Search

GoogLeNet is a pre-trained Convolutional Neural Network (CNN) that allows transfer learning and has achieved high recognition rates in image classification tasks. A Recurrence Plot (RP) is an imaging method that depicts the recurrence of the state space system using coloured points and lines in 2D images. This work contributes to facilitating time series feature extraction by proposing a method that applies the GoogLeNet to time series images obtained with RP. The developed method is tested using simulated time series and selected time series from the M3 competition dataset. The results shows that the transfer learning approach allowed the extraction of business time series features by means of a GoogLeNet fine-tuned using 100 simulated time series. The combination of GoogLeNet and RPs outperforms the alternative and easier combination of GoogLeNet and plots of the time series and support the convenience of the RP transformation step. This application of deep learning techniques to business time series imaging offers opportunity for further developments. [ABSTRACT FROM AUTHOR]

Depression is a pressing yet underdiagnosed issue in health management. Because depressed patients share their symptoms, life events, and treatments on digital platforms, information systems (IS) scholars resort to user-generated digital traces for depression detection. While they facilitate innovative information technology (IT) approaches to alleviate the social and economic burden of depression, most studies lack effective means to incorporate domain knowledge in depression detection systems or suffer from feature extraction difficulties. Following the design science research in IS, we propose a Deep-Knowledge-Aware Depression Detection system to detect social media users at risk of depression and explain the detection factors. We deploy extensive empirical analyses to evaluate our designed IT artifact, which shows domain knowledge greatly improves performance. Our work has significant implications for IS research in knowledge-aware machine learning, digital traces utilization, and generalizable design principles. Practically, the early detection and factor explanation from our IT artifact can assist depression management and enable large-scale assessment of the population's mental health. [ABSTRACT FROM AUTHOR]

Surface defect detection by machine vision has received increased attention concerning the quality control of universal joint bearings (UJB). The defect distribution and counting information are important for product quality optimisation. However, vision defect detection for UJB remains a challenging task due to the diversity of background textures and defect characters on multiple surfaces. In this study, a multi-surface defect detection (MSDD) method consisting of region segmentation, feature extraction and detection is proposed. First, defect regions are accurately localised by the proposed adaptive defect region segmentation algorithm, which suppresses the inference of background variety. Then, a novel defect feature named multimodal fusion shape descriptor that integrates the global information and local information of defects is constructed to generate the discriminative defect representation. Finally, a defect feature extraction ability transfer strategy based on the transfer learning mechanism is proposed to address the problem of insufficient defect samples. The experimental results show that our method achieves the best accuracy of 94.8% macro-F1 and processes 28 defect images per second. Besides, the application effect in the practical production line indicates that our method meets the accuracy and real-time requirements of MSDD for UJB. [ABSTRACT FROM AUTHOR]

The fast warning for financial risk of enterprises has always been a realistic demand for their managers. Currently, this mainly relies on expert experience to make comprehensive analysis from massive business data. Benefitting from the strong computational performance of deep learning, this paper proposes a fuzzy neural network (FNN)-based intelligent warning method for financial risk of enterprises. An improved FNN structure with time-varying coefficients and time-varying time lags is established to extract features of enterprises from complex financial context. The algorithm of fuzzy C-means and fuzzy clustering based on sample data are studied. In this paper, the fuzzy C-means algorithm is used to cluster the samples, the input sample set is preprocessed, a new set of learning samples is formed, and then the neural network is trained. The enterprise financial risk sample and its modular FNN model are established, and the evaluation of the enterprise financial risk sample is simulated. Then, a decision part is added following the FNN part to output the warning results. After that, we have also conducted a case study as simulation experiments to evaluate the proposed technical framework. The obtained results show that it can perform well in the fast warning of financial risk for enterprises. [ABSTRACT FROM AUTHOR]

Handwritten character recognition (HCR) remains a challenging pattern recognition problem despite decades of research, and lacks research on script independent recognition techniques. This is mainly because of similar character structures, different handwriting styles, diverse scripts, handcrafted feature extraction techniques, unavailability of data and code, and the development of script-specific deep learning techniques. To address these limitations, we have proposed a script independent deep learning network for HCR research, called HCR-Net, that sets a new research direction for the field. HCR-Net is based on a novel transfer learning approach for HCR, which partly utilizes feature extraction layers of a pre-trained network. Due to transfer learning and image augmentation, HCR-Net provides faster and computationally efficient training, better performance and generalizations, and can work with small datasets. HCR-Net is extensively evaluated on 40 publicly available datasets of Bangla, Punjabi, Hindi, English, Swedish, Urdu, Farsi, Tibetan, Kannada, Malayalam, Telugu, Marathi, Nepali and Arabic languages, and established 26 new benchmark results while performed close to the best results in the rest cases. HCR-Net showed performance improvements up to 11% against the existing results and achieved a fast convergence rate showing up to 99% of final performance in the very first epoch. HCR-Net significantly outperformed the state-of-the-art transfer learning techniques and also reduced the number of trainable parameters by 34% as compared with the corresponding pre-trained network. To facilitate reproducibility and further advancements of HCR research, the complete code is publicly released at https://github.com/jmdvinodjmd/HCR-Net. [ABSTRACT FROM AUTHOR]

Renal Cell Carcinoma (RCC) is the most common malignant tumor (85%) of kidney cancer and has a complex histological pattern and nuclear structure. The manual diagnosis of kidney cancer or any other cancer from histopathology image depends on the knowledge and experience of pathologists, and the pathologist's experience influences the results. According to studies, the kind of histology in kidney cancer is related to the prognosis and course of treatment. Since the kind of histology, molecular profile, and stage of the disease all affect how the disease is treated, there is an essential need to develop an automated system that can precisely analyze the histopathological images of the disease. This work demonstrates how a deep learning framework can be used to predict and classify associated grades of RCC from provided haematoxylin and eosin (H &E) images. The proposed model focuses on two important tasks- First to capture and extract associated features from the H &E images of five different grades. Second, to classify the new set of unseen H &E images into five separate grades using the obtained features. The proposed architecture has been tested and experimented on two independent datasets containing H &E stained histopathology images. The proposed architecture has been examined using the following performance metrics namely precision, recall, F1 - score, accuracy, Floating-point operations (FLOPs), and the total number of parameters. The obtained results show that the proposed architecture attains better results over seven state-of-the-art deep learning architectures on two different H &E stained histopathology image datasets. [ABSTRACT FROM AUTHOR]

Reconstruction of multi-view 3-dimensional images is essential in robotics and computer vision to obtain an accurate 3-dimensional representation of objects by analyzing the 2-dimensional input data. For reconstructing the 3-dimensional image, it is mandatory to analyze the 3-dimensional geometry features from multiple viewpoints. It includes feature extraction and transformation from 2-dimensional features to 3-dimensional volumetric meshes. However, the existing research cannot produce consistent reconstruction results for the same input images with different orders. Therefore, the deep learning-based Residual Network-50 model is developed for 3-dimensional image reconstruction from multi-view images in the present work. The proposed system model comprises a 2-dimensional and 3-dimensional network and a backpropagation layer. From the input image, 2-dimensional features are computed using a 2-dimensional network. Then, the metaheuristic Adaptive School of Fish Optimization is used to improve the neural network's output, determining the optimal weight that gives less classification error. Then, the testing process uses the deep autoencoder, which decodes the output of the training model. Residual Network-50 is used to reconstruct 2-dimensional images to 3-dimensional using single or multi-views. Finally, the experimental analysis is performed in Python. The experiment is performed on the ShapeNet dataset and compared with the existing works. The proposed model yields better accuracy, F-score and Intersection-over-Union values of 99.3%, 0.893 and 0.734, respectively, which is more efficient than other existing models. [ABSTRACT FROM AUTHOR]

With the development of highways, there has been an increase in toll evasion by unscrupulous individuals who employ various means to avoid paying fees. This has made it essential for highway management to focus on effective methods to prevent and manage toll evasion, as well as regulate toll-related operations. However, traditional auditing techniques are heavily reliant on manual toll data and lack sufficient auxiliary data such as video and pictures, making big data analysis and in-depth data analysis difficult. Consequently, these methods can only identify a portion of evasion behavior, with a significant number of potential evasion cases remaining unidentified. To address this issue, we propose utilizing vehicle re-identification to enable cross-camera identification and continuous tracking of vehicles, allowing for accurate calculation of their operating mileage. However, vehicle re-identification using images captured by highway cameras is a challenging task, primarily due to the low resolution of the images and the lack of relevant datasets. To tackle these problems, we construct a large-scale, low-resolution dataset based on images captured by cameras on highways. We also propose a Part-guided Feature-combination Network (PFNet) to analyze the dataset. PFNet adopts a three-stage approach, wherein each stage of the network requires independent training. In the first stage, the Pyramid Scene Parsing Network (PSPNet) is improved to segment the entire vehicle into parts. In the second stage, based on a comparative learning approach, the Siamese Feature Extraction Module (SFEM) module was created to extract features of pairs of vehicle parts. To train the SFEM, pairs of positive samples, which consist of the same vehicle component pairs captured by different cameras, and pairs of negative samples, which consist of different vehicle component pairs are needed. To ensure that the positive sample pairs are closer to each other in the embedding space, and the negative samples are as far away as possible from each other, contrast loss is utilized. Moreover, The SFEM incorporates VGG16 and a Multi-period Convolutional Block Attention Module (MCBAM), which enhances the learning ability of the conventional Convolutional Block Attention Module (CBAM). In the third stage, information propagation among the vehicle parts is achieved using a Multi-layer Perceptron (MLP), which calculates the similarity after combining the features of all parts. Our self-built test set demonstrates that PFNet is effective for continuous monitoring of vehicles on highways under multiple cameras, resulting in improved accuracy of vehicle re-identification, with Top1 accuracy of 93.61% and Top5 accuracy of 99.44%. [ABSTRACT FROM AUTHOR]

Deep learning-based models have recently shown a strong potential in Underwater Image Enhancement (UIE) that are satisfying and have the right colors and details, but these methods significantly increase the parameters and complexity of the image processing models and therefore cannot be deployed directly to the edge devices. Vision Transformers (ViT) based architectures have recently produced amazing results in many vision tasks such as image classification, super-resolution, and image restoration. In this study, we introduced a lightweight Context-Aware Vision Transformer (CAViT), based on the Mean Head tokenization strategy and uses a self-attention mechanism in a single branch module that is effective at simulating long-distance dependencies and global features. To further improve the image quality we proposed an efficient variant of our model which derived results by applying White Balancing and Gamma Correction methods. We evaluated our model on two standard datasets, i.e., Large-Scale Underwater Image (LSUI) and Underwater Image Enhancement Benchmark Dataset (UIEB), which subsequently contributed towards more generalized results. Overall findings indicate that our real-time UIE model outperforms other Deep Learning based models by reducing the model complexity and improving the image quality (i.e., 0.6 dB PSNR improvement while using only 0.3% parameters and 0.4% float operations). [ABSTRACT FROM AUTHOR]

For the safety and reliability of the system, Remaining Useful Life (RUL) prediction is considered in many industries. The traditional machine learning techniques must provide more feature representation and adaptive feature extraction. Deep learning techniques like Long Short-Term Memory (LSTM) achieved an excellent performance for RUL prediction. However, the LSTM network mainly relies on the past few data, which may only capture some contextual information. This paper proposes a hybrid combination of Convolution Neural Network (CNN) and LSTM (CNN+LSTM) to solve this problem. The proposed hybrid model predicts how long a machine can operate without breaking down. In the proposed work, 1D horizontal and vertical signals of the mechanical bearing are first converted to 2D images using Continuous Wavelet Transform (CWT). These 2D images are applied to CNN for key feature extraction. Ultimately, these key features are applied to the LSTM deep neural network for predicting the RUL of a mechanical bearing. A PRONOSTIA data is utilized to demonstrate the performance of the proposed model and compare the proposed model with other state-of-the-art methods. Experimental results show that our proposed CNN+LSTM-based hybrid model achieved higher accuracy (98%) with better robustness than existing methods. [ABSTRACT FROM AUTHOR]

Convolutional neural networks (CNNs) have demonstrated promising accuracy in segmenting concrete cracks under controlled conditions. However, these existing methods usually are challenging in addressing multi-scale patterns or complex scenarios. To address this pervasive issue, a modified U-Net network was established in the current study. The proposed model, namely U-Net-ASPP-CBAM, integrates the convolutional block attention module after a convolution operator to selectively focus on the crack information while disregarding irrelevant background details during feature extraction. Furthermore, U-Net-ASPP-CBAM network replaces a pooling layer using the atrous spatial pyramid pooling module to explore and fuse features across multiple scales, enhancing information capture for the segmentation of small objects. The performance of the proposed model has been validated by a self-built dataset comprising crack images with diverse complex backgrounds. And the segmentation effectiveness is assessed through evaluation indices, including precision, recall, F1 score, pixel accuracy, and mean intersection over union. The results show the proposed U-Net-ASPP-CBAM model outperforms other segmentation models. [ABSTRACT FROM AUTHOR]

The quality of weld seams is a core factor affecting the performance of welded components, significantly impacting the safety and reliability of weldment structures. It holds considerable importance in quality control and cost management. Therefore, weld seam inspection is an indispensable part of industrial manufacturing and a key topic in non-destructive testing. This paper comprehensively summarises the main research results in weld seam feature extraction and defect identification based on systematic and detailed literature research. Firstly, the paper compares current mainstream non-destructive testing technologies, such as ultrasonic testing, radiographic testing, and infrared thermography, regarding detection principles, performance accuracy, and efficiency. It summarises their applicable scopes, as well as their advantages and disadvantages, and briefly introduces image processing-based methods for weld seam feature extraction. Secondly, a discussion and analysis of weld defect recognition are carried out, focusing on intelligent recognition methods based on deep learning, such as convolutional neural networks, generative adversarial networks, YOLO, etc. Finally, the outlook and future research trends are presented from various perspectives, such as real-time, adaptability, intelligence, etc. to provide a theoretical basis and technical support for welding quality control and structural safety assessment. [ABSTRACT FROM AUTHOR]

Intelligent monitoring of tool condition is the key to ensuring workshop manufacturing efficiency, product machining quality, and accuracy, and is also an indispensable part of intelligent processing. In the face of complex and massive, multi-source heterogeneous, and low-value density machining process data, the monitoring method based on traditional machine learning is challenging to meet the development needs of intelligent manufacturing. In contrast, with its powerful data processing and automatic feature extraction capabilities, deep learning shows broad application prospects in tool condition intelligent monitoring. Given this, this paper first systematically introduces the components of tool condition intelligent monitoring framework based on deep learning. Subsequently, the basic principles, modeling process, and application status of the four most widely used deep learning models (deep belief network, stacked auto-encoder network, convolutional neural network, and recurrent neural network) in the field of tool condition monitoring are detailed, and the advantages and disadvantages of different models are comparably discussed. Finally, the challenges and prospects of the current tool condition intelligent monitoring based on deep learning are summarized. [ABSTRACT FROM AUTHOR]

Nowadays, people all over the world are facing several problems related to the deadly disease of breast cancer. The research on breast cancer detection using existing techniques shows lesser detection results due to the shortage of medical facilities and manual detection procedures. However, the disease diagnosis becomes very critical when it is detected in the critical or chronic stage. Early detection plays an essential role in the accurate detection and effective treatment of breast cancer, which further minimizes the death rate. Therefore, automated breast cancer detection based on deep learning is proposed in this paper by using histopathological images. The proposed approach involves five steps: image filtering, dual-stage segmentation, feature extraction, feature selection, and classification. Initially, image filtering is performed to execute image resizing, noise elimination, and contrast enhancement. The weighted Guided Image Filtering (Weighted GIF) approach is used for noise removal, and Transformed Optimal Gamma Correction (TOGC) is used for contrast enhancement. To obtain the cellular structures from histology/histopathology images, a dual-stage segmentation using Superpixel Mixed Clustering (SMC) is applied. Then, feature extraction is done by the Gray Level Co-occurrence Matrix with Three-dimensional space (GLCM -3D) and RGB color model to extract texture and color features. Then, the most significant features are selected using Stochastic Diffusion Dynamic Optimization (SDDO). Finally, breast histology images have been classified using Rapid Tri-Attention Residual Dense Capsule Network with Aquila Optimization (Rapid Tri-Net), further categorizing the histology images into various classes. The proposed approach is simulated in the Python platform using BreakHis and BACH datasets and evaluates the performance on the basis of f-measure, recall, accuracy, precision, and specificity. Rapid Tri-Net's performance is related to the recent prevailing framework to attain a fair comparison. As a result, the simulated results clearly showed that the proposed Rapid Tri-Net performed better than the existing approaches. [ABSTRACT FROM AUTHOR]

Autism spectrum disorder (ASD) is a disorder in neurological growth, which includes cognitive and behavioral impairment and it starts from infancy. However, the reason for ASD is still vague and no effective medical ways are used for its discovery. Earlier discovery of ASD is extensively beneficial for children's health sustainability. The classical detection models rely on expertise analysis which tends to be expensive and inaccurate. Thus, this paper presents an effectual autism diagnostic model with electroencephalogram (EEG) signals that are produced through the electrical activities of the brain for detecting ASD. The Gaussian filter is employed to abandon the noise. Various statistical features signal and spectral-based features are extracted and provided to DRN for enhanced efficiency. The ASD detection is undergone using Chronological Sewing Training Optimization-Deep Residual Network (CSTO-DRN) wherein DRN is pre-trained using CSTO algorithm by tuning finest weights. The CSTO is built by incorporating the Chronological concept with Sewing Training-Based Optimization (STBO). The CSTO-DRN provided finest accuracy of 88.6%, Negative Predictive Value (NPV) of 87.8%, Positive Predictive Value (PPV) of 89.4%, True negative rate (TNR) of 85%, True positive rate (TPR) of 88.9%, and F-Measure of 87.5%. Its execution can enhance the efficiency of detection and minimize cost and human intervention. [ABSTRACT FROM AUTHOR]

Intelligent farming has been an ever-increasing trend in agricultural production, with the development of artificial intelligence (AI) and Internet of Things (IoT). Rapid and accurate identification of cattle identity is of great significance to prevent the insurance fraud for the live cattle loans in the cattle industry. Among them, computer vision can be expected for the cattle face recognition in the modernization transformation of the livestock industry. Smart devices and systems can also be integrated to achieve the intelligent cattle management, feeding, and disease prevention. However, the traditional identification (such as ear tags and collars) has limited the large-scale production in recent years, due to the small differences in facial features among different cattle, the deep layers of the FaceNet network, slow inference speeds, and insufficient classification accuracy. In this study, a cattle face recognition was proposed using FaceNet, called VanillaFaceNet. Firstly, the backbone feature extraction network of FaceNet was replaced with the latest simplified network. VanillaNet-13. Dynamic activation and enhanced linear transformation of activation functions were proposed to improve the non-linearity of the network. Specifically, dynamic activation was fully utilized the expressive power of activation functions during training when dynamically adjusting, in order to flexibly adapt the variations in data distribution at different stages of training. Dynamic activation was used to merge the convolutional layers during inference phase. The computational load was reduced to improve the inference speed of networks. The performance and efficiency of model were then enhanced during training and inference. Activation functions with linear transformations were significantly enhanced the non-linearity through parallel stacking. Multiple activation functions were stacked in parallel, thus enabling each layer to capture more complex features. Additionally, spatial context information was embedded within the activation functions. The spatial relationships among features were better utilized to fit the complex feature distributions. Non-linearity and integration of spatial context information were achieved in a more accurate and efficient model when processing complex data. Secondly, DBCA (Dual-Branch Coordinate Attention) module was added into the global maximum pooling. Global average pooling was used to aggregate significant features of cattle faces, in order better represent the differences among cattle facial features. Therefore, the accuracy network was improved to recognize the cattle. Finally, a center loss was introduced to train the network with the center-triplet loss joint supervision, because the triplet loss was only reduced the inter-class differences among cattle. The intra-class separability of cattle was improved to compactly aggregate the same category of cattle. Thus, the accuracy of comparisons was improved among the same identities of cattle. Cattle face videos were collected at the Otai Ranch in Hohhot, Inner Mongolia Autonomous Region. An image dataset was constructed to train and test the model for the cattle face recognition. The experimental results show that VanillaFaceNet was achieved an accuracy of 88.21% in the cattle recognition, with a frame rate of 26.23 frames per second (FPS). Compared with FaceNet, MobileFaceNet, CenterFace, CosFace, and ArcFace, the model was improved the recognition accuracy by 2.99%, 9.58%, 6.26%, 3.85%, and 4.49%, respectively, and the inference speed by 2.67, 0.77, 0.10, 1.28, and 0.94 frames/s, respectively. The recognition accuracy and speed were greatly improved to fully meet the requirements of the ranch for the accuracy and real-time performance of cattle recognition. The excellent performance was achieved in the cattle recognition, suitable for the deployment on embedded devices, such as Jetson AGX Xavier. A better balance was also gained between accuracy and inference speed of cattle facial recognition. [ABSTRACT FROM AUTHOR]

Accurate and continuous identification of individual cattle is crucial to precision farming in recent years. It is also the prerequisite to monitor the individual feed intake and feeding time of beef cattle at medium to long distances over different cameras. However, beef cattle can tend to frequently move and change their feeding position during feeding. Furthermore, the great variations in their head direction and complex environments (light, occlusion, and background) can also lead to some difficulties in the recognition, particularly for the bio-similarities among individual cattle. Among them, AlignedReID++ model is characterized by both global and local information for image matching. In particular, the Dynamically Matching Local Information (DMLI) algorithm has been introduced into the local branch to automatically align the horizontal local information. In this research, the AlignedReID++ model was utilized and improved to achieve the better performance in cattle Re-Identification (ReID). Initially, Triplet Attention (TA) modules were integrated into the BottleNecks of ResNet50 Backbone. The feature extraction was then enhanced through cross-dimensional interactions with the minimal computational overhead. Since the TA modules in AlignedReID++ baseline model increased the model size and Floating Point Operations (FLOPs) by 0.005 M and 0.05 G, the rank-1 accuracy and mean Average Precision (mAP) were improved by 1.0 percentage points and 2.94 percentage points, respectively. Specifically, the rank-1 accuracies were outperformed by 0.86 percentage points and 0.12 percentage points, respectively, compared with the Convolution Block Attention Module (CBAM) and Efficient Channel Attention (ECA) modules, although 0.94 percentage points were lower than that of Squeeze-and-Excitation (SE) modules. The mAP metric values were exceeded by 0.22 percentage points, 0.86 percentage points, and 0.12 percentage points, respectively, compared with the SE, CBAM, and ECA modules. Additionally, the Cross-Entropy Loss function was replaced with the CosFace Loss function in the global branch of baseline model. CosFace Loss and Hard Triplet Loss were jointly employed to train the baseline model for the better identification on the similar individuals. AlignedReID++ with CosFace Loss was outperformed the baseline model by 0.24 percentage points and 0.92 percentage points in the rank-1 accuracy and mAP, respectively, whereas, AlignedReID++ with ArcFace Loss was exceeded by 0.36 percentage points and 0.56 percentage points, respectively. The improved model with the TA modules and CosFace Loss was achieved in a rank-1 accuracy of 94.42%, rank-5 accuracy of 98.78%, rank-10 accuracy of 99.34%, mAP of 63.90%, FLOPs of 5.45 G, Frames Per Second (FPS) of 5.64, and model size of 23.78 M. The rank-1 accuracies were exceeded by 1.84 percentage points, 4.72 percentage points, 0.76 percentage points, and 5.36 percentage points, respectively, compared with the baseline model, Part-based Convolutional Baseline (PCB), Multiple Granularity Network (MGN), and Relation-aware Global Attention (RGA), while the mAP metrics were surpassed 6.42 percentage points, 5.86 percentage points, 4.30 percentage points, and 7.38 percentage points, respectively. Meanwhile, the rank-1 accuracy was 0.98 percentage points lower than TransReID, but the mAP metric was exceeded by 3.90 percentage points. Moreover, the FLOPs of improved model were only 0.05 G larger than that of baseline model, while smaller than those of PCB, MGN, RGA, and TransReID by 0.68 G, 6.51 G, 25.4 G, and 16.55 G, respectively. The model size of improved model was 23.78 M, which was smaller than those of the baseline model, PCB, MGN, RGA, and TransReID by 0.03 M, 2.33 M, 45.06 M, 14.53 M, and 62.85 M, respectively. The inference speed of improved model on a CPU was lower than those of PCB, MGN, and baseline model, but higher than TransReID and RGA. The t-SNE feature embedding visualization demonstrated that the global and local features were achieve in the better intra-class compactness and inter-class variability. Therefore, the improved model can be expected to effectively re-identify the beef cattle in natural environments of breeding farm, in order to monitor the individual feed intake and feeding time. [ABSTRACT FROM AUTHOR]

This research presents a novel deep learning framework for MRI-based brain tumour (BT) detection. The input brain MRI image is first acquired from the dataset. Once the images have been obtained, they are passed to an image preprocessing step where a median filter is used to eliminate noise and artefacts from the input image. The tumour-tumour region segmentation module receives the denoised image and it uses RP-Net to segment the BT region. Following that, in order to prevent overfitting, image augmentation is carried out utilizing methods including rotating, flipping, shifting, and colour augmentation. Later, the augmented image is forwarded to the feature extraction phase, wherein features like GLCM and proposed EGDP formulated by including entropy with GDP are extracted. Finally, based on the extracted features, BT detection is accomplished based on the proposed deep convolutional belief network (DCvB-Net), which is formulated using the deep convolutional neural network and deep belief network.The devised DCvB-Net for BT detection is investigated for its performance concerning true negative rate, accuracy, and true positive rate is established to have acquired values of 93%, 92.3%, and 93.1% correspondingly. [ABSTRACT FROM AUTHOR]

Feature extraction is a core aspect in hyperspectral image classification, which can extract key information closely related to ground cover from complex scene, thus improving classification accuracy. Therefore, designing an effective feature extraction network is a hotspot and a challenge in the current research. In this paper, a feature extraction framework based on 3-D homogeneous attribute decomposition (3D-HAD) is proposed for HSI classification, which consists of the following key technologies. First, the principal component analysis algorithm is applied to the raw HSI to extract the principal components (PCs), and the raw HSI is clustered into many 3D superpixel blocks according to the first three PCs-based over-segmentation strategy. Then, a superpixel intrinsic attribute decomposition (SIAD) is designed to capture reflectance feature and suppress shading feature. Meanwhile, a metric entropy is introduced into the decomposition process to overcome the spectral-spatial weak assumption among pixels. Next, superpixel-guided recursive filtering is employed to preserve global details of HSI to enhance accuracy in HSI classification. Finally, the support vector machine classifier is used to obtain classification results of HSI. Experiments performed on several real hyperspectral data sets with limited training samples indicate that the proposed 3D-HAD method outperforms the classic, advance, and deep learning classification methods. [ABSTRACT FROM AUTHOR]

Printed circuit boards (PCBs) play a critical role in electronic products. Ensuring these products' long-term reliability and consistent performance requires effective PCB defect detection. Although existing deep learning models for PCB defect detection are not highly accurate, they often neglect capability considerations. This paper introduces a precise, fast, and lightweight defect detection model, CCG-YOLO, based on an enhanced YOLOv5 model to address this issue. The enhancements in CCG-YOLO can be summarized as follows: (1) Improved Backbone network: The feature extraction ability of the Backbone network is enhanced by introducing a C3HB module, which fosters spatial interaction capabilities. (2) Lightweight feature fusion network: A lightweight convolution structure called Ghost-Shuffle Convolution is incorporated in the feature fusion network, remarkably reducing model parameters while maintaining performance. (3) Efficient residual networking: To enhance model performance further, a CNeB module is introduced based on the ConvNeXt network, which replaces the C3 module in the Neck. CNeB improves model detection accuracy and reduces the number of model parameters. The combination of these enhancements results in impressive performance. CCG-YOLO achieves mean average precision (mAP@0.5) of 99.5% and 88.75% in mAP@0.5:0.95 on the TDD-Net public dataset. Compared with the original YOLOv5s algorithm, CCG-YOLO offers a 4.24% improvement in mAP@0.5:0.95, a 1 MB reduction in model size, a 0.472 M decrease in the number of parameters, a 0.6G floating point operation reduction in computational complexity, and a 120 frames per second real-time inference speed. These experimental results underscore that the proposed model excels in accuracy and speed and has a compact size for PCB defect detection. Moreover, CCG-YOLO is easily deployable on low-end devices, making it well-suited for meeting the real-time requirements of industrial defect detection. [ABSTRACT FROM AUTHOR]

The fast and nondestructive identification of mobile phone cases is a crucial task for law enforcement personnel in the field of forensic science. Attenuated total reflection-Fourier transform infrared spectroscopy (ATR-FTIR) has been employed in conjunction with feature extraction and deep learning algorithms to develop a method for distinguishing mobile phone cases. As preprocessing steps, multivariate scatter correction and standard normal variate were performed. Feature variables were extracted using the competitive adaptive reweighted sampling and shuffled frog leaping algorithm. Several supervised pattern recognition methods, namely ResNet50, VGG16, and multilayer perceptron (MLP), were utilized for constructing classifiers. The results indicated that the model's performance based on multivariate scattering correction (MSC) with competitive adaptive re-weighted sampling (CARS) surpassed that of other approaches. Notably, the ResNet50 model exhibited superior discriminative ability compared to the VGG16 and MLP models, achieving 100% accuracy in the differentiation of all samples. This designed method represents a potentially rapid and nondestructive approach for identifying mobile phone cases in forensic science. [ABSTRACT FROM AUTHOR]

Efficient character recognition in ancient handwritten Devanagari documents is crucial for societal advancements. Challenges such as overlapping characters, missing headlines, and over-inked stains further complicate the recognition process. In response, we propose a Capsule Network (CapsNet) with LSTM to address hierarchical temporal dependencies in Devanagari scripts, following initial implementation of a simple CNN. We also explored a combined CNN+LSTM architecture for character recognition, leveraging CNN's feature extraction capabilities with LSTM's sequential modeling to handle temporal dependencies in Devanagari scripts. Our experimentation involved a dataset of 10,825 characters from ancient Devanagari manuscripts, encompassing basic characters, modifiers, and conjuncts, classified into 399 classes. Testing various training–testing ratios (9:1, 8:2, and 7:3), we visually and statistically evaluated the experimental data, demonstrating the superiority of CapsNet and LSTM in handling these challenges. We calculated recognition accuracy, precision, and recall values, with CapsNet achieving a maximum accuracy of 95.98% after 30 epochs. This research underscores the effectiveness of CapsNet and LSTM in advancing character recognition for ancient Devanagari manuscripts. [Display omitted] • Conjuncts and modifiers from Devanagari collected manuscripts are highlighted. • CapsNet model was developed to analyse recognition accuracy of given manuscript. • CapsNet achieving a maximum accuracy of 95.98% after 30 epochs. [ABSTRACT FROM AUTHOR]

Diabetic retinopathy (DR), a leading cause of blindness in diabetic patients, necessitates the precise segmentation of lesions for the effective grading of lesions. DR multi-lesion segmentation faces the main concerns as follows. On the one hand, retinal lesions vary in location, shape, and size. On the other hand, the currently available multi-lesion region segmentation models are insufficient in their extraction of minute features and are prone to overlooking microaneurysms. To solve the above problems, we propose a novel deep learning method: the Multi-Scale Spatial Attention Gate (MSAG) mechanism network. The model inputs images of varying scales in order to extract a range of semantic information. Our innovative Spatial Attention Gate merges low-level spatial details with high-level semantic content, assigning hierarchical attention weights for accurate segmentation. The incorporation of the modified spatial attention gate in the inference stage enhances precision by combining prediction scales hierarchically, thereby improving segmentation accuracy without increasing the associated training costs. We conduct the experiments on the public datasets IDRiD and DDR, and the experimental results show that the proposed method achieves better performance than other methods. [ABSTRACT FROM AUTHOR]

Inferring gene regulatory networks through deep learning and causal inference methods is a crucial task in the field of computational biology and bioinformatics. This study presents a novel approach that uses a Graph Convolutional Network (GCN) guided by causal information to infer Gene Regulatory Networks (GRN). The transfer entropy and reconstruction layer are utilized to achieve causal feature reconstruction, mitigating the information loss problem caused by multiple rounds of neighbor aggregation in GCN, resulting in a causal and integrated representation of node features. Separable features are extracted from gene expression data by the Gaussian-kernel Autoencoder to improve computational efficiency. Experimental results on the DREAM5 and the mDC dataset demonstrate that our method exhibits superior performance compared to existing algorithms, as indicated by the higher values of the AUPRC metrics. Furthermore, the incorporation of causal feature reconstruction enhances the inferred GRN, rendering them more reasonable, accurate, and reliable. [ABSTRACT FROM AUTHOR]

This work aims to explore the application of an improved convolutional neural network (CNN) combined with Internet of Things (IoT) technology in art design education and teaching. The development of IoT technology has created new opportunities for art design education, while deep learning and improved CNN models can provide more accurate and effective tools for image processing and analysis. In order to enhance the effectiveness of art design teaching and students' creative expression, this work proposes an improved CNN model. In model construction, it increases the number of convolutional layers and neurons, and incorporates the batch normalization layer and dropout layer to enhance feature extraction capabilities and reduce overfitting. Besides, this work creates an experimental environment using IoT technology, capturing art image samples and environmental data using cameras, sensors, and other devices. In the model application phase, image samples undergo preprocessing and are input into the CNN for feature extraction. Sensor data are concatenated with image feature vectors and input into the fully connected layers to comprehensively understand the artwork. Finally, this work trains the model using techniques such as cross-entropy loss functions and L2 regularization and adjusts hyperparameters to optimize model performance. The results indicate that the improved CNN model can effectively acquire art sample data and student creative expression data, providing accurate and timely feedback and guidance for art design education and teaching, with promising applications. This work offers new insights and methods for the development of art design education. [ABSTRACT FROM AUTHOR]

Introduction: Advanced traffic monitoring systems face significant challenges in vehicle detection and classification. Conventional methods often require substantial computational resources and struggle to adapt to diverse data collection methods. Methods: This research introduces an innovative technique for classifying and recognizing vehicles in aerial image sequences. The proposed model encompasses several phases, starting with image enhancement through noise reduction and Contrast Limited Adaptive Histogram Equalization (CLAHE). Following this, contourbased segmentation and Fuzzy C-means segmentation (FCM) are applied to identify foreground objects. Vehicle detection and identification are performed using EfficientDet. For feature extraction, Accelerated KAZE (AKAZE), Oriented FAST and Rotated BRIEF (ORB), and Scale Invariant Feature Transform (SIFT) are utilized. Object classification is achieved through a Convolutional Neural Network (CNN) and ResNet Residual Network. Results: The proposed method demonstrates improved performance over previous approaches. Experiments on datasets including Vehicle Aerial Imagery from a Drone (VAID) and Unmanned Aerial Vehicle Intruder Dataset (UAVID) reveal that the model achieves an accuracy of 96.6% on UAVID and 97% on VAID. Discussion: The results indicate that the proposed model significantly enhances vehicle detection and classification in aerial images, surpassing existing methods and offering notable improvements for traffic monitoring systems. [ABSTRACT FROM AUTHOR]

Pneumonia is a serious inflammatory disease that causes lung ulcers, and it is one of the leading reasons for pediatric death in the world. Chest X-rays are perhaps the most commonly utilized modalities to recognize pneumonia. Generally, the illness could be analyzed by a specialist radiologist. But for some reason, the diagnosis may be subjective. Thus, the physicians must be guided by computer-aided diagnosis frameworks in this challenging task. In this study, we propose a combined deep learning architecture to identify pneumonia in chest radiography images. We first, use Adaptive Median Filter for images enhancement, then we employ a regularized Convolutional Neural Network for features extraction, and then we use Long Short Term Memory as a classifier. Finally, the attention mechanism is used to direct the network attention to relevant features. The suggested approach was tested on two publicly available pneumonia X-ray datasets provided by Kermany and the Radiological Society of North America. On the Kermany and RSNA datasets, the suggested technique attained accuracy rates of 99.91% and 88.86%, respectively. In the last stage of our experiments, we employed a Grad-CAM-based color visualization technique to precisely interpret the detection of pneumonia in radiological images. The results outperformed those of state-of-the-art approaches. [ABSTRACT FROM AUTHOR]

In recent years, human activity recognition (HAR) has been widely used in medical rehabilitation, smart home and other fields. Currently, the recognition performance highly depends on feature extraction and effective algorithm. On the one hand, traditional manual feature extraction and classification algorithms hinder the improvement of HAR. On the other hand, the latest deep learning technology can automatically process data and extract features, but it faces the problems of poor feature quality and information loss. In order to solve this problem, this paper proposes a new recognition method using only wearable sensor data. In the feature extraction stage, the axis information of each sensor is extracted separately into one-dimensional data, and information of all axes is integrated into a two-dimensional graph. Then, two deep convolutional neural network models are designed to train the features based on one-dimensional data and two-dimensional graph respectively. Finally, weighted voting method is used to get the classification results. Experiments have shown that the average recognition accuracy of the method in this paper is about 3% higher than that of other HAR deep neural network methods, which shown the advantage of the method in this paper in obtaining better recognition result with limited data. [ABSTRACT FROM AUTHOR]

Flotation is a crucial separation technique in coal beneficiation. However, the inability to quickly and accurately detect ash content during the flotation process hinders the development of flotation automation. This paper introduced a novel hybrid model based on convolutional neural networks, establishing the ash content in tailings. In this method, convolutional neural network is utilized to automatically extract features from tailings images. In the model training and optimization phase after inputting the augmented data, the optimal ResNet18 model is identified by comparing five model optimization methods and replacing the SoftMax classifier with the traditional SVR in ResNet18. Additionally, a PCA-SVR model for predicting tailings ash content is constructed for comparative analysis, employing six different feature combinations to validate the effectiveness of the proposed features, with optimal support vector parameters identified through random search and cross-validation algorithms. The prediction results show that the PCA-SVR model has the highest accuracy when the feature dimension is reduced to 16, and the prediction of ResNet18-SVR is more accurate compared to PCA-SVR. This indicates that the proposed method can predict the ash content more accurately than the traditional feature engineering methods and has a broad application prospect in the field of coal beneficiation. [ABSTRACT FROM AUTHOR]

Human action recognition has been widely used in fields such as human–computer interaction and virtual reality. Despite significant progress, existing approaches still struggle with effectively integrating hierarchical information and processing data beyond a certain frame count. To address these challenges, we introduce the Multi-AxisFormer (MAFormer) model, which is organized in terms of spatial, temporal, and channel dimensions of the action sequence, thereby enhancing the model’s understanding of correlations and intricate structures among and within features. Drawing on the Transformer architecture, we propose the Cross-channel Spatio-temporal Aggregation (CSA) structure for more refined feature extraction and the Multi-Axis Attention (MAA) module for more comprehensive feature aggregation. Moreover, the integration of Rotary Position Embedding (RoPE) boosts the model’s extrapolation and generalization abilities. MAFormer surpasses the known state-of-the-art on multiple skeleton-based action recognition benchmarks with the accuracy of 93.2% on NTU RGB+D 60 cross-subject split, 89.9% on NTU RGB+D 120 cross-subject split, and 97.2% on N-UCLA, offering a novel paradigm for hierarchical modeling in human action recognition. [ABSTRACT FROM AUTHOR]

Objective. To enable the registration network to be trained only once, achieving fast regularization hyperparameter selection during the inference phase, and to improve registration accuracy and deformation field regularity. Approach. Hyperparameter tuning is an essential process for deep learning deformable image registration (DLDIR). Most DLDIR methods usually perform a large number of independent experiments to select the appropriate regularization hyperparameters, which are time-consuming and resource-consuming. To address this issue, we propose a novel dynamic hyperparameter block, which comprises a distributed mapping network, dynamic convolution, attention feature extraction layer, and instance normalization layer. The dynamic hyperparameter block encodes the input feature vectors and regularization hyperparameters into learnable feature variables and dynamic convolution parameters which changes the feature statistics of the high-dimensional features layer feature variables, respectively. In addition, the proposed method replaced the single-level structure residual blocks in LapIRN with a hierarchical multi-level architecture for the dynamic hyperparameter block in order to improve registration performance. Main results. On the OASIS dataset, the proposed method reduced the percentage of | J ϕ | ⩽ 0 by 28.01 % , 9.78 % and improved Dice similarity coefficient by 1.17 % , 1.17 % , compared with LapIRN and CIR, respectively. On the DIR-Lab dataset, the proposed method reduced the percentage of | J ϕ | ⩽ 0 by 10.00 % , 5.70 % and reduced target registration error by 10.84 % , 10.05 % , compared with LapIRN and CIR, respectively. Significance. The proposed method can fast achieve the corresponding registration deformation field for arbitrary hyperparameter value during the inference phase. Extensive experiments demonstrate that the proposed method reduces training time compared to DLDIR with fixed regularization hyperparameters while outperforming the state-of-the-art registration methods concerning registration accuracy and deformation smoothness on brain dataset OASIS and lung dataset DIR-Lab. [ABSTRACT FROM AUTHOR]

Vessel density within tumor tissues strongly correlates with tumor proliferation and serves as a critical marker for tumor grading. Recognition of vessel density by pathologists is subject to a strong inter-rater bias, thus limiting its prognostic value. There are many challenges in the task of object detection in pathological images, including complex image backgrounds, dense distribution of small targets, and insignificant differences between the features of the target to be detected and the image background. To address these problems and thus help physicians quantify blood vessels in pathology images, we propose Pathological Images-YOLO (PI-YOLO), an enhanced detection network based on YOLOv7. PIYOLO incorporates the BiFormer attention mechanism, enhancing global feature extraction and accelerating processing for regions with subtle differences. Additionally, it introduces the CARAFE upsampling module, which optimizes feature utilization and information retention for small targets. Furthermore, the GSConv module improves the ELAN module, reducing model parameters and enhancing inference speed while preserving detection accuracy. Experimental results show that our proposed PI-YOLO network has higher detection accuracy compared to Faster-RCNN, SSD, RetinaNet, YOLOv5 network, and the latest YOLOv7 network, with a mAP value of 87.48%, which is 2.83% higher than the original model. We also validated the performance of this network on the ICPR 2012 mitotic dataset with an F1 value of 0.8678, outperforming other methods, demonstrating the advantages of our network in the task of target detection in complex pathology images. [ABSTRACT FROM AUTHOR]

This study investigates olfaction, a complex and not well-understood sensory modality. The chemical mechanism behind smell can be described by so far proposed two theories: vibrational and docking theories. The vibrational theory has been gaining acceptance lately but needs more extensive validation. To fill this gap for the first time, we, with the help of data-driven classification, clustering, and Explainable AI techniques, systematically analyze a large dataset of vibrational spectra (VS) of 3018 molecules obtained from the atomistic simulation. The study utlizes image representations of VS using Gramian Angular Fields and Markov Transition Fields, allowing computer vision techniques to be applied for better feature extraction and improved odor classification. Furthermore, we fuse the PCA-reduced fingerprint features with image features, which show additional improvement in classification results. We use two clustering methods, agglomerative hierarchical (AHC) and k-means, on dimensionality reduced (UMAP, MDS, t-SNE, and PCA) VS and image features, which shed further insight into the connections between molecular structure, VS, and odor. Additionally, we contrast our method with an earlier work that employed traditional machine learning on fingerprint features for the same dataset, and demonstrate that even with a representative subset of 3018 molecules, our deep learning model outperforms previous results. This comprehensive and systematic analysis highlights the potential of deep learning in furthering the field of olfactory research while confirming the vibrational theory of olfaction. [ABSTRACT FROM AUTHOR]

The difficulties in communication and hearing are an important concern for deaf–dumb people, which stop access to their essential and basic needs. Many findings have been made to address sign languages even though this challenging problem is not still solved. Many methods aimed to propose vision-based classifiers through identical pattern investigation tasks by obtaining the difficult handcraft feature descriptions of gestures from the gathered images. However, the efficacy of all those models is less for performing with a huge signbook captured from uncontrolled and complex background conditions. So, an effective Indian Sign Language (ISL) classification method is developed by an advanced deep learning approach. At first, the hand gesture images are obtained from the data source. Only the image of the hand, even from a complicated background, is extracted from the obtained image. The features are extracted using the Scale-Invariant Feature Transform (SIFT) method and Multiscale Vision Transformer (MVT). Then, the extracted features are fed to the Hybrid Convolution-based EfficientNet (HCEN) model. The hyper-parameters in the developed HCEN model are tuned using the implemented Adaptive Political Optimizer (APO) algorithm. The recognized hand signs are obtained from the suggested HCEN model. Various experiments are conducted to determine the performance of the suggested deep learning-based hand gesture recognition model. [ABSTRACT FROM AUTHOR]

Gait phase recognition systems based on surface electromyographic signals (EMGs) are crucial for developing advanced myoelectric control schemes that enhance the interaction between humans and lower limb assistive devices. However, machine learning models used in this context, such as Linear Discriminant Analysis (LDA) and Support Vector Machine (SVM), typically experience performance degradation when modeling the gait cycle with more than just stance and swing phases. This study introduces a generalized phasor-based feature extraction approach (PHASOR) that captures spatial myoelectric features to improve the performance of LDA and SVM in gait phase recognition. A publicly available dataset of 40 subjects was used to evaluate PHASOR against state-of-the-art feature sets in a five-phase gait recognition problem. Additionally, fully data-driven deep learning architectures, such as Rocket and Mini-Rocket, were included for comparison. The separability index (SI) and mean semi-principal axis (MSA) analyses showed mean SI and MSA metrics of 7.7 and 0.5, respectively, indicating the proposed approach's ability to effectively decode gait phases through EMG activity. The SVM classifier demonstrated the highest accuracy of 82% using a five-fold leave-one-trial-out testing approach, outperforming Rocket and Mini-Rocket. This study confirms that in gait phase recognition based on EMG signals, novel and efficient muscle synergy information feature extraction schemes, such as PHASOR, can compete with deep learning approaches that require greater processing time for feature extraction and classification. [ABSTRACT FROM AUTHOR]

With the development of educational technology, machine learning and deep learning provide technical support for traditional classroom observation assessment. However, in real classroom scenarios, the technique faces challenges such as lack of clarity of raw images, complexity of datasets, multi-target detection errors, and complexity of character interactions. Based on the above problems, a student classroom behavior recognition network incorporating super-resolution and target detection is proposed. To cope with the problem of unclear original images in the classroom scenario, SRGAN (Super Resolution Generative Adversarial Network for Images) is used to improve the image resolution and thus the recognition accuracy. To address the dataset complexity and multi-targeting problems, feature extraction is optimized, and multi-scale feature recognition is enhanced by introducing AKConv and LASK attention mechanisms into the Backbone module of the YOLOv8s algorithm. To improve the character interaction complexity problem, the CBAM attention mechanism is integrated to enhance the recognition of important feature channels and spatial regions. Experiments show that it can detect six behaviors of students—raising their hands, reading, writing, playing on their cell phones, looking down, and leaning on the table—in high-definition images. And the accuracy and robustness of this network is verified. Compared with small-object detection algorithms such as Faster R-CNN, YOLOv5, and YOLOv8s, this network demonstrates good detection performance on low-resolution small objects, complex datasets with numerous targets, occlusion, and overlapping students. [ABSTRACT FROM AUTHOR]

Speech emotion recognition (SER) is not only a ubiquitous aspect of everyday communication, but also a central focus in the field of human–computer interaction. However, SER faces several challenges, including difficulties in detecting subtle emotional nuances and the complicated task of recognizing speech emotions in noisy environments. To effectively address these challenges, we introduce a Transformer-based model called MelTrans, which is designed to distill critical clues from speech data by learning core features and long-range dependencies. At the heart of our approach is a dual-stream framework. Using the Transformer architecture as its foundation, MelTrans deciphers broad dependencies within speech mel-spectrograms, facilitating a nuanced understanding of emotional cues embedded in speech signals. Comprehensive experimental evaluations on the EmoDB (92.52%) and IEMOCAP (76.54%) datasets demonstrate the effectiveness of MelTrans. These results highlight MelTrans's ability to capture critical cues and long-range dependencies in speech data, setting a new benchmark within the context of these specific datasets. These results highlight the effectiveness of the proposed model in addressing the complex challenges posed by SER tasks. [ABSTRACT FROM AUTHOR]

With the wide application of wind energy as a clean energy source, to cope with the challenge of increasing maintenance difficulty brought about by the development of large-scale wind power equipment, it is crucial to monitor the operating status of wind turbines in real time and accurately identify the specific location of faults. In this study, a CNN-LSTM-based wind motor fault detection model is constructed for four types of typical faults, namely gearbox faults, electrical faults, yaw faults, and pitch faults of wind motors, combining CNN's advantages of excelling in feature extraction and LSTM's advantages of dealing with long-time sequence data, to achieve the simultaneous detection of multiple fault types. The accuracy of the CNN-LSTM-based wind turbine fault detection model reaches 90.06%, and optimal results are achieved for the effective discovery of yaw system faults, pitch system faults, and gearbox faults, obtaining 94.09%, 96.46%, and 97.39%, respectively. The CNN-LSTM wind turbine fault detection model proposed in this study improves the fault detection effect, avoids the further deterioration of faults, provides direction for preventive maintenance, reduces downtime loss due to restorative maintenance, and is essential for the sustainable use of wind turbines and maintenance of wind turbine service life, which helps to improve the operation and maintenance level of wind farms. [ABSTRACT FROM AUTHOR]

With the development of the power system, power users begin to use their own power supply in order to improve the power economy, but this also leads to the occurrence of the risk of self-provided power supply. The actual distribution network has few samples of power supply risk and it is difficult to identify the power supply risk by using conventional deep learning methods. In order to achieve high accuracy of self-provided power supply risk identification with small samples, this paper proposes a combination of transfer learning, convolutional block attention module (CBAM), and convolutional neural network (CNN) to identify the risk of self-provided power supply in an active distribution network. Firstly, in order to be able to further identify whether or not a risk will be caused based on completing the identification of the faulty line, we propose that it is necessary to identify whether or not the captive power supply on the faulty line is in operation. Second, in order to achieve high-precision identification and high-efficiency feature extraction, we propose to embed the CBAM into a CNN to form a CBAM-CNN model, so as to achieve high-efficiency feature extraction and high-precision risk identification. Finally, the use of transfer learning is proposed to solve the problem of low risk identification accuracy due to the small number of actual fault samples. Simulation experiments show that compared with other methods, the proposed method has the highest recognition accuracy and the best effect, and the risk recognition accuracy of active distribution network backup power is high in the case of fewer samples. [ABSTRACT FROM AUTHOR]

Infrared and visible image fusion integrates complementary information from different modalities into a single image, providing sufficient imaging information for scene interpretation and downstream target recognition tasks. However, existing fusion methods often focus only on highlighting salient targets or preserving scene details, failing to effectively combine entire features from different modalities during the fusion process, resulting in underutilized features and poor overall fusion effects. To address these challenges, a global and local four-branch feature extraction image fusion network (GLFuse) is proposed. On one hand, the Super Token Transformer (STT) block, which is capable of rapidly sampling and predicting super tokens, is utilized to capture global features in the scene. On the other hand, a Detail Extraction Block (DEB) is developed to extract local features in the scene. Additionally, two feature fusion modules, namely the Attention-based Feature Selection Fusion Module (ASFM) and the Dual Attention Fusion Module (DAFM), are designed to facilitate selective fusion of features from different modalities. Of more importance, the various perceptual information of feature maps learned from different modality images at the different layers of a network is investigated to design a perceptual loss function to better restore scene detail information and highlight salient targets by treating the perceptual information separately. Extensive experiments confirm that GLFuse exhibits excellent performance in both subjective and objective evaluations. It deserves note that GLFuse effectively improves downstream target detection performance on a unified benchmark. [ABSTRACT FROM AUTHOR]

Detecting clouds, snow, and lakes in remote sensing images is vital due to their propensity to obscure underlying surface information and hinder data extraction. In this study, we utilize Sentinel-2 images to implement a two-stage random forest (RF) algorithm for image labeling and delve into the factors influencing neural network performance across six aspects: model architecture, encoder, learning rate adjustment strategy, loss function, input image size, and different band combinations. Our findings indicate the Feature Pyramid Network (FPN) achieved the highest MIoU of 87.14%. The multi-head self-attention mechanism was less effective compared to convolutional methods for feature extraction with small datasets. Incorporating residual connections into convolutional blocks notably enhanced performance. Additionally, employing false-color images (bands 12-3-2) yielded a 4.86% improvement in MIoU compared to true-color images (bands 4-3-2). Notably, variations in model architecture, encoder structure, and input band combination had a substantial impact on performance, with parameter variations resulting in MIoU differences exceeding 5%. These results provide a reference for high-precision segmentation of clouds, snow, and lakes and offer valuable insights for applying deep learning techniques to the high-precision extraction of information from remote sensing images, thereby advancing research in deep neural networks for semantic segmentation. [ABSTRACT FROM AUTHOR]

Hyperspectral unmixing (HU) aims to decompose mixed pixels into a set of endmembers and corresponding abundances. Deep learning-based HU methods are currently a hot research topic, but most existing unmixing methods still rely on per-pixel training or employ convolutional neural networks (CNNs), which overlook the non-local correlations of materials and spectral characteristics. Furthermore, current research mainly focuses on linear mixing models, which limits the feature extraction capability of deep encoders and further improvement in unmixing accuracy. In this paper, we propose a nonlinear unmixing network capable of extracting global spatial-spectral features. The network is designed based on an autoencoder architecture, where a dual-stream CNNs is employed in the encoder to separately extract spectral and local spatial information. The extracted features are then fused together to form a more complete representation of the input data. Subsequently, a linear projection-based multi-head self-attention mechanism is applied to capture global contextual information, allowing for comprehensive spatial information extraction while maintaining lightweight computation. To achieve better reconstruction performance, a model-free nonlinear mixing approach is adopted to enhance the model's universality, with the mixing model learned entirely from the data. Additionally, an initialization method based on endmember bundles is utilized to reduce interference from outliers and noise. Comparative results on real datasets against several state-of-the-art unmixing methods demonstrate the superior of the proposed approach. [ABSTRACT FROM AUTHOR]

EEG-based emotion classification is considered to separate and observe the mental state or emotions. Emotion classification using EEG is used for medical, security and other purposes. Several deep learning and machine learning strategies are employed to classify the EEG emotion signals. They do not provide sufficient accuracy and have higher complexity and high error rate. In this manuscript, a novel Reinforced Spatio-Temporal Attentive Graph Neural Networks (RSTAGNN) and ContextNet for emotion classification with EEG signals is proposed (RSTAGNN-ContextNet-GWOA-EEG-EA). Here, the input EEG signals are taken from two benchmark datasets,namely DEAP and K-EmoCon datasets. Then, the input EEG signals are pre-processed,and the features are extracted utilizing ContextNet with Global Principal Component Analysis (GPCA). After that, the EEG signal emotions are classified using Reinforced Spatio-Temporal Attentive Graph Neural Networks method. RSTAGNN weight parameters are optimized under the Glowworm Swarm Optimization Algorithm (GWOA). The proposed model classifies the EEG signal emotions with high accuracy. The efficacy of the proposed method using the DEAP dataset attains higher accuracy by 24.05%, 12.64% related to existing systems, like Multi-domain feature fusion for emotion classification (DWT-SVM-EEGEA- DEAP), EEG emotion finding utilizing fusion mode of graph CNN with LSTM (GCNN-LSTM-EEG-EA-DEAP) respectively. The efficiency of the proposed method using the K-EmoCon dataset attains higher accuracy 32.64%, 15.65% related to existing systems, like Toward Robust Wearable Emotion Realization along Contrastive Representation Learning (CAT-EEG-EA-K-EmoCon) and Human Emotion Recognition using Physiological Signals (CATEEG- EA-K-EmoCon) respectively. [ABSTRACT FROM AUTHOR]

Accurate tool wear state identification models are essential to ensure manufacturing reliability and efficiency. Tool wear state recognition systems establish a mapping relationship with the tool state by extracting signal features. Therefore, this paper proposes an architecture for identifying the actual wear state of data unbalanced machining tools by applying the power of minimalism in deep learning networks, namely, VanillaNet, combined with recurrence plot encoding technology (RP). In this paper, the signal is preprocessed by RP, and the nonlinear one-dimensional time-sequential digital signal embedded in variable time-lag delay coordinate space in the phase space is converted into a two-dimensional (2D) color texture image, thereby achieving the feature extraction of tool wear. Then, the data-enhanced 2D recurrence coded image is used as the input to VanillaNet, and its minimalist network architecture is applied to establish the mapping relationship between tool wear states and wear features. This process reduces the state recognition time and achieves the fast recognition of tool wear states. The model in this paper achieves more than 95 % on all four classification metrics: accuracy, recall, F1 score, and precision in three sets of crossover experiments while reducing misclassification in the sharp wear phase. The proposed model also outperforms three DL-based methods, namely, CNN-Attention, AlexNet, and ResNet. [ABSTRACT FROM AUTHOR]

The physiological activities within cells are mainly regulated through protein–protein interactions (PPI). Therefore, studying protein interactions has become an essential part of researching protein function and mechanisms. Traditional biological experiments required for PPI prediction are expensive and time consuming. For this reason, many methods based on predicting PPI from protein sequences have been proposed in recent years. However, existing computational methods usually require the combination of evolutionary feature information of proteins to predict PPI docking situations. Because different relevant features of selected proteins are chosen, there may be differences in the predicted results for PPI. This article proposes a PPI prediction method based on the pretrained protein sequence model ProtBert, combined with the Bidirectional Gated Recurrent Unit (BiGRU) and attention mechanism. Only using protein sequence information and leveraging ProtBert's powerful ability to capture amino acid feature information, BiGRU is used for further feature extraction of the amino acid vectors output by ProtBert. The attention mechanism is then applied to enhance the focus on different amino acid features and improve the expression ability of protein sequence features, ultimately obtaining binary classification results for protein interactions. Experimental results show that our proposed ProtBert-BiGRU-Attention model has good predictive performance for PPI. Through relevant comparative experiments, it has been proven that our model performs well in protein binary prediction. Furthermore, through the ablation experiment of the model, different deep learning modules' contributions to the prediction have been demonstrated. [ABSTRACT FROM AUTHOR]

This paper proposes a novel efficient descriptor for finger knuckle print (FKP) identification. Our proposed descriptor is called MDL (Magnitude, Direction, and Local patterns), as it describes the rich structure of FKP images by jointly encoding three types of discriminative information, namely gradient magnitude, direction, and local patterns. This joint representation allows obtaining reliable features as gradient values that are merged with local features could offer abundant information on the FKP structure and the edges contained therein. Note that MDL is strengthened by extracting the features in a block-wise manner to include the spatial relationship information. Furthermore, we consider a multimodal learning scheme in which our MDL and deep features are extracted from major and minor FKP modalities. For the sake of efficiency, we opt for the VGG-F deep network, which has a small depth, to extract the deep features. The VGG-F is well-suited to the nature of FKP images (i.e., lines and edges without complicated backgrounds), where networks with a few layers/parameters can efficiently learn the intrinsic characteristics of the FKP without suffering from the overfitting. The discriminant correlation analysis (DCA) algorithm is used to perform feature fusion. DCA has the advantage of efficiently and jointly performing features fusion and dimensionality reduction. We conduct comprehensive experiments on a public finger knuckle dataset. Experimental results demonstrate the effectiveness of the proposed method, which has outperformed several related methods. The obtained results confirm the complementarity of our MDL and deep features. Moreover, the proposed method has shown a high robustness to occlusion scenario, where some parts of the FKP image are missed. [ABSTRACT FROM AUTHOR]

Brain tumor classification is a significant issue in computer aided diagnosis to make a convenient treatment. Deep learning techniques are surpassing traditional brain tumor classification methods in the medical image field with advanced hardware technology. In this article, a novel classification framework has been proposed that includes pre-trained fine-tuned EfficientNet B7. The framework focuses on a multi-class classification problem to distinguish among three common types of brain tumors, i.e., gliomas, meningiomas, and pituitary tumors. The proposed classification framework is trained and tested using six optimization algorithms SGD, RMSProp, Adagrad, Adam, Adadelta, and Nadam using two classifiers softmax and support vector machine (SVM). The best performance achieved by Nesterov-accelerated adaptive moment estimation (Nadam) optimizer using softmax classifier with test accuracy 99 % , average precision 99.3 % , average recall 97.3 % and average F1-score 98 % which outperformed other state-of-the-art architectures on Figshare MRI image data set. The proposed framework has also been evaluated, and performance has been compared using two more publicly available standard four-class brain tumor MRI data sets. [ABSTRACT FROM AUTHOR]

The ability of a person to communicate with other people and engage with the outside environment makes it crucial to a person's existence. It can be challenging for all the people who have hearing loss to communicate without the aid of lip reading, facial expression, or signals. Also, without instruction, it is challenging for them to comprehend lip movements, which makes it challenging for them to hear spoken words. The visual speech recognition devices for lip reading were developed to solve this issue and help the deaf understand speech. Recently, Visual Voice Recognition research has focused a lot on deep learning advancements. Here, we suggested the LSTM and CNN 3D hybrid model for MIRACL VC1 Dataset visual speech recognition. When a person's lip features are extracted for pattern detection, the text without audio is converted in this process. A model used to automatically recognize speech is called visual speech recognition. The proposed work stands apart from the previous work by incorporation the hybrid model, the combination of the models would yield a better and efficient accuracy rate comparatively and also the focal novelty of the proposed work was that the concatenation of the extracted features to provide an input to the system was a challenging one so has to manage the weights in the entire process of extraction for a huge number of dataset. It was suggested that two separate models, CNN 3D and LSTM, be combined to create hybrid model in mandate to progress the overall accuracy in this case. In comparison with the individual CNN 3D model, whose testing accuracy is 79%, and the LSTM model, whose accuracy was 85% when we employed the English Dataset, the hybrid model's training, test, and validation accuracy were all 98, 85, and 86%, respectively. [ABSTRACT FROM AUTHOR]

Hand pose estimation is a fundamental task in computer vision with applications in virtual reality, gesture recognition, autonomous driving, and virtual surgery. Keypoint detection often relies on deep learning methods and high-resolution feature map representations to achieve accurate detection. The HRNet framework serves as the basis, but it presents challenges in terms of extensive parameter count and demanding computational complexity due to high-resolution representations. To mitigate these challenges, we propose a lightweight keypoint detection network called LGCANet (Lightweight Ghost-Coordinate Attention Network). This network primarily consists of a lightweight feature extraction head for initial feature extraction and multiple lightweight foundational network modules called GCAblocks. GCAblocks introduce linear transformations to generate redundant feature maps while concurrently considering inter-channel relationships and long-range positional information using a coordinate attention mechanism. Validation on the RHD dataset and the COCO-WholeBody-Hand dataset shows that LGCANet reduces the number of parameters by 65.9% and GFLOPs by 72.6% while preserving the accuracy and improves the detection speed. [ABSTRACT FROM AUTHOR]