Search

The high surface reflectivity and concave-convex of hot stamping process printed matters lead to poor defect detection accuracy. To solve this problem, this paper proposes a defect detection model (OBL-BWO-SVM) for hot stamping process printed matters based on oppositional learning (OBL) beluga optimized (BWO) support vector machine (SVM). First, in terms of image acquisition, a new light source illumination method is proposed for the own characteristics of hot stamping, the gray level covariance matrix (GLCM) is applied to extract the image information to build a dataset, and principal component analysis (PCA) is applied to reduce the dimensionality of the dataset. Then, SVM is used as the base model and the category weights of SVM are adjusted for the sample data category imbalance problem. Then, the C and gamma parameters of SVM are optimized using OBL-BWO. Finally, 14400 images are used as the dataset, in which the proportions of defect-free samples and defective samples are 48.9% and 51.1%, respectively, and are divided into the training set and the test set according to the ratio of 7:3. The models in this paper and the commonly used models are tested on this dataset. The experimental results show that the accuracy of the model proposed in this paper is 96.11%, precision is 94.61%, recall is 95.75%, and F1 score is 95.18%. Compared with the commonly used methods, the method proposed in this paper has higher classification accuracy and can improve the accuracy of printed matter defect detection during hot stamping printing. [ABSTRACT FROM AUTHOR]

The registration of printed images and the original image before printing is the key step of high precision quality detection of printed image. In the printing process and the process of collecting printed images, rigid deformation, elastic deformation and brightness difference between the original image and the printed image are inevitably caused to the printed image. In order to improve the registration accuracy of printed images and deal with various deformations and color differences that may occur during the printing process, this paper proposes an innovative algorithm that combines feature registration and deformation optimization. The algorithm first uses the SIFT feature extraction algorithm to achieve global registration, and then combines the Active Demons deformation estimation algorithm to optimize the local deformation to correct the rigid and elastic deformation of the printed image. However, the deformation estimation algorithm is limited by the premise of the same brightness. Therefore, this paper performs superpixel segmentation on the original image before printing, and then introduces a nonlinear brightness compensation mechanism, thus overcoming the brightness constraint of the deformation estimation algorithm. In order to improve the registration efficiency, a reduction factor is adaptively introduced in the global registration stage, and the local deformation optimization registration stage cancels the iteration by designing the value of the normalization factor. The experimental results show that compared with SIFT, Active Demons and SIFT-ActiveDemons algorithms, the innovative fusion registration algorithm in this paper has the highest registration accuracy and can effectively correct the elastic deformation of printed images. [ABSTRACT FROM AUTHOR]

Ship object detection and recognition in remote sensing images (RSIs) is a challenging task due to the multi-scale and complex background characteristics of ship objects. Currently, convolution-based methods cannot adequately solve these problems. Firstly, this paper first applies the diffusion model to the task of ship object detection and recognition in RSIs, and proposes a new diffusion model for multi-scale ship object detection and recognition in remote sensing images (MSDiffDet). Secondly, in order to reduce the loss of multi-scale information in the feature extraction process, this paper proposes the Channel Fusion FPN (CF-FPN) based on FPN and constructs the Large-Scale Feature Enhancement Module (LSFEM), which further enhances the algorithm’s ability to extract large-scale ship object features and improves the detection accuracy of ship objects in RSIs. Finally, this paper prunes and reconstructs MobileNetV2 to obtain the Sparse MobileNetV2, which is used as the backbone network of the image encoder, which enhances detection accuracy while reducing the overall parameter count of the algorithm. The experimental results demonstrate that the MSDiffDet algorithm is effective in detecting and recognizing four types of remote sensing ship objects: aircraft carriers, warships, commercial ships, and submarines. The m A P 0.5 achieved a notable 89.8%. A significant improvement of 5.8% in m A P 0.5 is observed compared to the DiffusionDet algorithm, indicating the potential of the MSDiffDet algorithm for applications in remote sensing ship object detection and recognition. [ABSTRACT FROM AUTHOR]

Accurately segmenting organs, or diseased regions of varying sizes, is a challenge in medical image segmentation tasks with limited datasets. Although Transformer-based methods have self-attention mechanisms, excellent global modelling abilities and can effectively focus on crucial areas in medical images, they still face the intractable issues of computational complexity and excessive reliance on data. The Swin-Transformer has partly addressed the problem of computational complexity, however the method still requires large amounts of training data due to its lack of inductive bias capabilities. When applied to medical image segmentation tasks with small datasets, this leads to suboptimal performances. In contrast, the CNN-based methods can compensate for this limitation. To address this issue further, this paper proposes Swin-IBNet, which combines the Swin-Transformer with a CNN in a novel manner to imbue it with inductive bias capabilities, reducing its reliance on data. During the encoding process of Swin-IBNet, two novel and crucial modules, the feature fusion block (FFB) and the multiscale feature aggregation block (MSFA), are designed. The FFB is responsible for propagating the inductive bias capability to the Swin-Transformer encoder. Different from the previous use of multiscale features, MSFA efficiently leverages multiscale information from different layers through self-learning. This paper not only attempts to analyse the interpretability of the proposed Swin-IBNet but also performs more verifications on the public Synapse, ISIC 2018 and ACDC datasets. The experimental results show that Swin-IBNet is superior to the baseline method, Swin-Unet, and several state-of-the-art methods. Especially on the Synapse dataset, the DSC of Swin-IBNet surpasses that of Swin-Unet by 3.45%. [ABSTRACT FROM AUTHOR]

The integration of radar technology into smart furniture represents a practical approach to health monitoring, circumventing the concerns regarding user convenience and privacy often encountered by conventional smart home systems. Radar technology's inherent non-contact methodology, privacy-preserving features, adaptability to diverse environmental conditions, and high precision characteristics collectively establish it a compelling alternative for comprehensive health monitoring within domestic environments. In this paper, we introduce a millimeter (mm)-wave radar system positioned strategically behind a seat, featuring an algorithm capable of identifying unique cardiac waveform patterns for healthy subjects. These patterns are characterized by two peaks followed by a valley in each cycle, which can be correlated to Electrocardiogram (ECG), enabling effective cardiac waveform monitoring. The provided algorithm excels in discerning variations in heart patterns, particularly in individuals with prolonged corrected QT intervals, by minimizing high frequency breathing interference and ensuring accurate pattern recognition. Additionally, this paper addresses the influence of body movements in seated individuals, conducting a comprehensive study on heart rate variability and estimation. Experiment results demonstrate a maximum interbeat intervals (IBI) error of 30 milliseconds and an average relative error of 4.8% in heart rate estimation, showcasing the efficacy of the proposed method utilizing variational mode decomposition and a multi-bin approach. [ABSTRACT FROM AUTHOR]

In the farming industry, to cope with problems such as complex pasture environments and dense targets, which lead to increased difficulty in recognising and thus quickly classifying and automatically identifying cattle breeds to improve accuracy. In this paper, an RTDETR-Refa (RepConv Efficient Faster Attention) algorithm based on ResNet18 backbone network is proposed for cattle breed classification and identification. First, new improvements are made to the ResNet18 backbone network: the Faster-Block module is introduced to improve the feature extraction network and increase the computational speed without sacrificing the accuracy; the 1 × 1 convolution in the Faster-Block module is replaced by a 3 × 3 convolution using the RepConv reparameterised with the RepVGG block, which makes the algorithm more lightweight and improves the inference speed. Second, in order to enhance the feature transformation and classification, the Efficient Multiscale Attention (EMA) module is added after the Faster-Block module at different stages. Finally, the above improved Faster-Block module is used to replace the 4-layer BasicBlock after the 3 convolutional layers in the backbone network of ResNet18. The training test results of the RTDETR-Refa algorithm are compared with other classical models to validate the superiority of the RTDETR-Refa algorithm. The average accuracy of the RTDETR-Refa algorithm on the bovine The average accuracy of RTDETR-Refa algorithm on the classification training set is 91.6%, which is 0.8% higher than the original model and 0.9–5.2% higher than other classical models. The experimental results show that the RTDETR-Refa model proposed in this paper is able to identify and classify different breeds of cattle while guaranteeing similar detection speed, which proves the feasibility of convolutional neural networks in breed identification and classification. [ABSTRACT FROM AUTHOR]

Springback is one of the major shape defects in roll forming. It is difficult to predict springback accurately and efficiently because the process involves complicated deformation. In this paper, a high accuracy Support Vector Regression (SVR) algorithm based on the Simulated Annealing Particle Swarm Optimization algorithm (SAPSO) is proposed to predict springback. Firstly, simulations of the forming process of V-channel profile are carried out to investigate the springback at different fillet radii, yield strengths, uphill volumes and roll span spaces. Then with the data obtained, the accuracy of SAPSO-SVR, SVR, and Back Propagation Neural Network (BPNN) prediction models was tested. The experimental results show that SAPSO-SVR has the highest prediction accuracy, and its average absolute error is about 0.11 ∘ , which is 38.7% less than BPNN, and 61.8% less than SVR. Furthermore, in order to explore the applicability of emerging Artificial Intelligence (AI) models in investigating forming mechanisms, the relationship between forming parameters and springback was elucidated based on the prediction model. It is found that the roll span space had a small impact on springback, but a larger roll span space would reduce the impact of the uphill volume on springback. And applying an uphill volume of - 10 mm will minimize the variation of longitudinal strain in the cross section, thereby reducing the trend of flange inward shrinkage and leading to an increase in springback. The above conclusions are mutually verified with traditional theoretical research. This paper establishes a high accuracy prediction model and explores the springback mechanism, which can provide important theoretical references for future research on intelligent roll forming. [ABSTRACT FROM AUTHOR]

With the rapid development and deep application of artificial intelligence, modern air combat is incrementally evolving towards intelligent combat. Although deep reinforcement learning algorithms have contributed to dramatic advances in in air combat, they still face challenges such as poor interpretability and weak transferability of adversarial strategies. In this regard, this paper proposes a tactical intent-driven method for autonomous air combat behaviour generation. Firstly, this paper explores the mapping relationship between optimal strategies and rewards, demonstrating the detrimental effects of the combination of sparse rewards and dense rewards on policy. Built around this, the decision-making process of pilot behavior is analyzed, and a reward mapping model from intent to behavior is established. Finally, to address the problems of poor stability and slow convergence speed of deep reinforcement learning algorithms in large-scale state-action spaces, the dueling-noisy-multi-step DQN algorithm is devised, which not only improves the accuracy of value function approximation but also enhances the efficiency of space exploration and network generalization. Through experiments, the conflicts between sparse rewards and dense rewards are demonstrated. The superior performance and stability of the proposed algorithm compared to other algorithms are captured by our empirical results. More intuitively, the strategies under different intents exhibit strong interpretability and flexibility, which can provide tactical support for intelligent decision-making in air combat. [ABSTRACT FROM AUTHOR]

Aiming at the lightweight and real-time detection requirements of infrared pedestrians on edge devices, this paper proposes an improved infrared pedestrian recognition model based on You Only Look Once v5 nano. The model effectively reduces the number of parameters by introducing the Ghost Convolution and SlimNeck modules in the lightweight design. To enhance real-time performance, the model reduces the number of parameters and improves operation speed through the Performance-aware Approximation of Global Channel Pruning pruning strategy. After experimental analysis, a balance between operation speed and accuracy is achieved. The results show that compared to the original model, the proposed model decreases the number of parameters by 1.52 M, increases running speed by 21%, and decreases mAP@0.50 by only 0.4%. The running speed on Jetson Nano reaches 0.12 s per image. The proposed model effectively ensures high detection accuracy under lightweight and real-time requirements, providing technical support for the deployment of edge devices in scenarios such as autonomous driving. [ABSTRACT FROM AUTHOR]

The existing coal mine safety helmet detection method has problems such as low detection accuracy, susceptibility to environmental impact, poor real-time performance, and a large number of parameters. So, this paper proposes a Miner Helmet detection algorithm based on YOLO, abbreviated as MH-YOLO. First, the convolutional block attention mechanism (CBAM) is applied to improve the CSPDarkNet53 to 2-Stage FPN (C2f) module of the backbone network and enhance feature-extraction capability. Second, the MaxPooling (MP) module is used to replace the partial subsampling convolution of YOLOv8 to reduce the impact of unbalanced sample categories and improve the recall rate. In addition, a small target detection layer is added to further improve the small target characteristics by fusing shallow network features with deep network features. Finally, the ZoomCat and Scalseq Module (ZAS) feature-extraction module is used to improve the detection accuracy of small and overlapping targets. Training and testing were conducted on the public dataset CUMT-Helmet from China University of Mining and Technology and DsLMF + helmet from Xi’an University of Science and Technology. The proposed MH-YOLO achieves mAP50 values of 92.4% and 97.8%, respectively, surpassing the comparative networks. The detection time is 10.1 ms, enabling accurate and real-time detection of whether coal miners are wearing safety helmets. Source code is released in . [ABSTRACT FROM AUTHOR]

An electrocardiogram (ECG) is a crucial noninvasive medical diagnostic method that enables real-time monitoring of the electrical activity of the heart. ECGs hold a significant position in the rapid diagnosis and routine monitoring of cardiac diseases due to their user-friendly operation, prompt detection, broad range of diagnosable problems, and cost-effectiveness. However, thorough comprehension of ECG readings requires a high level of medical expertise due to the complex variations in ECG patterns, substantial interindividual differences, and numerous interfering factors. Consequently, current ECG machines and ECG Holters typically provide simplistic indications of ECG anomalies. Nonetheless, current ECG anomaly detection (EAD) algorithms lack precision; therefore, these medical devices cannot accurately report the specific types of diseases reflected in ECG results. In response to these challenges, this paper proposes enhancing the accuracy of electrocardiogram detection by improving algorithms. Therefore, we propose SAPTSTA-AnoECG, a PatchTST-based ECG anomaly detection method with subtractive attention and data augmentation. This method introduces a subtractive attention mechanism to make the Transformer architecture more suitable for time series data. We also use data augmentation to increase the robustness of the model. In addition, a patch-based approach is employed to reduce the algorithm’s computational complexity of the model. Furthermore, we introduce a new publicly available ECG dataset named HCE in this paper and conduct comparative experiments using this dataset along with the PTB-XL and CPSC 2018 datasets. The experimental results demonstrate the effectiveness of this method. [ABSTRACT FROM AUTHOR]

Recently, brain-inspired spiking neural networks (SNNs) have attracted great research attention owing to their inherent bio-interpretability, event-triggered properties and powerful perception of spatiotemporal information, which is beneficial to handling event-based neuromorphic datasets. In contrast to conventional static image datasets, event-based neuromorphic datasets present heightened complexity in feature extraction due to their distinctive time series and sparsity characteristics, which influences their classification accuracy. To overcome this challenge, a novel approach termed Neuromorphic Momentum Contrast Learning (NeuroMoCo) for SNNs is introduced in this paper by extending the benefits of self-supervised pre-training to SNNs to effectively stimulate their potential. This is the first time that self-supervised learning (SSL) based on momentum contrastive learning is realized in SNNs. In addition, we devise a novel loss function named MixInfoNCE tailored to their temporal characteristics to further increase the classification accuracy of neuromorphic datasets, which is verified through rigorous ablation experiments. Finally, experiments on DVS-CIFAR10, DVS128Gesture and N-Caltech101 have shown that NeuroMoCo of this paper establishes new state-of-the-art (SOTA) benchmarks: 83.6% (Spikformer-2-256), 98.62% (Spikformer-2-256), and 84.4% (SEW-ResNet-18), respectively. [ABSTRACT FROM AUTHOR]

This paper introduces a novel approach for training generative adversarial networks using federated machine learning. Generative adversarial networks have gained plenty of attention in the research community especially with their abilities to produce high quality synthetic data for a variety of use-cases. Yet, when combined with federated learning, those models suffer from degradation in both training time and quality of results. To address this challenge, this paper introduces a novel approach that uses hierarchical learning techniques to enable the efficient training of federated GAN models. The proposed approach introduces an innovative mechanism that dynamically clusters participant clients to edge servers as well as a novel multi-generator GAN architecture that utilizes non-identical model aggregation stages. The proposed approach has been evaluated on a number of benchmark datasets to measure its performance on higher numbers of participating clients. The results show that HFL-GAN outperforms other comparative state-of-the-art approaches in the training of GAN models in complex non-IID federated learning settings. [ABSTRACT FROM AUTHOR]

To facilitate flexible manufacturing, modern industries have incorporated numerous modular operations such as multi-robot services which can be expediently arranged or offloaded to other production resources. However, complex manufacturing projects often consist of multiple tasks with fixed sequences, posing a significant challenge for smart factories in efficiently scheduling limited robot resources to complete specific tasks. Additionally, when projects span across factories, ensuring faithful execution of contracts becomes another challenge. In this paper, we propose a modified combinatorial auction method combined with blockchain and edge computing technologies to organize project scheduling. Firstly, we transform efficient resource scheduling into a resource-constrained multi-project scheduling problem (RCPSP). Subsequently, the solution integrates combinatorial auction with random sampling (CA-RS) into smart contracts. Alongside security analysis, simulations are conducted using real data sets. The results indicate that the suggested CA-RS approach significantly enhances efficiency and security in resource arrangement within the industrial Internet of Things compared to baseline algorithms. [ABSTRACT FROM AUTHOR]

The current approach to fetal anomaly screening is based on biometric measurements derived from individually selected ultrasound images. In this paper, we introduce a paradigm shift that attains human-level performance in biometric measurement by aggregating automatically extracted biometrics from every frame across an entire scan, with no need for operator intervention. We use a neural network to classify each frame of an ultrasound video recording. We then measure fetal biometrics in every frame where appropriate anatomy is visible. We use a Bayesian method to estimate the true value of each biometric from a large number of measurements and probabilistically reject outliers. We performed a retrospective experiment on 1457 recordings (comprising 48 million frames) of 20-week ultrasound scans, estimated fetal biometrics in those scans and compared our estimates to real-time manual measurements. Our method achieves human-level performance in estimating fetal biometrics and estimates well-calibrated credible intervals for the true biometric value. [ABSTRACT FROM AUTHOR]

In the face of forest fire emergencies, fast and efficient dispatching of rescue vehicles is an important means of mitigating the damage caused by forest fires, and is an effective method of avoiding secondary damage caused by forest fires, minimizing the damage caused by forest fires to the ecosystem, and mitigating the losses caused by economic development. this paper takes the actual problem as the starting point, constructs a reasonable mathematical model of the problem, for the special characteristics of the emergency rescue vehicle scheduling problem of forest fires, taking into account the actual road conditions in the northern pristine forest area, through the analysis of the cost of paths between the forest area and the highway, to obtain the least obstructed rescue paths, to narrow the gap between the theoretical model and the problem of the actual. Improvement of ordinary genetic algorithm, design of double population strategy selection operation, the introduction of chaotic search initialization population, to improve the algorithm's solution efficiency and accuracy, through the northern pristine forest area of Daxing'anling real forest fire cases and generation of large-scale random fire point simulation experimental test to verify the effectiveness of the algorithm, to ensure that the effectiveness and reasonableness of the solution to the problem of forest fire emergency rescue vehicle scheduling program. It enriches the solution method of forest fire emergency rescue vehicle dispatching problem in Great Khingan area, which is of great significance to improve the emergency rescue capability in case of sudden forest fire. Through simulation experiments, the proposed Improved Genetic Algorithm (IGA) achieved an average rescue time reduction of 8.5% compared to conventional Genetic Algorithm (GA) and 3.5% compared to Improved Artificial Bee Colony (IABC) algorithm, with an average solution time of 9.4 ms. [ABSTRACT FROM AUTHOR]

Over the past two decades, machine learning (ML) has transformed manufacturing, particularly in optimizing production and quality control. A significant challenge in ML applications is obtaining sufficient training data, which data augmentation aims to address. While widely applied to image, text, and sound data, data augmentation for numerical data in manufacturing has seen limited investigation. This paper empirically compares three data augmentation techniques—generative adversarial networks, variational auto-encoders mixed with long-short-term memory, and warping—on four manufacturing datasets. It also provides a literature review, highlighting that generative models are the most common technique for numerical manufacturing data. Preliminary findings suggest that generative adversarial networks are effective for non-time-series numerical data, especially with datasets featuring many correlated model features, multiple machines, and sufficient instances and labels. This research enhances the understanding of data augmentation in manufacturing ML applications, emphasizing the need for tailored strategies. [ABSTRACT FROM AUTHOR]

To enhance enterprises' interactive exploration capabilities for unstructured chart data, this paper proposes a multimodal chart question-answering method. Facing the challenge of recognizing curved and irregular text in charts, we introduce Gaussian heatmap encoding technology to achieve character-level precise text annotation. Additionally, we combine a key point detection algorithm to extract numerical information from the charts and convert it into structured table data. Finally, by employing a multimodal cross-fusion model, we deeply integrate the queried charts, user questions, and generated table data to ensure that the model can comprehensively capture chart information and accurately answer user questions. Experimental validation has demonstrated that our method achieves a precision of 91.58% in chart information extraction and a chart question-answering accuracy of 82.24%, fully proving the significant advantages of our proposed method in enhancing chart text recognition and question-answering capabilities. Through practical enterprise application cases, our method has shown its ability to answer four types of chart questions, exhibiting mathematical reasoning capabilities and providing robust support for enterprise data analysis and decision-making. [ABSTRACT FROM AUTHOR]

Straightness is the basic measurement parameter in machining, and the traditional straightness measurement methods such as light gap method, table method, et al., have extremely low measurement efficiency and cannot achieve online real-time high-precision detection. Our research group has proposed a machine vision online detection based on 10 industrial camera arrays, which can obtain the surface profile straight line of the sucker rod by collecting the edge profile image of the sucker rod and performing morphological transformation. Compared with the traditional method, online nondestructive testing has good real-time performance and high accuracy, but due to the shortcomings of serious environmental pollution and strong noise in some industrial scenarios, the accuracy of the detected straight line is not high. This paper continues to discuss the straightness optimization design method, based on the probability t distribution of the data, reasonably select the sampled edge points, and remove the scene noise. Experiments show that, compared with the previous results, the straightness based on the probability and statistics method can strictly extract the contour sampling points with a confidence of 95%, and the accuracy is higher than that of the traditional Hough transform. [ABSTRACT FROM AUTHOR]

Intelligence technology has widely empowered education. As an example, Optical Character Recognition (OCR) can be used in smart education scenarios such as online homework correction and teaching data analysis. One of the fundamental yet challenging tasks is to recognize images of handwritten English text as editable text accurately. This is because handwritten text tends to have different writing habits as well as smearing and overlapping, resulting in hard alignment between the image and the real text. Additionally, the lack of data on handwritten text further leads to a lower recognition rate. To address the above issue, on the one hand, this paper extends the existing dataset and introduces hyphenated data annotation to provide data support for improving the robustness and discrimination of the model; on the other hand, a novel framework named Improved Multi-task Transformer based on Localization Mechanism Network (IMTLM-Net) is proposed for handwritten English text recognition. IMTLM-Net contains two parts, namely the encoding and decoding modules. The encoding module introduces a dual-stream processing mechanism. That is, in the simultaneous processing of text and images, a Vision Transformer (VIT) is utilized to encode images, and a Permutation Language Model (PLM) is designed for word arrangement. Two Multiple Head Attention (MHA) units are employed in the decoding module, focusing on text sequences and image sequences. Moreover, the localization mechanism (LM) is applied to enhance font structure feature extraction from image data, which in turn improves the model’s ability to capture complex details. Numerous experiments demonstrate that the proposed method achieves state-of-the-art results in handwritten text recognition. [ABSTRACT FROM AUTHOR]

To monitor the condition of cupping spots in real-time during the operation of the automatic cupping machine, reduce the influence of the surrounding environment on the image, and improve the segmentation accuracy of the cupping spots, this paper proposes a network called MCA-Deeplabv3+. Firstly, backbone network replaced by Mobilenetv2 to reduce the model size and improve feature extraction speed; Secondly, to further enhance the network’s feature extraction capabilities, we added dilated convolution channels and integrated the CA attention mechanism into the ASPP module; Finally, data augmentation and brightness adjustment are performed on the dataset to improve the generalization of the model in different environments. The experimental results show that, in comparison with other segmentation models, MCA-Deeplabv3+performs the best in cupping spot segmentation, with mIoU and mPA reaching 93.90% and 96.73%, respectively. The practicality and effectiveness of the cupping spot segmentation model presented in this paper are thoroughly demonstrated. [ABSTRACT FROM AUTHOR]

This paper proposes a fault diagnosis method for rotating machinery that integrates transfer learning with the ConvNeXt model (TL-CoCNN), addressing challenges such as small sample sizes and varying operating conditions. To meet the input requirements of the model while minimizing feature loss, an alternative approach to visualizing vibration data is introduced. Specifically, RGB images are synthesized from time-domain, frequency-domain, and time-frequency domain representations of the original signal, which are subsequently used as the input dataset. The fault diagnosis process leverages a pre-trained ConvNeXt model, initially trained on the ImageNet dataset, and fine-tunes its parameters using the synthesized RGB images to perform the fault classification task. Experimental results demonstrate that this data visualization method extracts more fault-related information compared to traditional time-domain and frequency-domain techniques, without the need to augment the sample size. The TL-CoCNN model achieves superior recognition accuracy when evaluated in terms of training time and model size across multiple test datasets. As an end-to-end fault diagnosis system, TL-CoCNN significantly enhances the feature representation capability of complex signals, showing promising potential for practical applications in fault detection and diagnosis. [ABSTRACT FROM AUTHOR]

Drone inspections are widely utilized in the detection of insulators in power lines. To address issues with traditional object detection algorithms, such as large parameter counts, low detection accuracy, and high miss rates, this paper proposes an insulator detection algorithm based on an improved YOLOv5 model. Firstly, in the backbone and neck networks, a lightweight CSP-SCConv module is employed to replace the original CSP-Darknet53 module, thereby reducing the parameter count and enhancing the feature extraction capabilities. Secondly, to broaden the image receptive field and improve feature fusion, a Receptive Field Block (RFB) model is incorporated into the neck network, replacing the original Spatial Pyramid Pooling Fast (SPPF) module. Additionally, a Lattice Structured Kernel (LSKBlock) attention mechanism is appended at the end of the neck network to further obtain richer semantic information. Finally, to flexibly improve the accuracy of bounding boxes of different sizes and enhance the robustness of the model, an loss function is utilized to replace the original Complete Intersection Over Union (CIOU) loss function. Experimental results demonstrate that the improved YOLOv5 model achieves a mean Average Precision (mAP) precision of 95.60%, with a parameter count of 18.36 M and a computational load of 30.10G, respectively. The Precision (P) and Recall (R) are 88.10% and 95.20%, providing strong support for deployment on mobile devices for real-time detection. [ABSTRACT FROM AUTHOR]

Symbolic anchoring is an important topic in robotics, as it enables robots to obtain symbolic knowledge from the perceptual information acquired through their sensors and maintain the link between that knowledge and the sensory data. In cognitive-based robots, this process of transforming sub-symbolic data generated by sensors to obtain and maintain symbolic knowledge is still an open problem. To address this issue, this paper presents SAILOR, a framework for symbolic anchoring integrated into ROS 2. SAILOR aims to maintain the link between symbolic data and perceptual data in real robots over time. It provides a semantic world modeling approach using two deep learning-based sub-symbolic robotic skills: object recognition and matching function. The object recognition skill allows the robot to recognize and identify objects in its environment, while the matching function enables the robot to decide if new perceptual data corresponds to existing symbolic data. This paper describes the proposed method and the development of the framework, as well as its integration in MERLIN2 (a hybrid cognitive architecture fully functional in robots running ROS 2) and the validation of SAILOR using public datasets and a real-world scenario. [ABSTRACT FROM AUTHOR]

Numerous applications have explored medical image classification using deep learning models. With the emergence of Explainable AI (XAI), researchers have begun to recognize its potential in validating the authenticity and correctness of results produced by black-box deep learning models. On the other hand, current diagnostic approaches for osteonecrosis face significant challenges, including difficulty in early detection, subjectivity in image interpretation, and reliance on surgical interventions without a comprehensive diagnostic foundation. This paper presents a novel Medical Computer-Aid-Diagnosis System—the Shadow Learning System framework—which integrates a convolutional neural network (CNN) with an Explainable AI method. This system not only performs conventional computer-aiding-diagnosis functions but also uniquely exploits misclassified data samples to provide additional medically relevant information from the machine learning model’s perspective, assisting doctors in their diagnostic process. The implementation of XAI techniques in our proposed system goes beyond merely validating CNN model results; it also enables the extraction of valuable information from medical images through an unconventional machine learning perspective. Our paper aims to enhance and extend the general structure and detailed design of the Shadow Learner System, making it more advantageous not only for human users but also for the deep learning model itself. A case study on femoral head osteonecrosis was conducted using our proposed system, which demonstrated improved accuracy and reliability in its prediction results. Experimental results interpreted using XAI methods are visualized to prove the confidence of our proposed model that generates reasonable results, confirming the effectiveness of the proposed model. [ABSTRACT FROM AUTHOR]

The modern decision-making environment is complex and dynamic. Global supply chain networks are increasingly exposed to unsystematic risks. Therefore, decision-makers need greater flexibility and precision to better manage uncertain market changes and complex international environments. To construct an accurate multi-criteria group decision-making (MCGDM) model, it is necessary to select appropriate evaluation criteria and identify alternative ranking methods based on specific decision problems. To develop a suitable MCGDM model for dynamic environments, this paper develops a new MCGDM model based on probabilistic interval-valued intuitionistic hesitant fuzzy sets (PIVIHFSs), regret theory, and the extended ELimination Et Choice Translating Reality (ELECTRE) III method. Firstly, this paper proposes two new aggregation operators, including the generalized probabilistic interval-valued intuitionistic hesitant fuzzy weighted averaging (GPIVIHFWA) operator and the generalized probabilistic interval-valued intuitionistic hesitant fuzzy weighted geometric (GPIVIHFWG) operator. To incorporate the decision-maker (DM)'s regret aversion, a bidirectional projection measure is proposed to calculate the advantages and disadvantages between two probabilistic interval-valued intuitionistic hesitant fuzzy elements (PIVIHFEs). The regret values of PIVIHFEs are determined using the bidirectional projection measure instead of utility values in the regret-rejoice function. Then, this paper constructs an extended ELECTRE III method and establishes a decision-making model based on the Borda rule for ranking and selecting the best alternatives. Finally, the effectiveness and robustness of the proposed model are verified through a numerical example, and the results are discussed through sensitivity analysis and comparative analysis. [ABSTRACT FROM AUTHOR]

Conventional feature extraction methods for speech emotion recognition often suffer from unidimensionality and inadequacy in capturing the full range of emotional cues, limiting their effectiveness. To address these challenges, this paper introduces a novel network model named Multi-Modal Speech Emotion Recognition Network (MMSERNet). This model leverages the power of multimodal and multiscale feature fusion to significantly enhance the accuracy of speech emotion recognition. MMSERNet is composed of three specialized sub-networks, each dedicated to the extraction of distinct feature types: cepstral coefficients, spectrogram features, and textual features. It integrates audio features derived from Mel-frequency cepstral coefficients and Mel spectrograms with textual features obtained from word vectors, thereby creating a rich, comprehensive representation of emotional content. The fusion of these diverse feature sets facilitates a robust multimodal approach to emotion recognition. Extensive empirical evaluations of the MMSERNet model on benchmark datasets such as IEMOCAP and MELD demonstrate not only significant improvements in recognition accuracy but also an efficient use of model parameters, ensuring scalability and practical applicability. [ABSTRACT FROM AUTHOR]

This paper presents a novel Compressive Sensing (CS) algorithm for Synthetic Aperture Radar (SAR) imaging, designed to overcome the resolution limitations found in conventional methods, particularly in low Signal-to-Noise Ratio (SNR) scenarios. By employing Millimeter-wave SAR, the proposed approach offers a robust solution for precise object detection and localization challenges. The study models SAR imaging using Linear Frequency Modulated signals to meet the requirements of the CS framework. Performance evaluations are conducted across various noise levels to assess the effectiveness of the proposed Modified-Orthogonal Matching Pursuit (M-OMP) algorithm. This algorithm is compared with other established algorithms, including Orthogonal Matching Pursuit, Compressive Sampling Matching Pursuit (CoSaMP), and Adaptive CoSaMP (A-CoSaMP). Results indicate that the M-OMP algorithm outperforms the other algorithms presented in this paper and surpasses existing methods in the literature in accurately identifying and localizing targets, even in low SNR conditions as - 10 dB. [ABSTRACT FROM AUTHOR]

Mapping the multi-view image group acquired by the sparse camera array into an element image array can effectively improve the display angle of stereoscopic source reconstruction images in the integral imaging display system, and can also avoid the crosstalk problem caused by the use of microlens array. However, the number of multi-view images collected by sparse camera array is small, and it is easy to lose details, which will increase the economic cost or time cost to collect sufficient number of multi-view images. Therefore, we propose a method to generate element image array based on mapping between a small number of multi-view images and homologous pixels. The pixel mapping relationship between multi-view image group and element image array is established. The experimental results show that the element image generated by the method proposed in this paper has a larger display view field and smooth transition of each view interval after stereoscopic source reconstruction. [ABSTRACT FROM AUTHOR]

Surface electromyography-based gesture recognition was widely applied in human–computer interaction, hand rehabilitation, prosthetic control, and other fields. Electromyography (EMG) signals-based gesture classification usually relies on handcrafted feature extraction with intense subjectivity or convolutional neural networks with redundant structures to extract features. This paper converts the raw EMG signals into Gramian Angular Difference Field (GADF) and Gramian Angular Summation Field images. Four models were used to classify the pictures: K-nearest Neighbors (KNN), Generalized Learning Systems, Binary Trees, and Convolutional Neural Networks using MobileNetv1, and the proposed method was verified by using the public dataset NinaproDB2. Experimental results: When the window size is 300 ms, the step size is 10 ms, and KNN are used as the classification model, the average accuracy of EMG signals classification based on the GADF method is 98.17%, and the accuracy of exercises B, C and D was 96.65%, 95.53%, and 98.02%, respectively. The recognition accuracy was 7.92%, 14.25%, and 4.279% higher than the provided baseline. [ABSTRACT FROM AUTHOR]

Speech emotion recognition is probably among the most exciting and dynamic areas of modern research focused on speech signals analysis, which allows estimating and classifying speakers' rich spectrum of emotions. The following paper aims to develop a novel deep learning (DL)-based model for detecting speech emotion variation to overcome several weaknesses of the existing intelligent data-driven approaches. A new architecture for a DL network, referred to as the RNN–CNN, is proposed and applied in this paper to perform the SER task by operating directly on raw speech signals. Specifically, the challenge was effectively combining an initial convolution layer with a wide kernel as an efficient way to address and mitigate the problems caused by noise found in raw speech signals. In this experimental analysis, the 3 databases used to evaluate the proposed RNN–CNN model are RML, RAVDESS, and SAVEE. The effectiveness of such methodologies can be detected with remarkable efficacy, whose improved accuracy rates depict contrasting trends from those findings of the previous works analyzed through respective datasets. This assessment has validated the robust performance and applicability of the suggested models for diverse speech databases and underlined their potential in further speech-based emotion recognition. [ABSTRACT FROM AUTHOR]

The classification of degradation in historical document images plays a pivotal role in their preservation and restoration. This paper introduces a novel approach for noise classification using classical machine learning techniques, specifically Multi-Layer Perceptrons (MLPs). We assembled a comprehensive dataset of historical documents from a range of public sources, from which global and local statistical features were extracted for MLP training and validation. Through extensive experimentation, we determined the optimal MLP architecture and evaluated its performance. The model was rigorously tested through both unblind and blind testing scenarios. Unblind testing, utilizing images from the same collections as the training set, achieved a robust accuracy of 97.22%. Blind testing, performed with the distinct PHIBD dataset, demonstrated a 90% accuracy, outperforming current state-of-the-art deep learning models. These results affirm the model’s robustness and its potential for practical application in historical document analysis. [ABSTRACT FROM AUTHOR]

In recent years, with the rapid development of deep learning, the application in the field of synthetic aperture radar (SAR) image ship detection has received increasing attention. Most of the existing algorithms suffer from the problems of more network parameters and larger computation, so they are usually difficult to be deployed on embedded devices with limited storage and computational resources. The lightweight detection network cannot achieve good detection accuracy in complex scenes due to its small number of parameters and insufficient feature extraction and fusion for small-sized ships. Aiming at the above problems, in this paper, based on the YOLOv5 detection algorithm, the backbone and neck networks are designed to be lightweight, and the improved spatial feature pyramid module SimSPPFC is proposed to replace the SPPF module in YOLOv5, and the convolutional block attention module (CBAM) is used to construct an efficient feature extraction module to improve the feature extraction and selection capability of the network, and the weighted average introduced to the weighted non-maximum suppression (NMS) is introduced to replace the traditional NMS as the post-processing algorithm to further improve the localization accuracy of the prediction box. Experiments show that the proposed algorithm achieves 97.7% AP50 and 66.8% AP50:95 on the SSDD dataset with 0.438 M number of parameters and 1.1GFLOPs computational complexity. The algorithm takes into account both the detection accuracy and efficiency, and realizes a good trade-off between algorithmic lightweighting and detection accuracy. [ABSTRACT FROM AUTHOR]

In the safety monitoring of workers in manufacturing enterprises, there is a problem of large amount of calculation in the identification model caused by multi-objective high concurrent behavior. In this paper, we propose a human behavior recognition model that combines multi-dimensional convolution and gated recurrent neural network from the perspective of model structure design. The single target human behavior data set of the factory was constructed by YOLOv7 target detection and BOT-SORT multi-target tracking. Human behavior recognition model 3-2DCNN-BIGRU, in the mixed spatio-temporal feature extraction layer, uses the advantages of 3DCNN in spatio-temporal feature extraction to extract spatio-temporal features; The 3-2DCNN is used to extract spatial features after dimension reduction to improve the computational complexity and reduce the complexity of the model; Using the idea of expansion convolution in the time convolution network, the receptive fields of 3DCNN and 3-2DCNN are increased, and the ability of spatio-temporal feature extraction of the model is enhanced. In the time feature enhancement layer, a bidirectional gated recurrent neural network is fused to enhance the model’s ability to extract time features, thereby improving the overall performance of the model. With fewer parameters, the accuracy on the Fall Dataset reaches 98.65%, which can effectively identify human behaviors such as walking, sitting, and falling in factories, ensuring the safety of workers. [ABSTRACT FROM AUTHOR]

This paper proposes a robust adaptive algorithm for smooth graph signal recovery which is based on generalized correntropy. A proper cost function is defined, which takes the smoothness and generalized correntropy into account. The generalized correntropy used in this paper employs the generalized Gaussian density (GGD) function as the kernel. The proposed adaptive algorithm is derived and a kernel width learning-based version of the algorithm is suggested. The simulation results confirm the performance of the proposed algorithm for learning the kernel-width to the fixed correntropy kernel version of the algorithm. Moreover, some theoretical analyses of the proposed algorithm are provided. In this regard, firstly, the convexity analysis of the cost function is discussed. Secondly, the uniform stability of the algorithm is investigated. Thirdly, the mean convergence analysis is also added. Finally, the computational complexity analysis of the algorithm is incorporated. In addition, some synthetic and real-world experiments show the efficiency of the proposed algorithm in comparison to some other adaptive algorithms in the literature of adaptive graph signal recovery. [ABSTRACT FROM AUTHOR]

Based on the traditional three-dimensional Hopfield neural network (HNN), this paper introduces the sine function into the cubic nonlinear memristor, constructs a new memristor and simulates its hysteresis loop. The memristor is simulated as a memristor synapse and a new four-dimensional memristive HNN (MHNN) model is constructed. We analyzed the stability of the equilibrium point of the model, plotted the Lyapunov exponent spectrum, bifurcation diagram, and phase diagram, and studied the dynamic characteristics of the MHNN model, the results show that the model has a wide range of coupling coefficients and memristor parameters. With the development of Internet of Things (IoT) technology, secure communication has become increasingly important. At present, video conferencing applications are becoming more and more widespread, and the hidden dangers of security and confidentiality should not be underestimated. Enterprise video conferencing often designs a large amount of trade secrets and private data in the process of office, facing the risk of being intercepted and data leakage. In this paper, a set of solutions to solve the encrypted transmission of enterprise video conferences is proposed. This scheme uses the proposed four-dimensional MHNN system to achieve color video encryption transmission on the Raspberry Pi platform, which has important research significance in IoT video encryption. [ABSTRACT FROM AUTHOR]

To meet the real-time and high-precision detection requirements in actual yarn production, this paper introduces a novel lightweight yarn detection method named PHL-YOLO, an optimized version of the YOLOv8 algorithm. Initially, we developed the PEC2F module by fusing PConv (partial convolution) and EMA (efficient multi-scale attention) components to enhance the model's feature extraction capabilities while concurrently reducing computational demands and the number of parameters. Subsequently, we deployed the HWD (Haar wavelet downsampling) module for downsampling, which effectively diminishes image resolution without sacrificing the critical information necessary for precise detection. Additionally, to address the excessive parameters and computational complexity in the detection head of YOLOv8, we introduced the LSCDH (lightweight shared convolutional detection head) module, which significantly simplifies the structure of the detection head. Finally, we applied a structured pruning technique, GSPMD (group-level structured pruning method with DepGraph), to further refine the model. The experimental outcomes indicate that our enhanced model PHL-YOLO has achieved a 1.0% increase in precision, a 2.5% improvement in recall, and a 2.3% increase in mAP (mean average precision), along with a 26 FPS performance boost over the original model. Furthermore, our model’s computational load and parameter count are only 29.6% and 10.2% of those in the original YOLOv8 model, respectively. PHL-YOLO not only effectively reduces model complexity, but also upholds high performance levels, providing a valuable reference for rapid and accurate yarn detection in real-world production scenarios. [ABSTRACT FROM AUTHOR]

Probabilistic forecasting offers insights beyond point estimates, supporting more informed decision-making. This paper introduces the Neural Quantile Function with Recurrent Neural Networks (NQF-RNN), a model for multistep-ahead probabilistic time series forecasting. NQF-RNN combines neural quantile functions with recurrent neural networks, enabling applicability across diverse time series datasets. The model uses a monotonically increasing neural quantile function and is trained with a continuous ranked probability score (CRPS)-based loss function. NQF-RNN’s performance is evaluated on synthetic datasets generated from multiple distributions and six real-world time series datasets with both periodicity and irregularities. NQF-RNN demonstrates competitive performance on synthetic data and outperforms benchmarks on real-world data, achieving lower average forecast errors across most metrics. Notably, NQF-RNN surpasses benchmarks in CRPS, a key probabilistic metric, and tail-weighted CRPS, which assesses tail event forecasting with a narrow prediction interval. The model outperforms other deep learning models by 5% to 41% in CRPS, with improvements of 5% to 53% in left tail-weighted CRPS and 6% to 34% in right tail-weighted CRPS. Against its baseline model, DeepAR, NQF-RNN achieves a 41% improvement in CRPS, indicating its effectiveness in generating reliable prediction intervals. These results highlight NQF-RNN’s robustness in managing complex and irregular patterns in real-world forecasting scenarios. [ABSTRACT FROM AUTHOR]

Predicting future sales volumes of fashion industry products is challenging due to rapid market changes and limited historical sales data for recent products. As traditional forecasting methods and machine learning models often fail to address this problem, we propose a novel autoregressive multimodal transformer architecture to anticipate the sales volume of brand-new apparel items by capturing trends among interrelated attributes. In this paper, we utilize authentic data from a fashion company that includes a limited amount of historical time-series sales data and several influencing factors like product image, textual descriptions, and temporal attributes. To mitigate the data inadequacies, we investigate the impact of integrating exogenous knowledge from an e-tailer site filtered with fashion apparel products. Also, we found that employing the zero-shot forecasting approach further aids in forecasting with minimal time-series sales data. Our approach achieves the values of 1.546 and 16.42 in terms of MAE and WAPE, respectively, by leveraging exogenous data compared to existing benchmark models. This study demonstrates the potential of our autoregressive multimodal transformer to predict sales volumes with more precision, and it highlights the importance of incorporating the zero-shot forecasting approach in the dynamic fashion industry. [ABSTRACT FROM AUTHOR]

Feature correspondences identification between consecutive frames is a critical prerequisite in the monocular Visual-Inertial Navigation System (VINS). In this paper, we propose a novel self-adaptation feature point correspondences identification algorithm in terms of IMU-aided information fusion at the level of feature tracking for nonlinear optimization framework-based VINS. This method starts with an IMU pre-integration predictor to predict the pose of each new coming frame. In weak texture scenes and motion blur situations, in order to increase the number of feature correspondences and improve the track lengths of feature points, we introduce a novel predicting-matching based feature point tracking strategy to build new matches. On the other hand, the predicted pose is incorporated into the outliers rejection step to deal with mismatch caused by dynamic objects. Finally, the proposed self-adaptation feature correspondences identification algorithm is implemented based on VINS-Fusion and validated through public datasets. The experimental results show that it effectively improves the accuracy and tracking length of feature matching, and demonstrates better performance in terms of camera pose estimation as compared to state-of-the-art approaches. [ABSTRACT FROM AUTHOR]

In the domain of medical science, emotion recognition based on electroencephalogram (EEG) has been widely used in emotion computing. Despite the prevalence of deep learning in EEG signals analysis, standard convolutional and recurrent neural networks fall short in effectively processing EEG data due to their inherent limitations in capturing global dependencies and addressing the non-linear and unstable characteristics of EEG signals. We propose a dual transfer learning method based on 3D Convolutional Neural Networks (3D-CNN) with a Vision Transformer (ViT) to enhance emotion recognition. This paper aims to utilize 3D-CNN effectively to capture the spatial characteristics of EEG signals and reduce data covariance, extracting shallow features. Additionally, ViT is incorporated to improve the model’s ability to capture long-range dependencies, facilitating deep feature extraction. The methodology involves a two-stage process: initially, the front end of a pre-trained 3D-CNN is employed as a shallow feature extractor to mitigate EEG data covariance and transformer biases, focusing on low-level feature detection. The subsequent stage utilizes ViT as a deep feature extractor, adept at modeling the global aspects of EEG signals and employing attention mechanisms for precise classification. We also present an innovative algorithm for data mapping in transfer learning, ensuring consistent feature representation across both spatio-temporal dimensions. This approach significantly improves global feature processing and long-range dependency detection, with the integration of color channels augmenting the model’s sensitivity to signal variations. In a 10-fold cross-validation experiment on the DEAP, experimental results demonstrate that the proposed method achieves classification accuracies of 92.44 % and 92.85 % for the valence and arousal dimensions, and the accuracies of four-class classification across valence and arousal are HVHA: 88.01 % , HVLA: 88.27 % , LVHA: 90.89 % , LVLA: 78.84 % . Similarly, it achieves an accuracy of 98.69 % on the SEED. Overall, this methodology not only holds substantial potential in advancing emotion recognition tasks but also contributes to the broader field of affective computing. [ABSTRACT FROM AUTHOR]

Path planning in automatic parking is a significant challenge due to constrained parking spaces and numerous obstacles. To enhance both the safety and efficiency of the planned path, this paper proposes a bidirectional hybrid A* algorithm for narrow spaces with a high density of obstacles. A vehicle obstacle avoidance method that incorporates rectangular expansion through numerical analysis is proposed to achieve collision-free navigation. Meanwhile, a safety cost is integrated into the hybrid A* search algorithm to maintain a sufficient safety distance between the planned path and obstacles. Additionally, to enhance the efficiency of path planning, a bidirectional search method is combined with the hybrid A* algorithm, with the addition of a bidirectional cohesive item cost. Finally, simulation experiments are conducted to generate parking paths for both vertical and parallel parking scenarios. The simulation results indicate that the proposed algorithm obtains a sufficient safety distance, reduced search time, and fewer expanded nodes. Meanwhile, the stability and adaptability of the proposed method are analyzed. The comparison with other algorithms suggests that the proposed algorithm provides a larger safety distance and shorter search time. [ABSTRACT FROM AUTHOR]

Fuzzy c-means clustering (FCM) has gained widespread application because of its ability to capture uncertain information in data effectively. However, attributed to the prior assumption of identical distribution, traditional FCM is sensitive to noise and cluster size. Modified methods incorporating local spatial information can enhance the robustness to noise. However, they tend to balance cluster sizes, resulting in poor performance when dealing with imbalanced data. Modified methods learning the statistical characteristics of data are feasible to handle imbalanced data. However, they are often sensitive to noise due to the ignorance of local information. Aiming at the lack of method that can simultaneously alleviate the sensitivity to noise and cluster size, a double fuzzy relaxation local information c-means clustering algorithm (DFRLICM) is proposed in this paper. Firstly, sample relaxation is introduced to explore potential clustering results and enhance inter-class separability. Secondly, to cooperate with the relaxation, we design fuzzy weights to record the imbalance situation of data clusters, enhancing the capability of algorithm in dealing with imbalanced data. Thirdly, we introduce fuzzy factor to account for the preservation of local structures in data and improve the robustness of algorithm. Finally, we integrate the three elements into a unified model framework to achieve the combination optimization of robustness to noise and insensitivity to cluster size simultaneously. Extensive experiments are conducted and the results demonstrate that the proposed algorithm indeed achieves a balance between robustness to noise and insensitivity to cluster size. [ABSTRACT FROM AUTHOR]

Manually-designed network architectures for thermal infrared pedestrian tracking (TIR-PT) require substantial effort from human experts. AlexNet and ResNet are widely used as backbone networks in TIR-PT applications. However, these architectures were originally designed for image classification and object detection tasks, which are less complex than the challenges presented by TIR-PT. This paper makes an early attempt to search an optimal network architecture for TIR-PT automatically, employing single-bottom and dual-bottom cells as basic search units and incorporating eight operation candidates within the search space. To expedite the search process, a random channel selection strategy is employed prior to assessing operation candidates. Classification, batch hard triplet, and center loss are jointly used to retrain the searched architecture. The outcome is a high-performance network architecture that is both parameter- and computation-efficient. Extensive experiments proved the effectiveness of the automated method. [ABSTRACT FROM AUTHOR]

LiDAR Simultaneous Localization and Mapping (SLAM) plays a crucial role in intelligent robotics, finding extensive applications in autonomous driving and exploration. The traditional feature-based LiDAR SLAM holds a prominent position due to its robustness and accuracy. However, these methods still exhibit limitations in point cloud preprocessing and feature extraction. In this paper, we introduce an effective LiDAR SLAM method to address these issues. Specifically, we propose a novel Concentric Cluster Model (CCM) for clustering point clouds, aiming to preserve stable point clouds and eliminate the unstable ones. Additionally, we propose a Step-by-step Feature Extraction (SFE), which significantly enhances the effect of traditional feature extraction methods. We test the proposed SLAM method on several sequences of the KITTI odometry, M2DGR, and M2DGR-plus datasets. Experimental results show that our method achieves superior accuracy compared to several state-of-the-art LiDAR SLAM methods, while maintaining real-time performance. [ABSTRACT FROM AUTHOR]

Bearings are critical components in machinery, and accurately predicting their remaining useful life (RUL) is essential for effective predictive maintenance. Traditional RUL prediction methods often rely on manual feature extraction and expert knowledge, which face specific challenges such as handling non-stationary data and avoiding overfitting due to the inclusion of numerous irrelevant features. This paper presents an approach that leverages Continuous Wavelet Transform (CWT) for feature extraction, a Channel-Temporal Mixed MLP (CT-MLP) layer for capturing intricate dependencies, and a dynamic attention mechanism to adjust its focus based on the temporal importance of features within the time series. The dynamic attention mechanism integrates multi-head attention with innovative enhancements, making it particularly effective for datasets exhibiting non-stationary behaviour. An experimental study using the XJTU-SY rolling bearings dataset and the PRONOSTIA bearing dataset revealed that the proposed deep learning algorithm significantly outperforms other state-of-the-art algorithms in terms of RMSE and MAE, demonstrating its robustness and accuracy. [ABSTRACT FROM AUTHOR]

Internet of Things (IoT)-based health monitoring is a key focus in Healthcare 4.0 for continuously tracking Vital signs and other health biometrics. While IoT technologies like sensors, connectivity, hardware, software, and batteries have been extensively studied, there has been a lack of emphasis on evaluating human-centered health capabilities using long-term data collected by the Internet of Medical Things (IoMT). This paper tackles the problem of human-centered Vital signs-based health long-term assessment by proposing a hybrid approach that includes model-free multi-profile monitoring, optimization models, control charts, and profile capability indices. In this new approach, a curves dissimilarity index-based profile monitoring method is employed to address the complexity and outlier elimination issues in the traditional regression model-based profile monitoring approach, and a mixed-integer linear mathematical model is developed to assign human-centered tolerance to the dissimilarity index of each Vital sign profile. Also, control charts are employed to monitor the health of individuals according to each Vital sign, and overall. In addition, the profile capability index and non-confirming probability values are interpreted as long-term health measures. The efficacy of the newly proposed method is demonstrated in a simulation study based on an individual's six Vital signs generated data set using MATLAB, CPLEX, and MINITAB. [ABSTRACT FROM AUTHOR]

This paper identifies and analyzes applications in which Large Language Models (LLMs) can make Internet of Things (IoT) networks more intelligent and responsive through three case studies from critical topics: DDoS attack detection, macroprogramming over IoT systems, and sensor data processing. Our results reveal that the GPT model under few-shot learning achieves 87.6% detection accuracy, whereas the fine-tuned GPT increases the value to 94.9%. Given a macroprogramming framework, the GPT model is capable of writing scripts using high-level functions from the framework to handle possible incidents. Moreover, the GPT model shows efficacy in processing a vast amount of sensor data by offering fast and high-quality responses, which comprise expected results and summarized insights. Overall, the model demonstrates its potential to power a natural language interface. We hope that researchers will find these case studies inspiring to develop further. Article Highlights: Large language models' performance surpassed traditional machine learning models' on IoT cyber threat detection. Large language model demonstrated sufficient understanding in complicated IoT macroprogramming framwork. Large language models can contribute to automatic data processing by producing high-quality code scripts. [ABSTRACT FROM AUTHOR]

Neurodynamic observations indicate that the cerebral cortex evolved by self-organizing into functional networks, These networks, or distributed clusters of regions, display various degrees of attention maps based on input. Traditionally, the study of network self-organization relies predominantly on static data, overlooking temporal information in dynamic neuromorphic data. This paper proposes Temporal Self-Organizing (TSO) method for neuromorphic data processing using a spiking neural network. The TSO method incorporates information from multiple time steps into the selection strategy of the Best Matching Unit (BMU) neurons. It enables the coupled BMUs to radiate the weight across the same layer of neurons, ultimately forming a hierarchical self-organizing topographic map of concern. Additionally, we simulate real neuronal dynamics, introduce a glial cell-mediated Glial-LIF (Leaky Integrate-and-fire) model, and adjust multiple levels of BMUs to optimize the attention topological map.Experiments demonstrate that the proposed Self-organizing Glial Spiking Neural Network (SG-SNN) can generate attention topographies for dynamic event data from coarse to fine. A heuristic method based on cognitive science effectively guides the network's distribution of excitatory regions. Furthermore, the SG-SNN shows improved accuracy on three standard neuromorphic datasets: DVS128-Gesture, CIFAR10-DVS, and N-Caltech 101, with accuracy improvements of 0.3%, 2.4%, and 0.54% respectively. Notably, the recognition accuracy on the DVS128-Gesture dataset reaches 99.3%, achieving state-of-the-art (SOTA) performance. [ABSTRACT FROM AUTHOR]

Brainprint recognition technology, regarded as a promising biometric technology, encounters challenges stemming from the time-varied, low signal-to-noise ratio of brain signals, such as electroencephalogram (EEG). Steady-state visual evoked potentials (SSVEP) exhibit high signal-to-noise ratio and frequency locking, making them a promising paradigm for brainprint recognition. Consequently, the extraction of time-invariant identity information from SSVEP EEG signals is essential. In this paper, we propose an Attentive Multi-sub-band Depth Identity Embedding Learning Network for stable cross-session SSVEP brainprint recognition. To address the issue of low recognition accuracy across sessions, we introduce the Sub-band Attentive Frequency mechanism, which integrates the frequency-domain relevant characteristics of the SSVEP paradigm and focuses on exploring depth-frequency identity embedding information. Also, we employ Attentive Statistic Pooling to enhance the stability of frequency domain feature distributions across sessions. Extensive experimentation and validation were conducted on two multi-session SSVEP benchmark datasets. The experimental results show that our approach outperforms other state-of-art models on 2-second samples across sessions and has the potential to serve as a benchmark in multi-subject biometric recognition systems. [ABSTRACT FROM AUTHOR]