more reliable and accurate method for diagnosing and treating brain tumors. Evaluations using metrics such as accuracy, sensitivity, specificity, F1-Score, and AUC demonstrate that our method is superior to existing approaches, providing a more reliable and accurate way to diagnose brain tumors. [ABSTRACT FROM AUTHOR]

The generation of renewable energy encounters numerous obstacles, chiefly the unpredictability of renewable sources. When new energy generation prediction is lower than expected, it needs to be supplemented by other energy sources, which may lead to instability in the power grid if the deviation is large. When the prediction of new energy generation exceeds expectations, it will lead to energy waste. To address these issues, this paper proposes a Deep Neural Network-based fusion framework, which can improve the prediction accuracy of new energy and achieve a low-carbon, economical, and stable power grid. Within this structure, feature engineering is conducted initially. Subsequently, a combination of traditional tree algorithms like the Gradient Boosting Decision Tree, linear approaches such as the Least Squares Method, and nonlinear neural networks, for instance, Recurrent Neural Networks, are employed for individual model regression purposes. In the final step, both the original time-series data and the outcomes from the individual models are integrated into a deep neural network to derive the ultimate forecasting outcomes. By using our method, the electricity cost has been reduced by 26.5% and the carbon emissions have been decreased by 14.2%. Experiments have been carried out using actual community data, confirming the effectiveness of the proposed approach. The findings indicate that the integration of DNN with traditional and modern machine learning techniques can significantly improve the forecasting of renewable energy generation. This advancement contributes to the creation of a more sustainable, economical, and stable power grid. [ABSTRACT FROM AUTHOR]

Structure and properties of the electrode/electrolyte interface significantly influence the electrochemical processes of energy storage and conversion, yet the challenge lies in accurate description of both molecular characteristics and external field effects. Here, we develop a mesoscopic thermodynamic model that calculates the thermodynamic properties of electrolytes based on chemical potential, and its efficiency is enhanced by a deep neural network. The deep neural network enhanced mesoscopic thermodynamic (DeepMT) model effectively bridges the gap between micro‐level characteristics of ions and macro‐level effects of external field, enabling precise presentation of ion density distributions over complex conditions. Our result indicates that the DeepMT model not only demonstrates a computational efficiency improvement of approximately four orders of magnitude over direct theoretical calculations, but also accurately predicts interface properties including ion adsorption, surface charge, and differential capacitance through the statistical analysis of density distributions. [ABSTRACT FROM AUTHOR]

This work focuses on an intelligent management framework for vocal music media resources, with use of deep neural networks and web crawlers. The web driver module is used to simulate the Google browser access mode to collect and preprocess the data of the China vocal music network webpage project, and the delay code is added to solve the problem of Internet Protocol (IP) restrictions on crawling webpages. After parsing, the content of the project description is obtained and temporarily stored in the local file through the panda’s module to realize the automatic collection of project data. Membership functions of four basic types of emotions are designed. By utilizing the analytic hierarchy process, the experience and wisdom of many music experts are gathered. The weights of activity and the evoking force in determining the music emotion membership value are discussed, trying to make the membership function closer to reality. Through the application of the music emotion analysis model, the music emotion analysis technology is combined with the recommendation algorithm, and the music recommendation algorithm combined with the emotion analysis is realized. The experimental test proves that the algorithm has higher accuracy than the traditional music recommendation algorithm. [ABSTRACT FROM AUTHOR]

Simple Summary: In the duck breeding process, the rapid and accurate identification of gender is currently a key challenge. This study proposes a method based on sound signal discrimination to address this issue. The vocalizations of one-day-old male and female ducklings were extracted and the signal features were obtained through professional processing. Classification models were established using a backpropagation neural network (BPNN), a deep neural network (DNN), and a convolutional neural network (CNN). In total, 70% of the dataset was used for training, and 30% was used for testing. The test results showed that the BPNN model had the highest accuracy rate of 93.33% in identifying male ducks, while the CNN model had the highest accuracy rate of 95% in identifying female ducks. From the perspective of overall model performance, the CNN model performed the best, with the highest accuracy rate and F1-score of 84.15% and 84.32%, respectively. This study can provide a technical reference and a foundation for future voice recognition technologies for different poultry breeds and applications. Gender recognition is an important part of the duck industry. Currently, the gender identification of ducks mainly relies on manual labor, which is highly labor-intensive. This study aims to propose a novel method for distinguishing between males and females based on the characteristic sound parameters for day-old ducks. The effective data from the sounds of day-old ducks were recorded and extracted using the endpoint detection method. The 12-dimensional Mel-frequency cepstral coefficients (MFCCs) with first-order and second-order difference coefficients in the effective sound signals of the ducks were calculated, and a total of 36-dimensional feature vectors were obtained. These data were used as input information to train three classification models, include a backpropagation neural network (BPNN), a deep neural network (DNN), and a convolutional neural network (CNN). The training results show that the accuracies of the BPNN, DNN, and CNN were 83.87%, 83.94%, and 84.15%, respectively, and that the three classification models could identify the sounds of male and female ducks. The prediction results showed that the prediction accuracies of the BPNN, DNN, and CNN were 93.33%, 91.67%, and 95.0%, respectively, which shows that the scheme for distinguishing between male and female ducks via sound had high accuracy. Moreover, the CNN demonstrated the best recognition effect. The method proposed in this study can provide some support for developing an efficient technique for gender identification in duck production. [ABSTRACT FROM AUTHOR]

In the cigarette manufacturing industry, machine vision and artificial intelligence algorithms have been employed to improve production efficiency by detecting product defects. However, achieving both high accuracy and real-time defect detection for cigarettes with complex patterns remains a challenge. To address these issues, this study proposes a model based on RESNET18, combined with a feature enhancement algorithm, to improve detection accuracy. Additionally, a method is designed to deploy the model on a field-programmable gate array (FPGA) with high parallel processing capabilities to achieve high-speed detection. Experimental results demonstrate that the proposed detection model achieves a detection accuracy of 95.88% on a cigarette filter defect dataset with an end-to-end detection speed of only 9.38 ms. [ABSTRACT FROM AUTHOR]

UAVs are commonly utilized for the detection of insulator faults in transmission lines. The successful execution of such missions depends on two pivotal factors; the detection of insulator faults, which reduces losses in transmission lines, and the monitoring of the status of the UAV during the mission. Due to the vulnerability of UAV wings to defects, particularly lagging defects, it is crucial for the UAV to promptly land if it has loose wings to prevent severe damage. This article employs vibration data acquired from a wing-mounted sensor on the UAV to monitor its status. To determine the state of the UAV, a novel classifier is introduced, which aggregates votes from support vector machines (SVM), probabilistic neural networks (PNN), and deep neural networks (DNN) using Type-1 and Interval Type-2 Takagi–Sugeno–Kang Fuzzy System. Furthermore, the UAV-captured images of insulator in transmission lines are utilized for insulator fault detection. A similar approach is adopted for insulator fault detection, except that Multilayer Perceptron (MLP) is used instead of PNN, and ResNet-50 (Residual Network) is employed for feature extraction from insulator images. The software's cell phone interface, designed specifically for mobile devices, presents the graphical representation of UAV status and insulator fault detection. Furthermore, the generalizability of the proposed classifier for other applications is evaluated using two test datasets sourced from the UCI Machine Learning repository. The findings indicate that the proposed method performs better than renowned classifiers such as PNN, MLP, ANFIS, Fuzzy classifier, SVM, and ResNet-50. [ABSTRACT FROM AUTHOR]

This study introduces a deep self-organizing map neural network based on level-set (LS-SOM) for the customization of a shoe-last defined from plantar pressure imaging data. To alleviate the over-segmentation problem of images, which refers to segmenting images into more subcomponents, a domain-based segmentation model of plantar pressure images was constructed. The domain growth algorithm was subsequently modified by optimizing its parameters. A SOM with 10, 15, 20, and 30 hidden layers was compared and validated according to domain growth characteristics by using merging and splitting algorithms. Furthermore, we incorporated a level set segmentation method into the plantar pressure image algorithm to enhance its efficiency. Compared to the literature, this proposed method has significantly improved pixel accuracy, average cross-combination ratio, frequency-weighted cross-combination ratio, and boundary F1 index comparison. Using the proposed methods, shoe lasts can be designed optimally, and wearing comfort is enhanced, particularly for people with high blood pressure. [ABSTRACT FROM AUTHOR]

The objective of this work is to predict the methane hydrate phase boundary equilibrium temperature in the presence of ionic liquids (ILs) using machine learning techniques to overcome the limitations of the existing empirically proposed models. To achieve the objectives of this work, five deep neural networks (DNN) algorithms; Adadelta, Ftrl, Adagrad, Adam, and RMSProp coupled with six activation functions (elu, leaky relu, sigmoid, relu, tanh, and selu) were used on 610 experimental datasets from literature. The independent variables used to predict the ILs methane hydrate boundary temperature were pressure (2.39–100.43 MPa), concentration (0.10–50 wt.%), and ILs molecular weight (91.11–339.50 gmol−1). The study revealed that Adadelta DNN optimization algorithm and elu activation functions gave the best predictions with an average RMSE of 0.6727 and 0.6989, respectively. The findings suggest that the use of Adadelta coupled with elu accurately predicts the methane hydrate phase boundary condition in the presence of ionic liquids. The excellent performance of Adadelta and elu resides in their ability to predict exponential data trends which is the fundamental behavior of hydrate phase behavior condition. This work pioneered the use of machine learning techniques to predict hydrate behavior conditions in IL systems. Thus, the findings in this work will enhance the development of simple hydrate phase behavior properties predictive software for IL systems. [ABSTRACT FROM AUTHOR]

Filter pruning is an essential method for compressing and accelerating deep neural networks on computationally restricted devices. Despite recognizing the high correlation between filter redundancy and samples, existing methods primarily focus on independently searching for optimal subnetworks from individual input while ignoring the relationships among different inputs. In this paper, we propose a novel approach called Gated Filter Pruning based on Sample Manifold Relationships, which exploits and aligns the manifold relationships of all samples during training to obtain an optimal subnetwork. Firstly, we introduce a Gated Filter Normalization Module (GFNM) that excavates the manifold information of each sample, applicable to the operator level without adding many additional parameters. GFNM incorporates explainable control variables jointly optimized with convolutional weights, explicitly determining the competition and cooperation among filters during training. Subsequently, Manifold Regularized Pruning Module (MRPM) measures the manifold relationships between samples and subnetworks, efficiently regularizing the solution space of sample-network pairs. The manifold relationships between samples and subnetworks are aligned in training to derive an effective subnetwork for all input samples. Extensive experimental results validate the effectiveness of our method, demonstrating competitive performance in terms of accuracy and computational cost compared to state-of-the-art (SOTA) methods. [ABSTRACT FROM AUTHOR]

Dropout refers to the phenomenon of students leaving school before completing their degree or program of study. Dropout is a major concern for educational institutions, as it affects not only the students themselves but also the institutions' reputation and funding. Dropout can occur for a variety of reasons, including academic, financial, personal, and social factors. Therefore, understanding the factors that contribute to dropout and developing effective strategies to prevent it is a critical challenge for educational institutions. In this study, we propose a hybrid deep learning model based on Long Short-Term Memory and Deep Neural Network algorithms for school dropout prediction. The proposed model was compared with previous works and several other machine learning algorithms, including Deep Neural Network (DNN), K-Nearest Neighbors (KNN), Naive Bayes (NB), Multi-Layer Perceptron (MLP), Decision Trees (DT), Support Vector Machine (SVM), and Random Forest (RF). The results showed that the proposed DNN-LSTM model outperforms the other models in terms of accuracy and efficiency. [ABSTRACT FROM AUTHOR]

This study proposes a Privacy-Preserving Deep Neural Network for Credit Risk Prediction (PPDNN-CRP) framework that leverages homomorphic encryption (HE) to ensure data privacy throughout the credit risk prediction process. The PPDNN-CRP framework employs the Paillier homomorphic encryption scheme to secure sensitive loan application data during both the training and inference phases. Implemented using TensorFlow for deep neural network operations and TenSEAL for HE, the framework uses the Kaggle loan dataset to evaluate its performance. The results show that PPDNN-CRP achieved an accuracy of 80.48%, demonstrating competitive performance compared to Privacy-Preserving Logistic Regression (PPLR) at 77.23% and a slight decrease from the non-private DNN-CRP model at 86.18%. The model exhibited strong metrics with a precision of 0.84, recall of 0.91, F1-score of 0.87, and an AUC of 0.74. Security analysis confirms that PPDNN-CRP effectively defends against various privacy attacks, including poisoning, evasion, membership inference, model inversion, and model extraction, through robust encryption techniques and privacy measures. This framework offers a promising approach for achieving high-quality credit risk prediction while maintaining data privacy and complying with legal and ethical standards. [ABSTRACT FROM AUTHOR]

Humidity sensing has become an essential part of the process industry and the likewise essential device for domestic appliances. Humidity sensors are classified into capacitive and resistive based on their operating principles, capacitance-based humidity sensors dominate the market for their ease of use, simplicity of working, and reduced cost. In this investigation, a series of carbon fiber reinforced epoxy composites has been developed for use as a dielectric material in capacitance-based humidity sensors due to their low-cost, ease of fabrication coupled with excellent mechanical and electrical properties. The major limitation of carbon fiber-reinforced composites is their susceptibility to delamination, loss of stiffness, and overall failure of the fiber structures. To overcome these limitations, cross-linking and structural modification of the laminates was accomplished by gamma (γ) radiation exposure at dose levels of 1000 and 3000 kGy. γ-radiation exposure increases the volume resistivity and decreases the AC conductivity at higher frequencies due to reduction in free volume and hindered charge carrier transport. Radiation exposure increases the capacitance at 1000 kGy of γ-radiation by 6–32 times, and up to 5 times at 3000 kGy over the values non-irradiated laminate in the frequency range of 100–1000 Hz. The increase in capacitance due to radiation exposure is attributed to increase in dipoles resulting from orientation polarization, and chain scission. Higher sensitivity of 500 pF/%RH in achieved in laminate exposed to 1000 kGy, and 380 pF/%RH in laminate exposed to 3000 kGy over 200 pF/%RH of the base laminate. γ-radiation exposure does not alter the response time and it ranges from 60 to 65 s and the hysteresis is not affected by γ-radiation. Long term drift in capacitance is minimal and is unaffected by γ-radiation. [ABSTRACT FROM AUTHOR]

In this work, small-sized CaO–PdNps nanocomposites were synthesized using Lawsonia inermis leaf extract as a reducing and capping agent. X-ray diffraction (XRD), transmission electron microscopy (TEM), scanning electron microscopy (SEM), and Fourier transform infrared spectroscopy (FTIR) were used to analyze the produced nanocomposites. Based on these investigations, the average particle size was found to be 8±12nm, with a predominantly spherical morphology. The resulting nanocomposite was then tested for catalytic activity in methylene blue reduction. The mass of the catalyst and dye, the concentration of sodium borohydride ([NaBH4]), and the reaction time were all optimized using a central composite design (CCD). A significant correlation between predicted and experimental degradation percentages was shown by the CCD model (R2=0.9901), which also proved statistically significant (p-value=0.0016). Furthermore, a deep learning neural network (DLNN) was used to improve the prediction accuracy over and beyond CCD. There was a good connection between the two models as a result of using the CCD output as DLNN validation data. Mean absolute error (MAE), mean squared error (MSE), root mean squared error (RMSE), mean absolute percentage error (MAPE), and R2 were used to assess the performance of the DLNN. The results (MAE=0.001, MSE=0.01, RMSE=0.001, MAPE=0.001, R2=0.999) validated that all four metrics were successful in forecasting the percentage of degradation. [ABSTRACT FROM AUTHOR]

In this study, our goal is to integrate classical mathematical models with deep neural networks by introducing two novel deep neural network models for image segmentation known as Double-well Nets. Drawing inspirations from the Potts model, our models leverage neural networks to represent a region force functional. We extend the well-known MBO (Merriman--Bence--Osher) scheme to solve the Potts model. The widely recognized Potts model is approximated using a doublewell potential and then solved by an operator-splitting method, which turns out to be an extension of the well-known MBO scheme. Subsequently, we replace the region force functional in the Potts model with a UNet-type network, which is data-driven and is designed to capture multiscale features of images, and also introduce control variables to enhance effectiveness. The resulting algorithm is a neural network activated by a function that minimizes the double-well potential. What sets our proposed Double-well Nets apart from many existing deep learning methods for image segmentation is their strong mathematical foundation. They are derived from the network approximation theory and employ the MBO scheme to approximately solve the Potts model. By incorporating mathematical principles, Double-well Nets bridge the MBO scheme and neural networks, and offer an alternative perspective for designing networks with mathematical backgrounds. Through comprehensive experiments, we demonstrate the performance of Double-well Nets, showcasing their superior accuracy and robustness compared to state-of-the-art neural networks. Overall, our work represents a valuable contribution to the field of image segmentation by combining the strengths of classical variational models and deep neural networks. The Double-well Nets introduce an innovative approach that leverages mathematical foundations to enhance segmentation performance. [ABSTRACT FROM AUTHOR]

A few decades ago, quality assessment was a significant challenge in agriculture, even with tiny amounts of agricultural products. They are subject to diseases, pesticides, and environmental variables that have led to challenges to quality and safety. The quality evaluation of these items serves a crucial function in quality control and increases productivity before consumption. This paper proposes a deep neural network for solving Indian tomato diversity problems based on the assessment of their features. This study was carried out using four different categories of south Indian cherry tomatoes, i.e., spot, BER, calyx, and non-calyx. Thirty (30) relevant features were taken from an RGB-HIS color space: textural color moments, gray level co-occurrence matrix (GLCM), and defective proportion. They were then normalized in MS-EXCEL with a Min-Max function and fed into a multilayer perception deep neural network (MLP-DNN) and a convolutional neural network (CNN), respectively. The MLP-DNN was made with ten (10) hidden layers and was used with a fine-tuned convolutional neural network (CNN) of Alex net architecture with eight (8) learned layers to do grading and detection separately using MATLAB 2018a software. The results of training, testing, and validating both networks showed that they did an excellent job of solving the two-class problem (good and bad), with an overall prediction accuracy of more than 97% and a bit of loss function in both cases. However, the algorithms can be improved and employed to detect defects and sort fruits or vegetables of all kinds in the agro-industry with a larger dataset to reduce overfitting and increase validation accuracy in the future. [ABSTRACT FROM AUTHOR]

To better apply deep convolutional neural networks for expression recognition in UAV-enabled B5G/6G networks, we propose a deep network expression recognition method based on transfer learning and fine-tuning on facial expression datasets. Initially, we train our model on a large-scale facial attribute dataset and subsequently fine-tune it on a facial expression dataset. This strategy effectively lowers the high costs and dependence associated with annotating facial expression datasets, while enhancing the accuracy and training speed of our model. This method is particularly suited for UAVs equipped with on-board cameras and image processing capabilities, enabling real-time expression recognition for various applications such as crowd monitoring, search and rescue, and human-UAV interaction. Compared to traditional methods that train exclusively on facial expression datasets, our method significantly reduces the number of training iterations on facial expression datasets and significantly improves the generalization ability of the model, especially for UAV applications. We use a large-scale facial attribute dataset, which is more closely related to the facial expression recognition task, as our source dataset, forming a contrast with methods that typically use a facial recognition dataset as the transfer learning source dataset. The experimental results on the CK + dataset, integrated with UAV-enabled B5G/6G networks, show that our method achieves a facial expression recognition accuracy of 97.6%, a significant improvement over the 97.3% accuracy rate of methods that only train on facial expression datasets, with less training time as well. [ABSTRACT FROM AUTHOR]

Precoding is a critical signal processing technique used in wireless communication systems to enhance transmission performance. This paper initially provides a brief overview of various conventional precoding algorithms. The exploration includes various nonlinear precoding algorithms which outperform linear techniques in high signal-to-noise ratio (SNR) scenarios. Moreover, the paper delves into practical considerations and provides insights into selecting the most suitable technique for specific communication scenarios. The deep learning-based hybrid precoder is designed using the MDL-AltMin algorithm and simulated results show a spectral efficiency of (SE) 40.96 dB compared with different precoding algorithms. The precoder model is modified with a Hybrid Precoding Net algorithm with a spectral efficiency of 21.07 dB for high values of SNR such as −30 and again compared with deep learning precoders. The HPNet model is proposed with a spectral efficiency of 23.04 dB with and without CSI (Channel State Information) at the transmitter. Furthermore, the HPNet model is compared with optimal digital precoders. [ABSTRACT FROM AUTHOR]

Robotic fruit harvesting has great potential to revolutionize agriculture, but detecting cherry tomatoes in farming environments still faces challenges in accuracy and efficiency. To overcome the shortcomings of existing cherry tomato detection methods for harvesting, this study introduces a deep-learning-based cherry tomato detection scheme for robotic harvesting in greenhouses using multimodal RGB-D perception and an improved YOLOv7-tiny Cherry Tomato Detection (YOLOv7-tiny-CTD) network, which has been modified from the original YOLOv7-tiny by eliminating the "Objectness" output layer, introducing a new "Classness" method for the prediction box, and incorporating a new hybrid non-maximum suppression. Acquired RGB-D images undergo preprocessing such as color space transformation, point cloud normal vector angle computation, and multimodal regions of interest segmentation before being fed into the YOLOv7-tiny-CTD. The proposed method was tested using an AGV-based robot in a greenhouse cherry tomato farming facility. The results indicate that the multimodal perception and deep learning method improves detection precision and accuracy over existing methods while running in real time, and the robot achieved over 80% successful picking rates in two-trial mode in the greenhouse farm, showing promising potential for practical harvesting applications. [ABSTRACT FROM AUTHOR]

As a crucial spatial decision support tool, Geographic Information Systems (GISystems) are widely used in fields such as digital watersheds, resource management, environmental assessment, and regional governance, with their core strength lying in the integration of geographic simulation models from various disciplines, enabling the analysis of complex geographical phenomena and the resolution of comprehensive spatial problems. With the rapid advancement of artificial intelligence, deep neural network-based geographic simulation models (DNN-GSMs) have increasingly replaced traditional models, offering significant advantages in simulation accuracy and inference speed, and have become indispensable components in GISystems. However, existing integration methods do not adequately account for the specific characteristics of DNN-GSMs, such as their formats and input/output data types. To address this gap, we propose a novel tight integration framework for DNN-GSMs, comprising four key interfaces: the data representation interface, the model representation interface, the data conversion interface, and the model application interface. These interfaces are designed to describe spatial data, the simulation model, the adaptation between spatial data and the model, and the model's application process within the GISystem, respectively. To validate the proposed method, we construct a spatial morphology simulation model based on CNN-LSTM, integrate it into a GISystem using the proposed interfaces, and conduct a series of predictive experiments on island morphology evolution. The results demonstrate the effectiveness of the proposed integration framework for DNN-GSMs. [ABSTRACT FROM AUTHOR]

An effective approach to addressing the speech separation problem is utilizing a time–frequency (T-F) mask. The ideal binary mask (IBM) and ideal ratio mask (IRM) have long been widely used to separate speech signals. However, the IBM is better at improving speech intelligibility, while the IRM is better at improving speech quality. To leverage their respective strengths and overcome weaknesses, we propose an ideal threshold-based mask (ITM) to combine these two masks. By adjusting two thresholds, these two masks are combined to jointly act on speech separation. We list the impact of using different threshold combinations on speech separation performance under ideal conditions and discuss a reasonable range for fine tuning the thresholds. By using masks as a training target, to evaluate the effectiveness of the proposed method, we conducted supervised speech separation experiments applying a deep neural network (DNN) and long short-term memory (LSTM), the results of which were measured by three objective indicators: the signal-to-distortion ratio (SDR), signal-to-interference ratio (SIR), and signal-to-artifact ratio improvement (SAR). Experimental results show that the proposed mask combines the strengths of the IBM and IRM and implies that the accuracy of speech separation can potentially be further improved by effectively leveraging the advantages of different masks. [ABSTRACT FROM AUTHOR]

Financial fraud detection has been an urgent technical demand in cyberspace. It highly relies on clear extraction and deep representation toward complex relationships inside financial social networks. As consequence, this study combines both knowledge graph and deep learning to deal with such issue. Thus, an intelligent financial fraud detection model based on knowledge graph guidance and deep neural network is proposed in this paper. First, a new knowledge graph based on financially related systems is constructed, which includes multiple entities and relationships. Then, an adversarial learning-based neural network structure is formulated to extract financial attributes. Finally, the detection results can be output according to the extracted factors. Empirically, the proposal is implemented on a real-world dataset for performance evaluation. The experimental results show that it has higher accuracy and effectiveness compared to traditional fraud detection methods. The proposed detection model can not only identify known fraudulent behaviors, but also predict potential fraud patterns based on consumer habits, thereby improving the security and reliability of financial transactions. It can also update the knowledge graph in real-time, enabling it to cope with emerging fraud methods and variants. [ABSTRACT FROM AUTHOR]