Search

Purpose: Paper-based artifacts hold significant cultural and social values. However, paper is intrinsically fragile to microorganisms, such as mold, due to its cellulose composition, which can serve as a microorganisms' nutrient source. Mold not only can damage papers' structural integrity and pose significant challenges to conservation works but also may subject individuals attending the contaminated artifacts to health risks. Current approaches for strain identification usually require extensive training, prolonged time for analysis, expensive operation costs, and higher risks of secondary damage due to sampling. Thus, in current conservation practices with mold-contaminated artifacts, little pre-screening or strain identification was performed before mold removal, and the cleaning techniques are usually broad-spectrum rather than strain-specific. With deep learning showing promising applications across various domains, this study investigated the feasibility of using a convolutional neural network (CNN) for fast in-situ recognition and classification of mold on paper. Methods: Molds were first non-invasively sampled from ancient Xuan Paper-based Chinese books from the Qing and Ming dynasties. Strains were identified using molecular biology methods and the four most prevalent strains were inoculated on Xuan paper to create mockups for image collection. Microscopic images of the molds as well as their stains situated on paper were collected using a compound microscope and commercial microscope lens for cell phone cameras, which were then used for training CNN models with a transfer learning scheme to perform the classification of mold. To enable involvement and contribution from the research community, a web interface that actuates the process while providing interactive features for users to learn about the information of the classified strain was constructed. Moreover, a feedback functionality in the web interface was embedded for catching potential classification errors, adding additional training images, or introducing new strains, all to refine the generalizability and robustness of the model. Results & Conclusion: In the study, we have constructed a suite of high-confidence classification CNN models for the diagnostic process for mold contamination in conservation. At the same time, a web interface was constructed that allows recurrently refining the model with human feedback through engaging the research community. Overall, the proposed framework opens new avenues for effective and timely identification of mold, thus enabling proactive and targeted mold remediation strategies in conservation. [ABSTRACT FROM AUTHOR]

The use of Artificial Intelligence (AI) to preserve and promote cultural heritage has experienced significant growth in recent years. Among the various areas of cultural heritage, numismatics have emerged as a particularly promising field where we can develop AI solutions. Numismatics refers to the study of coins, tokens, paper money, and medals, which play a critical role in understanding human history and culture. However, there are still limited resources available to help researchers and collectors in the identification of coins. This is due to the vast number of coins in circulation, which presents a significant challenge in developing smart tools for classification tasks. This paper aims to provide a contribution to this setting. In particular, we start by creating a new dataset called EURO-Coin, which consists of images showing the side of coins with reliefs and is designed to facilitate the training and testing of AI models for euro coin classification. Then, we propose two approaches that leverage Convolutional Neural Networks and self-attention layers to classify the country and value of the coins. In our experiments, we obtain an accuracy of 86.9% for country classification and an accuracy of 96.4% for value classification. Finally, we conduct an ablation study to evaluate the impact of the preprocessing activities and attention layers in our approaches. [ABSTRACT FROM AUTHOR]

This paper provides an overview of the applications of Convolutional Neural Networks (CNN) in the railway maintenance industry. Our research covers specifically the subdomain of railway track maintenance. In this study, we have analyzed the state-of-the-art of CNNs applied to railway track maintenance by conducting an extensive literature review, summarizing different tasks and problems related to the topic and presenting solutions based on CNNs with a special emphasis on the data used to create these models. The results of our research show different applications of CNNs within the scope, including the detection of defects in the surface of railway rails and railway track components, such as fasteners, joints, sleepers, switches and crossings, as well as the recognition of track components, and the continuous monitoring of railway tracks. The architecture of CNNs is fitting to learning spatial hierarchies of features directly from the input data, making them of great use for Computer Vision and other applications related to the topic at hand. The implementation of IoT devices and smart sensors aid the collection of real-time data which can be used to feed powerful CNN models to recognize patterns and identify complex events related to the maintenance of railway tracks. This and more insights are discussed in detail within the contents of this paper. [ABSTRACT FROM AUTHOR]

Forest fire is a severe natural disaster, which leads to the destruction of forest ecology. At present, fire detection technology represented by convolutional neural network is widely used in forest resource protection, which can realize rapid analysis. However, in forest flame and smoke detection tasks, due to continuous expansion of the target range, a better detection effect cannot be achieved. This paper proposes an improved YOLOX method for multi-scale forest fire detection. This method proposes a novel feature pyramid model to reduce the information loss of high-level forest fire feature maps and enhance the representation ability of feature pyramids. Moreover, the method applies a small object data augmentation strategy to enrich the forest fire dataset, making it more suitable for the actual forest fire scene. According to the experimental results, the mAP of the model proposed in this paper reaches 79.64%, which is about 4.89% higher than the baseline network YOLOX. The method improves the accuracy of forest fire detection, reduces false alarms, and is suitable for real scenarios of forest fires. [ABSTRACT FROM AUTHOR]

With the establishment of the "double carbon" goal, various industries are actively exploring ways to reduce carbon emissions. Cloud data centers, represented by cloud computing, often have the problem of mismatch between load requests and resource supply, resulting in excessive carbon emissions. Based on this, this paper proposes a complete method for cloud computing carbon emission prediction. Firstly, the convolutional neural network and bidirectional long-term and short-term memory neural network (CNN-BiLSTM) combined model are used to predict the cloud computing load. The real-time prediction power is obtained by real-time prediction load of cloud computing, and then the carbon emission prediction is obtained by power calculation. Develop a dynamic server carbon emission prediction model, so that the server carbon emission can change with the change of CPU utilization, so as to achieve the purpose of low carbon emission reduction. In this paper, Google cluster data is used to predict the load. The experimental results show that the CNN-BiLSTM combined model has good prediction effect. Compared with the multi-layer feed forward neural network model (BP), long short-term memory network model (LSTM), bidirectional long short-term memory network model (BiLSTM), modal decomposition and convolution long time series neural network model (CEEMDAN-ConvLSTM), the MSE index decreased by 52 % , 50 % , 34 % and 45 % respectively. [ABSTRACT FROM AUTHOR]

Defect detection systems based on machine vision have been widely used as an essential part of intelligent manufacturing systems. However, in traditional object detection methods that rely on images as input, differences in defect areas, blurred images, and complex background interference can seriously impair detection accuracy. To meet these challenges, this paper proposed a dual-path neural network based on shearlet transform (STDPNet) by taking advantage of shearlet transform in multi-scale analysis and combining it with the improved object detection algorithm proposed in this paper. First, images are multi-scale and multi-directional decomposed with shearlet transform, and multi-directional sub-band information is input to the detection network instead of image information. Then, this paper proposed a dual-path object detection network for the differences between different frequency bands and introduced a transfer learning strategy between paths to improve the model performance. Finally, the training results on the NEU surface defect public dataset show that the mean average precision of STDPNet achieves 86.81% at a detection speed of 44.45 f/s, which exceeds that of Faster R-CNN by 12%. Experiments on different datasets prove that the accuracy is significantly superior to other models, and the proposed method is more advantageous compared to other models in large, fuzzy, and indistinguishable defect types. [ABSTRACT FROM AUTHOR]

Complex signals can be viewed as compositions of numerous sine waves with different frequencies and amplitudes. As the fundamental unit of perceiving image features, traditional Convolutional Neural Networks (CNNs) typically employ convolutional kernels without direct constraints. However, human perception of the world is inherently structured, relevant studies have indicated that considering the potential underlying correlation structures among data features or variables holds significant value, with particular emphasis on symmetric transformations in representation learning gaining attention in the field of neuroscience. Inspired by the process of human visual concept perception and classic image feature detection operators, when the convolution kernels have a similar scale to the objects, they produce larger responses. This paper introduces constraints on model parameters using centrosymmetric convolutional kernels. Under the influence of nonlinear combinations and local connections in neural networks, these kernels can diversely represent or perceive local image features. Experimental comparisons are conducted, and it is observed that the approach proposed in this paper not only enhances the convergence and accuracy of model, but also reduces the model solution space, resulting in an interpretable perception of local concepts. Furthermore, comparative experiments with classic deep CNNs have demonstrated comparable performance. [ABSTRACT FROM AUTHOR]

In this paper, a method based on transfer learning is proposed to recover the three-dimensional shape of cultural relics faces from a single old photo. It can simultaneously reconstruct the three-dimensional facial structure and align the texture of the cultural relics with fewer training samples. The UV position map is used to represent the three-dimensional shape in space and act as the output of the network. A convolutional neural network is used to reconstruct the UV position map from a single 2D image. In the training process, the human face data is used for pre-training, and then a small amount of artifact data is used for fine-tuning. A deep learning model with strong generalization ability is trained with fewer artifact data, and a three-dimensional model of the cultural relic face can be reconstructed from a single old photograph. The methods can train more complex deep networks without a large amount of cultural relic data, and no over-fitting phenomenon occurs, which effectively solves the problem of fewer cultural relic samples. The method is verified by restoring a Chinese Terracotta Warrior with facial cracks and a Buddha with a broken nose. Other applications can be used in the fields such as texture recovery, facial feature extraction, and three-dimensional model estimation of the damaged cultural relics or sculptures in the photos. [ABSTRACT FROM AUTHOR]

Most face recognition networks based on convolutional neural networks are easily affected by nonlinear factors of expression and posture, and the single task cannot adapt to multi-task scenarios. In addition, there is still room for further utilization of convolution kernels. Therefore, this paper proposes a multi-task aided face recognition network with convolution kernel spatial collaboration. The network is based on GhostNet, and the auxiliary task branch of expression and posture is added, which reduces the influence of nonlinear factors of expression and posture on face recognition, and enables the network to recognize expression and posture so that the network can adapt to the situation of more tasks. On this basis, the convolution of convolution is improved to increase its generality on grouped convolution, and it is applied to the cheap operation of Ghost Module so that the network can further use the convolution kernel to learn additional feature maps. Finally, a multi-task feature fusion module is proposed, which combines the shared features extracted from the backbone network with the features of the branch network, and further utilizes the features between different tasks to improve the performance of each task. The proposed method was compared with the existing methods based on deep learning on different datasets. The experimental results show that the proposed method has better recognition performance in face recognition, and the model size and inference speed are close to the lightweight network, which is suitable for more tasks. [ABSTRACT FROM AUTHOR]

With the development of image feature matching technology, feature matching algorithms based on deep learning have achieved excellent results, but in scenarios with low texture or extreme perspective changes, the matching accuracy is still difficult to guarantee. In this paper, a superresolution reconstruction method based on a Residual-ESPCN (efficient subpixel convolutional neural network) approach is proposed based on LoFTR (local feature matching with transformers). The superresolution method is used to improve the interpolation method used in ASFF (adaptive spatial feature fusion) to increase the image resolution, enhance the detailed information of the image, and make the extracted features richer. Then, ASFF is introduced into the local feature extraction module of LoFTR, which can alleviate the inconsistency problem of information transmission between different scale features of the feature pyramid and lessen the amount of information lost during transmission from low- to high-resolution levels. Moreover, to improve the adaptability of the algorithm to different scenarios, OTSU is introduced to adaptively calculate the threshold of feature matching. The experimental results show that in different indoor or outdoor scenarios, our proposed algorithm for matching features can effectively improve the adaptability of feature matching and can achieve good results in terms of the area under the curve (AUC), accuracy and recall. [ABSTRACT FROM AUTHOR]

This paper deals with the problem of creating a keyword spotting (KWS) system with real-world audio data. The paper describes the different methods used to build KWS systems, deep learning models such as convolutional neural networks (CNN), transformers, etc. The paper also discusses the mainstream dataset for training and testing KWS models, Google Speech Commands. We conduct experiments on Google Speech Commands dataset and propose our method of creating a KWS dataset and that helps neural networks achieve better results in training on relatively small amounts of data. We also introduce an idea of a hybrid KWS inference system architecture that uses voice detection and light-weight speech recognition framework in attempt to boost its computational performance and accuracy. We conclude by noting that KWS is an important challenge in the field of speech recognition, and suggest that their method can be used to improve the performance of KWS systems in the circumstances of low amounts of training data. We also note that future research could focus on bettering the process of evaluating the models and improving the overall performance of KWS systems. [ABSTRACT FROM AUTHOR]

Due to the uncertainty of weather conditions and the nonlinearity of high-dimensional data, as well as the need for a continuous and stable power supply to the power system, traditional regression analysis and time series forecasting methods are no longer able to meet the high accuracy requirements of today's PV power forecasting. To significantly improve the prediction accuracy of short-term PV output power, this paper proposes a short-term PV power forecasting method based on a hybrid model of temporal convolutional networks and gated recurrent units with an efficient channel attention network (TCN-ECANet-GRU) using the generated data of an Australian PV power station as the research object. First, temporal convolutional networks (TCNs) are used as spatial feature extraction layers, and an efficient channel attention network (ECANet) is embedded to enhance the feature capture capability of the convolutional network. Then, the GRU is used to extract the timing information for the final prediction. Finally, based on the experimental validation, the TCN-ECANet-GRU method generally outperformed the other baseline models in all four seasons of the year according to three performance assessment metrics: the normalized root mean square error (RMSE), normalized mean absolute error (MAE) and coefficient of determination (R2). The best RMSE, MAE and R2 reached 0.0195, 0.0128 and 99.72%, respectively, with maximum improvements of 11.32%, 8.57% and 0.38%, respectively, over those of the suboptimal model. Therefore, the model proposed in this paper is effective at improving prediction accuracy. Using the proposed method, this paper concludes with multistep predictions of 3, 6, and 9 steps, which also indicates that the proposed method significantly outperforms the other models. [ABSTRACT FROM AUTHOR]

In modern deep learning, character recognition in images is a very important field of study due to its has many real life applications. The goal of this paper is to create the state-of-the-art character recognition model using a stacking ensemble of convolution neural networks (CNNs).To develop the proposed ensemble model, we evaluated several CNN models. The models were judged on how well they performed on the Chars74k dataset. The dataset contains 74,103 images divided into 62 classes with labels [A-Z], [a-z], and [0-9]. The accuracy distribution based on the dataset's subgroups (uppercase, lowercase, and digit) is shown in results. The proposed ensemble model achieves state-of-the-art performance with a maximum accuracy of 92.31% on complete dataset, 99.22% on Uppercase alphabets, 98.66% on Lowercase alphabets, 99.77% on Digits, 91.97% on Uppercase+Lowercase alphabets. On the complete and partial datasets, a comparison report between the proposed model and other existing approaches is also displayed. A comparative study of the proposed work and the previous methods is also shown in this paper, in order to demonstrate the effectiveness of the proposed work. [ABSTRACT FROM AUTHOR]

In many countries, agriculture plays a major role in the economy. The health of the crop is therefore very important, but there are many plant diseases that are difficult to diagnose. A close inspection is necessary in many cases, or an expert's advice is required. As a result, it is important to address diseases in plants. Several attempts have been made to develop programs that detect diseases in plants because of the ever-rising growth of computer vision and deep learning. This paper focuses on detecting diseases in guava fruits. We use a dataset with pictures of 4 common diseases found in the fruit namely Phytopthra, Red Rust, Scab, Styler and Root. With the help of transfer learning and Convolutional Neural Networks (CNN) this paper train various models on the dataset and compare the results. The results we evaluated using accuracy, precision, Recall and F1 score. We had multiple models with test accuracy of 99%, with highest accuracy of 99.62% for DenseNet169. The results of this study are also compared with previous methods. According to the results, the proposed methods achieved better results than the previous approach. [ABSTRACT FROM AUTHOR]

This paper aims to propose a table tennis target detection (TD) method based on deep learning (DL) and multi-scale feature fusion (MFF) to improve the detection accuracy of the ball in table tennis competition, optimize the training process of athletes, and improve the technical level. In this paper, DL technology is used to improve the accuracy of table tennis TD through MFF guidance. Initially, based on the FAST Region-based Convolutional Neural Network (FAST R-CNN), the TD is carried out in the table tennis match. Then, through the method of MFF guidance, different levels of feature information are fused, which improves the accuracy of TD. Through the experimental verification on the test set, it is found that the mean Average Precision (mAP) value of the target detection algorithm (TDA) proposed here reaches 87.3%, which is obviously superior to other TDAs and has higher robustness. The DL TDA combined with the proposed MFF can be applied to various detection fields and can help the application of TD in real life. [ABSTRACT FROM AUTHOR]

An edge-cloud computing collaborative dust concentration detection architecture is proposed for real-time operation of intelligent algorithms to reduce the warning delay. And, an end-to-end three-channel convolutional neural network (E2E-SCNN) method is proposed in the paper to facilitate intelligent monitoring and management of dust concentration in tobacco production workshops. This model, which includes three sub-networks-a local feature branch, a global feature branch, and a spatial feature branch, learns the detail texture, overall layout, and spatial distribution information of the input image respectively. A fusion of the three complementary features is performed at the end of the network for the final dust concentration regression prediction. The design, when compared with the single network structure that directly regresses the entire image, is shown to more fully represent the overall information of the image and enhance monitoring performance. A richly annotated image dataset of tobacco production workshops is constructed to verify the effectiveness of the proposed method. The prediction error of E2E-SCNN is compared with existing image estimation algorithms, dual-channel networks, and other methods on this dataset using indicators such as Mean Absolute Error (MAE) and R 2 . It is shown by the results that excellent performance is achieved by the E2E-SCNN algorithm, significantly surpassing other comparison methods. The paper demonstrates that the accuracy and robustness of dust concentration prediction can be greatly improved by using a three-channel convolutional neural network spatial information monitoring framework. This achievement provides an effective means for dust supervision and governance during the tobacco production process and offers a technical route that can be referred to for image analysis tasks in other similar fields. [ABSTRACT FROM AUTHOR]

Purpose: In the winding process of chemical fiber production, the winding machine was responsible for winding the post-treated primary fiber. The winding machine clamp head was used as the direct winding part, and its running state would directly affect the quality and production efficiency of the chemical fiber cake. With the increasing complexity and precision of chemical fiber equipment, the traditional method of fault diagnosis relying on manual experience has been unable to meet the needs of intelligent production. The winding equipment also has some problems, such as few fault samples, strong background noise interference and large speed fluctuation. Therefore, the study of a better intelligent fault diagnosis method is of great significance to improve the winding reliability of the clamp head and ensure the quality and production efficiency of the chemical fiber cake. Methods: In this paper, a fault diagnosis method based on Empirical Mode Decomposition Order Spectrum of vibration signal and Dual-Channel Convolutional Neural Network (EMDOS-DCCNN) was proposed and the corresponding fault diagnosis model was designed. The trigger interval sampling scheme was designed according to the characteristics of the changing working conditions of the winder. The vibration data at the base of the winder's clamp head was collected by data acquisition equipment from Zhongli Winder in a chemical fiber production plant in Zhejiang Province as the real measured data for rolling bearing diagnosis. The strong background noise and the influence of the variable speed of the winder were removed by Empirical Mode Decomposition (EMD) reconstruction of the original signal and order spectrum analysis. The dual-channel convolutional neural network model was used for fault diagnosis, and the dual-channel data was fused to further improve the fault diagnosis accuracy. Order analysis and validation set confusion matrix were used to evaluate the diagnostic model proposed in this paper. Results and Conclusion: The model was trained with vibration data collected at three rotational speeds and verified on data sets collected at other rotational speeds. The results showed that the fault diagnosis model of rolling bearings based on EMDOS-DCCNN could effectively remove noise, and a single diagnosis model could be used to diagnose rolling bearings at different rotational speeds. In the model evaluation experiment, the accuracy of the dual-channel CNN model was stable at 98.6%, which was 3.1% higher than that of the single-channel model. It had high fault identification accuracy and adaptability at different speeds, and the model was suitable for variable working conditions and had better robustness. [ABSTRACT FROM AUTHOR]

English speech incorporates numerous features associated with the speaker's emotions, offering valuable cues for emotion recognition. This paper begins by briefly outlining preprocessing approaches for English speech signals. Subsequently, the Mel-frequency cepstral coefficient (MFCC), energy, and short-time zero-crossing rate were chosen as features, and their statistical properties were computed. The resulting 250-dimensional feature fusion was employed as input. A novel approach that combined gated recurrent unit (GRU) and a convolutional neural network (CNN) was designed for emotion recognition. The bidirectional GRU (BiGRU) method was enhanced through jump-joining to create a CNN-Skip-BiGRU model as an emotion recognition method for English speech. Experimental evaluations were conducted using the IEMOCAP dataset. The findings indicated that the fusion features exhibited superior performance in emotion recognition, achieving an unweighted accuracy rate of 70.31% and a weighted accuracy rate of 70.88%. In contrast to models like CNN-long short-term memory (LSTM), the CNN-Skip-BiGRU model demonstrated enhanced discriminative capabilities for different emotions. Moreover, it stood favorably against several existing emotion recognition methods. These results underscore the efficacy of the improved method in English speech emotion identification, suggesting its potential practical applications. [ABSTRACT FROM AUTHOR]

Segmentation of whole slide histological images through the classification of tissue types of small fragments is an extremely relevant task in digital pathology, necessary for the development of methods for automatic analysis of whole slide histological images. The extremely large resolution of such images also makes the task of increasing image resolution relevant, which allows storing images at a reduced resolution and increasing it if necessary. Annotating whole slide images by histologists is complex and time-consuming, so it is important to make the most efficient use of the available data, both labeled and unlabeled. In this paper we propose a novel neural network method to simultaneously solve the problems of super-resolution of histological images from ×20 optical magnification to ×40 and classifying image fragments into tissue types at ×20 magnification. The use of a single encoder as well as the proposed neural network training scheme allows to achieve better results on both tasks compared to existing approaches. The PATH-DT-MSU WSS2v2 dataset presented for the first time in this paper was used for training and testing the method. On the test sample, an accuracy value of 0.971 and a balanced accuracy value of 0.916 were achieved in the classification task on 5 tissue types, for the super-resolution task, values of PSNR = 32.26 and SSIM = 0.89 were achieved. The source code of the proposed method is available at: https://github.com/Kukty/WSI_SR_CL. [ABSTRACT FROM AUTHOR]

The models based on convolutional neural network have achieved excellent results in image super-resolution by acquiring prior knowledge from a large number of images, but such models still have problems such as the features between layers in the depth network cannot be effectively fused, the number of parameters is too large, and cross-channel feature learning is impossible. Based on this, a deep recursive residual channel attention network (DRRCAN) model was proposed in this paper. To solve the problem that the information between different layers in the deep network cannot be fused effectively, this paper constructs a channel feature fusion module, which can effectively fuse the feature information of different layers. To solve the problem that the parameters increase sharply due to the increase of network depth, recursive blocks are adopted in this paper, which greatly reduces the number of parameters in the deep network. The channel attention is integrated to enable the model to learn features across channels. In addition, to avoid gradient explosion or disappearance, residual modules, long skip connections are introduced to improve the stability and generalization ability of the model. Extensive benchmark evaluations validate the superiority of the proposed DRRCAN model compared with existing algorithms. [ABSTRACT FROM AUTHOR]

This paper proposes an improved strategy for the MobileNetV2 neural network(I-MobileNetV2) in response to problems such as large parameter quantities in existing deep convolutional neural networks and the shortcomings of the lightweight neural network MobileNetV2 such as easy loss of feature information, poor real-time performance, and low accuracy rate in facial emotion recognition tasks. The network inherits the characteristics of MobilenetV2 depthwise separated convolution, signifying a reduction in computational load while maintaining a lightweight profile. It utilizes a reverse fusion mechanism to retain negative features, which makes the information less likely to be lost. The SELU activation function is used to replace the RELU6 activation function to avoid gradient vanishing. Meanwhile, to improve the feature recognition capability, the channel attention mechanism (Squeeze-and-Excitation Networks (SE-Net)) is integrated into the MobilenetV2 network. Experiments conducted on the facial expression datasets FER2013 and CK + showed that the proposed network model achieved facial expression recognition accuracies of 68.62% and 95.96%, improving upon the MobileNetV2 model by 0.72% and 6.14% respectively, and the parameter count decreased by 83.8%. These results empirically verify the effectiveness of the improvements made to the network model. [ABSTRACT FROM AUTHOR]

Intelligent transport systems (ITS) provide various cooperative edge cloud services for roadside vehicular applications. These applications offer additional diversity, including ticket validation across transport modes and vehicle and object detection to prevent road collisions. Offloading among cooperative edge and cloud networks plays a key role when these resources constrain devices (e.g., vehicles and mobile) to offload their workloads for execution. ITS used different machine learning and deep learning methods for decision automation. However, the self-autonomous decision-making processes of these techniques require significantly more time and higher accuracy for the aforementioned applications on the road-unit side. Thus, this paper presents the new offloading ITS for IoT vehicles in cooperative edge cloud networks. We present the augmented convolutional neural network (ACNN) that trains the workloads on different edge nodes. The ACNN allows users and machine learning methods to work together, making decisions for offloading and scheduling workload execution. This paper presents an augmented federated learning scheduling scheme (AFLSS). An algorithmic method called AFLSS comprises different sub-schemes that work together in the ITS paradigm for IoT applications in transportation. These sub-schemes include ACNN, offloading, scheduling, and security. Simulation results demonstrate that, in terms of accuracy and total time for the considered problem, the AFLSS outperforms all existing methods. [ABSTRACT FROM AUTHOR]

Accurate and up-to-date information about the building stock is fundamental to better understand and mitigate the impact caused by catastrophic earthquakes, as seen recently in Turkey, Syria, Morocco and Afghanistan. Planning for such events is necessary to increase the resilience of the building stock and to minimize casualties and economic losses. Although in several parts of the world new constructions follow more strict compliance with modern seismic codes, a large proportion of existing building stock still demands a more detailed and automated vulnerability analysis. Hence, this paper proposes the use of computer vision deep learning models to automatically classify buildings and create large scale (city or region) exposure models. Such approach promotes the use of open databases covering most cities in the world (cf. OpenStreetMap, Google Street View, Bing Maps and satellite imagery), Therefore providing valuable geographical, topological and image data that may cheaply be used to extract valuable information to feed exposure models. Our previous work using deep learning models achieved, in line with the results from other projects, high classification accuracy concerning building materials and number of storeys. This paper extends the approach by: (i) implementing four CNN-based models to perform classification of three sets of different/extended buildings’ characteristics; (ii) training and comparing the performance of the four models for each of the sets; (iii) comparing the risk assessment results based on data extracted from the best CNN-based model against the results obtained with traditional ground data. In brief, the best accuracy obtained with the three tested sets of buildings’ characteristics is higher than 80%. Moreover, it is shown that the error resulting from using exposure models fed by automatic classification is not only acceptable, but also far outweighs the time and costs of obtaining a manual and specialised classification of building stocks. Finally, we recognize that automatic assessment of certain complex buildings’ characteristics compares to similar limitations of traditional assessments performed by specialized civil engineers, typically related with the identification of the number of storeys and the construction material. However, the identified limitations do not show worse results when compared against the use of manual buildings’ assessment.Article Highlights: Implement an AI/ML framework for automating the collection of buildings’ façades pictures annotated with several characteristics required by Exposure Models. Collect, process and filter a 4.239 pictures dataset of buildings’ façades, which was made publicly available. Train, validate and test several Deep Learning models using 3 sets of building characteristics to produce exposure models with accuracies above 80%. Use heatmaps to show which image areas are more activated for a given prediction, thus helping to explain classification results. Compare simulation results using the predicted exposure model and a manually created exposure model, for the same set of buildings. [ABSTRACT FROM AUTHOR]

Scene classification and recognition have always been one of the most challenging tasks of scene understanding due to the inherent ambiguity in visual scenes. The core of scene classification and recognition tasks is scene representation. Deep learning advances in computer vision, especially deep CNNs, have significantly improved scene representation in the last decade. Deep convolutional features extracted from deep CNNs provide discriminative representations of the images and are widely used in various computer vision tasks, such as scene classification. Deep convolutional features capture the appearance characteristics of the image and the spatial information about different image regions. Meanwhile, the semantic and context information obtained from high-level concepts about scene images, such as objects and their relationships, can significantly contribute to identifying scene images. Therefore, in this paper, we divide visual scenes into two categories, object-based and layout-based. Object-based scenes are scenes that have scene-specific objects and, based on those objects, can be described and identified. In contrast, the layout-based scenes do not have scene-specific objects and are described and identified based on the appearance and layout of the image. This paper proposes a new neural network model for representing and classifying visual scenes, which we call G-CNN (GNN-CNN). The proposed model includes two modules, feature extraction and feature fusion, and the feature extraction module composes of visual and semantic branches. The visual branch is responsible for extracting deep CNN features from the image, and the semantic branch is responsible for extracting semantic GNN features from the scene graph corresponding to the image. The feature fusion module is a novel two-stream neural network that fuses the CNN and GNN feature vectors to produce a comprehensive representation of the scene image. Finally, a fully-connected classifier classified the obtained comprehensive feature vector into one of the pre-defined categories. The proposed model has been evaluated on three benchmark scene datasets, UIUC Sports, MIT67, and SUN397, and obtained classification accuracy of 99.91%, 96.01%, and 85.32%, respectively. In addition, a new dataset named Scene40, which has been introduced in our previous paper, is also used for further evaluation of the proposed method. The comparison results based on classification accuracy criteria show that the proposed model can outperform the best previous methods on three benchmark scene datasets. [ABSTRACT FROM AUTHOR]

Furnace temperature and oxygen content are important parameters reflecting the combustion inside a circulating fluidized bed (CFB) boiler. Accurately predicting boiler output is a complex task due to the high noise and nonsmoothness of actual boiler input and output data. In this paper, a new hybrid convolutional neural network (CNN), bidirectional long short-term memory (biLSTM) network, and squeezing and excitation (SE) network prediction model is proposed to significantly improve the prediction accuracy of oxygen content and furnace temperature by combining the advantages of multiple deep learning networks. This network can extract spatiotemporal characteristics of input parameters such as coal feed to effectively predict boiler furnace temperature and oxygen content. CNNs can extract complex features such as dynamic and static nonlinearities between multiple variables affecting the furnace temperature and oxygen content, as well as high noise. The biLSTM network layer can efficiently handle the temporal information of irregular trends in modeling time series components; SE can extract the important information between channels through the feature relationships between channels for better overfeature extraction. The CNN-biLSTM-SE model can effectively solve the problem of nonlinear mapping complexity between inputs and outputs. Experiments show that the proposed CNN-biLSTM-SE model outperforms existing methods. The experimental results showed that the average MAPE errors for oxygen content prediction were CNN-biLSTM-SE (0.038), CNN-biLSTM with attention mechanism (AM) (0.043), CNN-biLSTM (0.051), CNN-LSTM (0.051), biLSTM (0.051), RNN (0.051), LSTM(0.0052), and CNN(0.0054). Extensive experiments in CFB boilers with oxygen content and furnace temperature show that the proposed CNN-biLSTM-SE model achieves better results in terms of goodness-of-fit, generalization ability and accuracy. [ABSTRACT FROM AUTHOR]

Over the past decade, the use of biometrics in security systems and other applications has grown in popularity. ECG signals in particular are attracting increased attention due to their characteristics, which are required for a trustworthy identification system. The majority of ECG-based person identification systems are evaluated without considering the health-state of the individuals. Few person identification systems consider person-by-person health-state annotation. This paper proposes a person identification system considering the health-state annotated ECG signals where each person's beats overlap among variant arrhythmia classes. This overlapping between the normal class and other arrhythmia classes grants the ability to isolate normal beats in the train set from the Arrhythmic beats in the test set. Therefore, this paper investigates the effect of arrhythmic heartbeats on biometric recognition. An effective lightweight CNN based on depth-wise separable convolution (DWSC) is proposed to enhance the performance of person identification for several common arrhythmia types using the MITBIH dataset. The proposed methodology has been tested on nine arrhythmia types and presents how different types of arrhythmia affect ECG-based biometric systems differently. The experimental results show excellent recognition performance (99.28%) on normal heartbeats and (93.81%) on arrhythmic heartbeats, outperforming other models in terms of mean accuracy. [ABSTRACT FROM AUTHOR]

In the actual production process of oil fields, the real-time and accurate acquisition of dynamic recorder data from the pumping unit is of great significance for the diagnosis of well failures. The traditional method of obtaining the card of the dynamometer usually includes installing a load sensor on the auxiliary head of the pumping unit. However, due to the harsh environment of the oil field production site, these load sensors often suffer from damage, distortion, and aging, resulting in large measurement errors and low reliability. This paper proposes a mixed model of pumping based on motor electrical parameter data and CNN convolutional neural network, which has good consistency with actual data in terms of predictive performance. Thus, the highlights of this paper can be summed up in two points: (1) Based on the mathematical model of the AC motor, the speed of the motor and the torque output of the motor are accurately estimated. (2) The convolutional neural network is introduced to compensate for the errors caused by the defects of the pumping unit mechanism model. [ABSTRACT FROM AUTHOR]

Identifying the root causes of part-to-part variation is a central problem in most six-sigma programs, especially of modern manufacturing processes. This is challenging as the sources and patterns of the variation are often unknown or previously unidentified. A small literature aims to address this problem by discovering unknown, previously unidentified variation sources, in a manner that helps understand their nature, from only a sample of measurement data. However, the common solution of this literature is unideal for this objective in terms of both methodology and metrology aspects. This paper proposes a convolutional generative modeling framework for optical scanning data to address this limitation. The proposed approach can correctly discover the true variation sources and visualize their individual patterns in two manufacturing examples, without any prior knowledge of the variation. The approach also outperforms state-of-the-art methods in these examples. [ABSTRACT FROM AUTHOR]

Sound event detection involves identifying sound categories in audio and determining when they start and end. However, in real-life situations, sound events are usually not isolated. When one sound event occurs, there are often other related sound events that take place as co-occurrences or successive occurrences. The timing relationship of sound events can reflect their characteristics. Therefore, this paper proposes a sound event detection method for traffic scenes based on a graph convolutional network, which considers this timing relationship as a form of multimodal information. The proposed method involves using the acoustic event window method to obtain co-occurrences or successive occurrences of relationship information in the sound signal while filtering out possible noise relationship information. This information is then represented as a graphical structure. Next, the graph convolutional neural network is improved to balance relationship weights between neighbors and itself and to avoid excessive smoothing. It is used to learn the relationship information in the graph structure. Finally, the convolutional recurrent neural network is used to learn the acoustic feature information of sound events, and the relationship information of sound events is obtained by multi-modal fusion to enhance the performance of sound event detection. The experimental results show that using multi-modal information with the proposed method can effectively improve the performance of the model and enhance the perception ability of smart cars in their surrounding environment while driving. [ABSTRACT FROM AUTHOR]

Epilepsy is one of the most well-known neurological disorders globally, leading to individuals experiencing sudden seizures and significantly impacting their quality of life. Hence, there is an urgent necessity for an efficient method to detect and predict seizures in order to mitigate the risks faced by epilepsy patients. In this paper, a new method for seizure detection and prediction is proposed, which is based on multi-class feature fusion and the convolutional neural network-gated recurrent unit-attention mechanism (CNN-GRU-AM) model. Initially, the Electroencephalography (EEG) signal undergoes wavelet decomposition through the Discrete Wavelet Transform (DWT), resulting in six subbands. Subsequently, time–frequency domain and nonlinear features are extracted from each subband. Finally, the CNN-GRU-AM further extracts features and performs classification. The CHB-MIT dataset is used to validate the proposed approach. The results of tenfold cross validation show that our method achieved a sensitivity of 99.24% and 95.47%, specificity of 99.51% and 94.93%, accuracy of 99.35% and 95.16%, and an AUC of 99.34% and 95.15% in seizure detection and prediction tasks, respectively. The results show that the method proposed in this paper can effectively achieve high-precision detection and prediction of seizures, so as to remind patients and doctors to take timely protective measures. [ABSTRACT FROM AUTHOR]

The automated replacement of empty tubes in the yarn bank is a critical step in the process of automatic winding machines with yarn banks, as the real-time detection of depleted yarn on spools and accurate positioning of empty tubes directly impact the production efficiency of winding machines. Addressing the shortcomings of traditional methods, such as poor adaptability and low sensitivity in optical and visual tube detection, and aiming to reduce the computational and detection time costs introduced by neural networks, this paper proposes a lightweight yarn spool detection model based on YOLOv8. The model utilizes Darknet-53 as the backbone network, and due to the dense spatial distribution of yarn spool targets, it incorporates large selective kernel units to enhance the recognition and positioning of dense targets. To address the issue of excessive focus on local features by convolutional neural networks, a bi-level routing attention mechanism is introduced to capture long-distance dependencies dynamically. Furthermore, to balance accuracy and detection speed, a FasterNeck is constructed as the neck network, replacing the original convolutional blocks with Ghost convolutions and integrating with FasterNet. This design minimizes the sacrifice of detection accuracy while achieving a significant improvement in inference speed. Lastly, the model employs weighted IoU with a dynamic focusing mechanism as the bounding box loss function. Experimental results on a custom yarn spool dataset demonstrate a notable improvement over the baseline model, with a high-confidence mAP of 94.2% and a compact weight size of only 4.9 MB. The detection speed reaches 223FPS, meeting the requirements for industrial deployment and real-time detection. [ABSTRACT FROM AUTHOR]

In the realm of prognosticating the remaining useful life (RUL) of pivotal components, such as aircraft engines, a prevalent challenge persists where the available historical life data often proves insufficient. This insufficiency engenders obstacles such as impediments in performance degradation feature extraction, inadequacies in capturing temporal relationships comprehensively, and diminished predictive accuracy. To address this issue, a 1D CNN-GRU prediction model for few-shot conditions is proposed in this paper. In pursuit of more comprehensive data feature extraction and enhanced RUL prognostication precision, the Convolutional Neural Network (CNN) is selected for its capacity to discern high-dimensional features amid the intricate dynamics of the data. Concurrently, the Gated Recurrent Unit (GRU) network is leveraged for its robust capability in extracting temporal features inherent within the data. We combine the two to construct a CNN-GRU hybrid network. Moreover, the integration of data distribution alongside correlation and monotonicity indices is employed to winnow the input of multi-sensor monitoring parameters into the CNN-GRU network. Finally, the engine RULs are predicted by the trained model. In this paper, experiments are conducted on a sub-dataset of the National Aeronautics and Space Administration (NASA) C-MAPSS multi-constraint dataset to validate the effectiveness of the method. Experimental results have demonstrated that this method has high accuracy in RUL prediction tasks, which can powerfully demonstrate its effectiveness. [ABSTRACT FROM AUTHOR]

The intelligent appearance quality classification method for Auricularia auricula is of great significance to promote this industry. This paper proposes an appearance quality classification method for Auricularia auricula based on the improved Faster Region-based Convolutional Neural Networks (improved Faster RCNN) framework. The original Faster RCNN is improved by establishing a multiscale feature fusion detection model to improve the accuracy and real-time performance of the model. The multiscale feature fusion detection model makes full use of shallow feature information to complete target detection. It fuses shallow features with rich detailed information with deep features rich in strong semantic information. Since the fusion algorithm directly uses the existing information of the feature extraction network, there is no additional calculation. The fused features contain more original detailed feature information. Therefore, the improved Faster RCNN can improve the final detection rate without sacrificing speed. By comparing with the original Faster RCNN model, the mean average precision (mAP) of the improved Faster RCNN is increased by 2.13%. The average precision (AP) of the first-level Auricularia auricula is almost unchanged at a high level. The AP of the second-level Auricularia auricula is increased by nearly 5%. And the third-level Auricularia auricula AP is increased by 1%. The improved Faster RCNN improves the frames per second from 6.81 of the original Faster RCNN to 13.5. Meanwhile, the influence of complex environment and image resolution on the Auricularia auricula detection is explored. [ABSTRACT FROM AUTHOR]

With the rapid development of artificial intelligence, hand gesture detection has gradually become a research hotspot in human–computer interaction. However, the traditional hand gesture detection has low robustness and detection accuracy, and brings the problem of privacy protection. Therefore, this paper proposes a Faster Region-based Convolutional Neural Network (F-RCNN) hand gesture detection method based on Frequency Modulated Continuous Wave (FMCW) radar with 3-Dimensions deep convolutional Generative Adversarial Networks (3-DCGAN). In particular, the range Doppler and angle of the hand gestures are calculated by FMCW radar. Then, the semantic label maps of the Range-Time-Map, Doppler-Time-Map and Angle-Time-Map images are respectively sent to the 3-DCGAN to expand the datasets. After that, the original and the 3-DCGAN generated images are sent to F-RCNN for jointly training. The classifier function is designed, and a learning ranking based assessment called Rank based Quality Score (RQS) is applied to improve the detection performance. The experimental results show that the mean average precision reaches to 80.8%. Moreover, the RQS for hand gesture detection is as high as 96.3%, which is increased by 5.8% compared to the traditional F-RCNN. [ABSTRACT FROM AUTHOR]

In convolutional neural networks, the second-order representation can effectively enhance the nonlinear modeling capability of first-order models, while the scale of parameters also increases significantly, resulting in overfitting and low efficiency. In this paper, a lightweight attention module with second-order information as well as a compression strategy of information matrix are developed to improve the classification ability of typical models without affecting or even improving the convergence speed. Specifically, the main contributions are (1) a novel module based on second-order channel attention is proposed, which utilizes multilayer one-dimensional convolution for feature representation, and directly learns the long-distance correlation between feature channels by adding a small number of parameters; (2) a Cholesky-DLT compression strategy is established for the second-order information matrix, which significantly reduces the parameter scale of the fully connected layer. These two modules can be easily integrated into end-to-end network structures and then applied to typical network models. The experimental results show that the combination of the proposed modules with the most common used first-order and second-order models can effectively improve the generalization performance without increasing the involved parameters. [ABSTRACT FROM AUTHOR]

In the field of remote sensing image interpretation, automatically extracting water body information from high-resolution images is a key task. However, facing the complex multi-scale features in high-resolution remote sensing images, traditional methods and basic deep convolutional neural networks are difficult to effectively capture the global spatial relationship of the target objects, resulting in incomplete, rough shape and blurred edges of the extracted water body information. Meanwhile, massive image data processing usually leads to computational resource overload and inefficiency. Fortunately, the local data processing capability of edge computing combined with the powerful computational resources of cloud centres can provide timely and efficient computation and storage for high-resolution remote sensing image segmentation. In this regard, this paper proposes PMNet, a lightweight deep learning network for edge-cloud collaboration, which utilises a pipelined multi-step aggregation method to capture image information at different scales and understand the relationships between remote pixels through horizontal and vertical spatial dimensions. Also, it adopts a combination of multiple decoding branches in the decoding stage instead of the traditional single decoding branch. The accuracy of the results is improved while reducing the consumption of system resources. The model obtained F1-score of 90.22 and 88.57 on Landsat-8 and GID remote sensing image datasets with low model complexity, which is better than other semantic segmentation models, highlighting the potential of mobile edge computing in processing massive high-resolution remote sensing image data. [ABSTRACT FROM AUTHOR]

Accurate prediction of sea level height is critically important for the government in assessing sea level risk in coastal areas. However, due to the nonlinear, time-varying and highly uncertain characteristics of sea level change data, sea level prediction is challenging. To improve the accuracy of sea level prediction, this paper uses a new swarm intelligence algorithm named the sparrow search algorithm (SSA), which can imitate the foraging behavior and antipredation behavior of sparrows, to determine optimal solutions. To avoid the algorithm falling into a local optimal situation, this paper integrates the sine–cosine algorithm and the Cauchy variation strategy into the SSA to obtain an algorithm named the SCSSA. The SCSSA is used to optimize the parameter values of the CNN-BiLSTM (convolutional neural network combined with bidirectional long short-term memory neural network) model; finally, a combined neural network model (named SCSSA-CNN-BiLSTM) is proposed. In this paper, the time series data of seven tidal stations located in coastal China are used for experimental analysis. First, the SCSSA-CNN-BiLSTM model is compared with the CNN-BiLSTM model to predict the time series data of SHANWEI Station. With respect to the training and test sets of data, the SCSSA-CNN-BiLSTM model outperforms the other models on all the evaluation metrics. In addition, the remaining six tide station datasets and five neural network models, including the SCSSA-CNN-BiLSTM model, are used to further study the performance of the proposed prediction model. Four evaluation indices including the root mean squared error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE) and coefficient of determination (R2) are adopted. For six stations, the RMSE, MAE, MAPE and R2 of SCSSA-CNN-BiLSTM model are ranged from 20.9217 ~ 27.8427 mm, 9.4770 ~ 17.8603 mm, 0.1322% ~ 0.2482% and 0.9119 ~ 0.9759, respectively. The experimental analysis results show that the SCSSA-CNN-BiLSTM model makes effective predictions at all stations, and the prediction performance is better than that of the other models. Even though the combination of SCSSA algorithm may increase the complexity of the model, indeed the proposed model is a new prediction method with good accuracy and robustness for predicting sea level change. [ABSTRACT FROM AUTHOR]

Deep convolutional neural networks can effectively improve the performance of single-image super-resolution reconstruction. Deep networks tend to achieve better performance than others. However, the deep CNNs will lead to a dramatic increase in the size of parameters, limiting its application on embedding and resource-constrained devices, such as smart phone. To address the common problems of blurred image edges, inflexible convolution kernel size selection and slow convergence during training procedure due to redundant network structure in image super-resolution algorithms, this paper proposes a lightweight single-image super-resolution network that fuses multi-level features. The components are mainly two-level nested residual blocks. To better extract features and reduce the number of parameters, each residual block adopts an asymmetric structure. Firstly, it expands twice and then compresses the number of channels twice. Secondly, in the residual block, the feature information of different channels is weighted and fused by adding an autocorrelation weight unit. The quality of the reconstructed image of the proposed method is superior to the existing image super-resolution reconstruction methods in both subjective perception and objective evaluation indicators, and the reconstruction performance is better when the factor is large. [ABSTRACT FROM AUTHOR]

Purpose: The main purpose of this paper is to change the structure of the SDP to include more fault information. Furthermore, improve the diagnostic accuracy and anti-noise performance of the bearing fault diagnosis method based on SDP. Methods: First, a multi-interval asymmetric dot pattern (MADP) is proposed by modifying the expression of SDP. Then, the improved SDP (multi-modal multi-interval asymmetric dot pattern, MMADP) is established by the MADP method fused with the multiple effective IMF components which are obtained through CEEMDAN decomposition. Finally, a bearing fault diagnosis model is established based on MMADP and convolutional neural network. Results: The effectiveness of the proposed fault diagnosis method is validated on the CWRU dataset. The results indicate that under Gaussian white noise with a signal-to-noise ratio (SNR) of above 4 dB and − 6 dB, the accuracy of the proposed fault diagnosis method reaches 100 and 93.3%, respectively. Conclusion: In this paper, a method (MADP) transforming time series signals into images is proposed, and a method for fault diagnosis of rolling bearings is formed through combination of CEEMDAN and CNN. The bearing fault diagnosis method has good anti-noise performance, and the MADP has potential value in the processing of sound signals. [ABSTRACT FROM AUTHOR]

The development of the Internet and big data has led to the emergence of graphs as an important data representation structure in various real-world scenarios. However, as data size increases, computational complexity and memory requirements pose significant challenges for graph embedding. To address this challenge, this paper proposes a multilevel embedding refinement framework (MERIT) based on large-scale graphs, using spectral distance-constrained graph coarsening algorithms and an improved graph convolutional neural network model that addresses the over-smoothing problem by incorporating initial values and identity mapping. Experimental results on large-scale datasets demonstrate the effectiveness of MERIT, with an average AUROC score 8% higher than other baseline methods. Moreover, in a node classification task on a large-scale graph with 126,825 nodes and 22,412,658 edges, the framework improves embedding quality while enhancing the runtime by 25 times. The experimental findings highlight the superior efficiency and accuracy of the proposed approach compared to other graph embedding methods. [ABSTRACT FROM AUTHOR]

Purpose: The mechanical fault diagnosis method based on deep learning mainly uses single-scale convolution kernels to extract fault features, which is difficult to extract fault feature information comprehensively. Under strong noise conditions, the performance of fault diagnosis using single-scale convolution kernels will decrease sharply. Methods: An intelligent fault diagnosis method based on multi-scale deep convolution neural network (MSD-CNN) model and data enhancement for strong noise is proposed in this paper. By the multi-scale cascade convolution kernels in the MSD-CNN, the multi-scale information of the original fault signal is extracted, and the ELU activation function is used to retain the negative information contained in the multi-scale information. By the data enhancement method for strong noise, the number and diversity of the MSD-CNN model training samples are improved, which enables the model to learn deeper features in the training stage. Results and Conclusion: Compared six common fault diagnosis models, the proposed model achieves the optimal diagnostic accuracy of 99.98% and 99.66% under normal conditions and strong noise conditions, respectively, which verifies the effectiveness of the proposed method. [ABSTRACT FROM AUTHOR]

This paper proposes an approach to pruning the parameters of convolutional neural networks using unsupervised pretraining. The authors demonstrate that the proposed approach makes it possible to reduce the number of configurable parameters of a convolutional neural network without loss of generalization ability. A comparison of the proposed approach and existing pruning techniques is made. The capabilities of the proposed algorithm are demonstrated on classical CIFAR10 and CIFAR100 computer vision samples. [ABSTRACT FROM AUTHOR]

Deep neural networks perform better than shallow neural networks, but the former tends to be deeper or wider, introducing large numbers of parameters and computations. We know that networks that are too wide have a high risk of overfitting and networks that are too deep require a large amount of computation. This paper proposed a narrow-deep ResNet, increasing the depth of the network while avoiding other issues caused by making the network too wide, and used the strategy of knowledge distillation, where we set up a trained teacher model to train an unmodified, wide, and narrow-deep ResNet that allows students to learn the teacher’s output. To validate the effectiveness of this method, it is tested on Cifar-100 and Pascal VOC datasets. The method proposed in this paper allows a small model to have about the same accuracy rate as a large model, while dramatically shrinking the response time and computational effort. [ABSTRACT FROM AUTHOR]

Sleep is essential for human physical and mental health. Sleep disorders are a significant threat to human health, and a large number of people in the world suffer from sleep disorders. Effective detection of sleep disorders is essential for the treatment of sleep disorders. Questionnaires and scale assessments are traditional methods of sleep disorder detection, which are subjective, time-consuming and prone to misdiagnosis. To detect sleep disorders quickly and accurately, a Multimodal Attention-Fusion Convolutional Neural Network is proposed in this paper. The network uses electroencephalography, electrooculography, electrocardiography, and electromyography signals to automatically identify healthy and five sleep disorders, namely insomnia, narcolepsy, periodic leg movement, rapid eye movement behaviour disorder and nocturnal frontal lobe epilepsy. First, multiple convolutional neural network branches are used to extract time-invariant features of multimodal signals. Then, a multi-scale attention module based on dilated convolutional networks and a squeeze and excite block is proposed for further extracting features with different scales and fusing feature information. Finally, a prediction module consisting of fully connected layers is used to detect sleep disorders. The accuracy, F1 score, and Kappa coefficient obtained on the Cyclic Alternating Pattern sleep dataset are 99.56%, 99.49% and 0.9942, respectively. Compared to the existing state-of-the-art studies, the method proposed in this paper has higher performance in sleep disorder detection. [ABSTRACT FROM AUTHOR]

In recent years, the field of bionics has attracted the attention of numerous scholars. Some models combined with biological vision have achieved excellent performance in computer vision and image processing tasks. In this paper, we propose a new bio-inspired lightweight contour detection network (BLCDNet) by combining parallel processing mechanisms of bio-visual information with convolutional neural networks. The backbone network of BLCDNet simulates the parallel pathways of ganglion cell–lateral geniculate nucleus and primary visual cortex (V1) area, realizing parallel processing and step-by-step extraction of input information, effectively extracting local features and detailed features in images, and thus improving the overall performance of the model. In addition, we design a depth feature extraction module combining depth separable convolution and residual connection in the decoding network to integrate the output of the backbone network, which further improves the performance of the model. We conducted a large number of experiments on BSDS500 and NYUD datasets, and the experimental results show that the BLCDNet proposed in this paper achieves the best performance compared with traditional methods and previous biologically inspired contour detection methods. In addition, BLCDNet still outperforms some VGG-based contour detection methods without pre-training and with fewer parameters, and it is competitive among all of them. The research in this paper also provides a new idea for the combination of biological vision and convolutional neural networks. [ABSTRACT FROM AUTHOR]

Objectives: This study aims to enhance supervised human activity recognition based on spatiotemporal graph convolutional neural networks by addressing two key challenges: (1) extracting local spatial feature information from implicit joint connections that is unobtainable through standard graph convolutions on natural joint connections alone. (2) Capturing long-range temporal dependencies that extend beyond the limited temporal receptive fields of conventional temporal convolutions. Methods: To achieve these objectives, we propose three novel modules integrated into the spatiotemporal graph convolutional framework: (1) a connectivity feature extraction module that employs attention to model implicit joint connections and extract their local spatial features. (2) A long-range frame difference feature extraction module that captures extensive temporal context by considering larger frame intervals. (3) A coordinate transformation module that enhances spatial representation by fusing Cartesian and spherical coordinate systems. Findings: Evaluation across multiple datasets demonstrates that the proposed method achieves significant improvements over baseline networks, with the highest accuracy gains of 2.76 % on the NTU-RGB+D 60 dataset (Cross-subject), 4.1 % on NTU-RGB+D 120 (Cross-subject), and 4.3 % on Kinetics (Top-1), outperforming current state-of-the-art algorithms. This paper delves into the realm of behavior recognition technology, a cornerstone of autonomous systems, and presents a novel approach that enhances the accuracy and precision of human activity recognition. [ABSTRACT FROM AUTHOR]

Integrating energy systems with information systems in smart grids offers a promising avenue for combating electricity theft by leveraging real-time data insights. Suspicious activity indicative of theft can be identified through anomalous consumption patterns observed in smart networks. However, a smart model is required for capturing and analysing the data intelligently to accurately detect electricity theft. In the paper, electricity theft has been detected using an encoder-decoder-based classifier that integrates two models of convolutional neural networks (CNN). The aim is to scan the strength of the data and built a smart model that analysed the connections in complex data and determine the pattern of theft. The model comprises three compartments: the auto-encoder, the wide convolutional neural network (1-D CNN model), and the deep convolutional neural network (2-D CNN model). The auto-encoder has been trained on the complex and in-depth linkage between the theft data and the normal data as it removes noise and unnecessary information. The 1-D CNN model gathers relevant connections and general features, while the 2-D CNN model determines the rate at which energy theft occurs and differentiates between the energy-stealing consumers and normal consumers. The efficacy of the approach is underscored by its superiority over traditional deep learning and machine learning techniques. This paper elucidates the distinct advantages and applications of the proposed model in combating electricity theft within smart grid environments. [ABSTRACT FROM AUTHOR]

Time series analysis is widely applied in action recognition, anomaly detection, and weather forecasting. Time series forecasting remains a key challenge due to the complexity of temporal patterns, overlapping changes within sequences, and the need for advanced predictive models to forecast longer sequences in many scenarios. In this study, a model named the decomposed dimension time-domain convolutional neural network (DDTCN) is proposed. This model is specifically designed to address the challenges associated with long time series data. This paper presents a dimension temporal convolutional network (DTCN) module, which has a strong ability to capture variable correlations, and an adaptive strategy is introduced. Specifically, the model proposed in this paper first decomposes time series trends and is combined with the DTCN to extract variable correlations, thus achieving accurate predictions for complex time series and providing a powerful solution for long-term series forecasting. Experiments are conducted on multiple long-term series datasets covering five practical applications: energy, transportation, economics, weather, and health care. The proposed model is extensively evaluated and compared with traditional time series prediction methods and several benchmarks. The experimental results demonstrate that the proposed model outperforms state-of-the-art methods in most tasks involving multiple long-term series forecasting. Additionally, a pig price dataset is generated to predict agricultural economic trends, where compared to that of state-of-the-art methods, the DDTCN achieves a reduction in prediction error of 25.36%. Hence, this model holds promising prospects for wide-ranging applications. [ABSTRACT FROM AUTHOR]

This paper proposes a forward layer-wise learning algorithm for CNNs in classification problems. The algorithm utilizes the Separation Index (SI) as a supervised complexity measure to evaluate and train each layer in a forward manner. The proposed method explains that gradually increasing the SI through layers reduces the input data's uncertainties and disturbances, achieving a better feature space representation. Hence, by approximating the SI with a variant of local triplet loss at each layer, a gradient-based learning algorithm is suggested to maximize it. Inspired by the NGRAD (Neural Gradient Representation by Activity Differences) hypothesis, the proposed algorithm operates in a forward manner without explicit error information from the last layer. The algorithm's performance is evaluated on image classification tasks using VGG16, VGG19, AlexNet, and LeNet architectures with CIFAR-10, CIFAR-100, Raabin-WBC, and Fashion-MNIST datasets. Additionally, the experiments are applied to text classification tasks using the DBPedia and AG's News datasets. The results demonstrate that the proposed layer-wise learning algorithm outperforms state-of-the-art methods in accuracy and time complexity. [ABSTRACT FROM AUTHOR]

The majority of the current widely used algorithms for fatigue detection rely on shallow learning to extract fatigue characteristics and use a single feature to determine the level of fatigue. The accuracy of detection is greatly affected by individual and environmental differences, and there are certain limitations in complex scenes. To improve the accuracy and real-time performance of the fatigue detection algorithm, a new driver fatigue detection algorithm based on multi-feature fusion is proposed. This paper employs two cameras to capture photos of the driver and the road, respectively, and a lightweight convolutional neural network to extract features from the driver's face, including the eyes, mouth, and head, as well as lane departure features from the road images. The four fatigue features are analyzed and fused to comprehensively detect the driver's fatigue state. The experimental results show that the multi-feature fusion-based driver fatigue detection algorithm can not only detect the driver's fatigue condition accurately but also classify the fatigue state according to the degree of fatigue, which is useful for making effective pre-warning system. [ABSTRACT FROM AUTHOR]