To enhance speech emotion recognition capability,this study constructs a speech emotion recognition model integrating the adaptive acoustic mixup(AAM)and improved coordinate and shuffle attention(ICASA)methods.The AAM...To enhance speech emotion recognition capability,this study constructs a speech emotion recognition model integrating the adaptive acoustic mixup(AAM)and improved coordinate and shuffle attention(ICASA)methods.The AAM method optimizes data augmentation by combining a sample selection strategy and dynamic interpolation coefficients,thus enabling information fusion of speech data with different emotions at the acoustic level.The ICASA method enhances feature extraction capability through dynamic fusion of the improved coordinate attention(ICA)and shuffle attention(SA)techniques.The ICA technique reduces computational overhead by employing depth-separable convolution and an h-swish activation function and captures long-range dependencies of multi-scale time-frequency features using the attention weights.The SA technique promotes feature interaction through channel shuffling,which helps the model learn richer and more discriminative emotional features.Experimental results demonstrate that,compared to the baseline model,the proposed model improves the weighted accuracy by 5.42%and 4.54%,and the unweighted accuracy by 3.37%and 3.85%on the IEMOCAP and RAVDESS datasets,respectively.These improvements were confirmed to be statistically significant by independent samples t-tests,further supporting the practical reliability and applicability of the proposed model in real-world emotion-aware speech systems.展开更多
Brain tumors require precise segmentation for diagnosis and treatment plans due to their complex morphology and heterogeneous characteristics.While MRI-based automatic brain tumor segmentation technology reduces the b...Brain tumors require precise segmentation for diagnosis and treatment plans due to their complex morphology and heterogeneous characteristics.While MRI-based automatic brain tumor segmentation technology reduces the burden on medical staff and provides quantitative information,existing methodologies and recent models still struggle to accurately capture and classify the fine boundaries and diverse morphologies of tumors.In order to address these challenges and maximize the performance of brain tumor segmentation,this research introduces a novel SwinUNETR-based model by integrating a new decoder block,the Hierarchical Channel-wise Attention Decoder(HCAD),into a powerful SwinUNETR encoder.The HCAD decoder block utilizes hierarchical features and channelspecific attention mechanisms to further fuse information at different scales transmitted from the encoder and preserve spatial details throughout the reconstruction phase.Rigorous evaluations on the recent BraTS GLI datasets demonstrate that the proposed SwinHCAD model achieved superior and improved segmentation accuracy on both the Dice score and HD95 metrics across all tumor subregions(WT,TC,and ET)compared to baseline models.In particular,the rationale and contribution of the model design were clarified through ablation studies to verify the effectiveness of the proposed HCAD decoder block.The results of this study are expected to greatly contribute to enhancing the efficiency of clinical diagnosis and treatment planning by increasing the precision of automated brain tumor segmentation.展开更多
Reliable traffic flow prediction is crucial for mitigating urban congestion.This paper proposes Attentionbased spatiotemporal Interactive Dynamic Graph Convolutional Network(AIDGCN),a novel architecture integrating In...Reliable traffic flow prediction is crucial for mitigating urban congestion.This paper proposes Attentionbased spatiotemporal Interactive Dynamic Graph Convolutional Network(AIDGCN),a novel architecture integrating Interactive Dynamic Graph Convolution Network(IDGCN)with Temporal Multi-Head Trend-Aware Attention.Its core innovation lies in IDGCN,which uniquely splits sequences into symmetric intervals for interactive feature sharing via dynamic graphs,and a novel attention mechanism incorporating convolutional operations to capture essential local traffic trends—addressing a critical gap in standard attention for continuous data.For 15-and 60-min forecasting on METR-LA,AIDGCN achieves MAEs of 0.75%and 0.39%,and RMSEs of 1.32%and 0.14%,respectively.In the 60-min long-term forecasting of the PEMS-BAY dataset,the AIDGCN out-performs the MRA-BGCN method by 6.28%,4.93%,and 7.17%in terms of MAE,RMSE,and MAPE,respectively.Experimental results demonstrate the superiority of our pro-posed model over state-of-the-art methods.展开更多
随着光伏发电在全球能源体系中占比不断提升,超短期光伏发电量预测对电力系统调度与安全运行至关重要。然而,光伏发电量受多因素影响,具有显著随机性与波动性。为此,提出了一种基于TCN-BiLSTM-Attention模型的超短期光伏发电量预测方法...随着光伏发电在全球能源体系中占比不断提升,超短期光伏发电量预测对电力系统调度与安全运行至关重要。然而,光伏发电量受多因素影响,具有显著随机性与波动性。为此,提出了一种基于TCN-BiLSTM-Attention模型的超短期光伏发电量预测方法。首先通过皮尔逊相关分析筛选关键特征,并利用孤立森林算法检测异常值,结合线性插值法和标准化完成数据预处理。随后,通过时间卷积网络(Temporal Convolutional Network,TCN)提取时序特征,再利用双向长短期记忆网络(Bidirectional Long Short-Term Memory,BiLSTM)网络捕获前后向时间依赖关系,并在输出端引入注意力机制聚焦关键时间步特征。最后,在Desert Knowledge Australia Solar Centre(DKASC)数据集上的对比实验表明,与传统LSTM、BiLSTM模型相比,提出的TCN-BiLSTM-Attention模型在预测精度、稳定性等方面均表现出一定优势。展开更多
Salient object detection(SOD)models struggle to simultaneously preserve global structure,maintain sharp object boundaries,and sustain computational efficiency in complex scenes.In this study,we propose SPSALNet,a task...Salient object detection(SOD)models struggle to simultaneously preserve global structure,maintain sharp object boundaries,and sustain computational efficiency in complex scenes.In this study,we propose SPSALNet,a task-driven two-stage(macro–micro)architecture that restructures the SOD process around superpixel representations.In the proposed approach,a“split-and-enhance”principle,introduced to our knowledge for the first time in the SOD literature,hierarchically classifies superpixels and then applies targeted refinement only to ambiguous or error-prone regions.At the macro stage,the image is partitioned into content-adaptive superpixel regions,and each superpixel is represented by a high-dimensional region-level feature vector.These representations define a regional decomposition problem in which superpixels are assigned to three classes:background,object interior,and transition regions.Superpixel tokens interact with a global feature vector from a deep network backbone through a cross-attention module and are projected into an enriched embedding space that jointly encodes local topology and global context.At the micro stage,the model employs a U-Net-based refinement process that allocates computational resources only to ambiguous transition regions.The image and distance–similarity maps derived from superpixels are processed through a dual-encoder pathway.Subsequently,channel-aware fusion blocks adaptively combine information from these two sources,producing sharper and more stable object boundaries.Experimental results show that SPSALNet achieves high accuracy with lower computational cost compared to recent competing methods.On the PASCAL-S and DUT-OMRON datasets,SPSALNet exhibits a clear performance advantage across all key metrics,and it ranks first on accuracy-oriented measures on HKU-IS.On the challenging DUT-OMRON benchmark,SPSALNet reaches a MAE of 0.034.Across all datasets,it preserves object boundaries and regional structure in a stable and competitive manner.展开更多
In daily life,keyword spotting plays an important role in human-computer interaction.However,noise often interferes with the extraction of time-frequency information,and achieving both computational efficiency and rec...In daily life,keyword spotting plays an important role in human-computer interaction.However,noise often interferes with the extraction of time-frequency information,and achieving both computational efficiency and recognition accuracy on resource-constrained devices such as mobile terminals remains a major challenge.To address this,we propose a novel time-frequency dual-branch parallel residual network,which integrates a Dual-Branch Broadcast Residual module and a Time-Frequency Coordinate Attention module.The time-domain and frequency-domain branches are designed in parallel to independently extract temporal and spectral features,effectively avoiding the potential information loss caused by serial stacking,while enhancing information flow and multi-scale feature fusion.In terms of training strategy,a curriculum learning approach is introduced to progressively improve model robustness fromeasy to difficult tasks.Experimental results demonstrate that the proposed method consistently outperforms existing lightweight models under various signal-to-noise ratio(SNR)conditions,achieving superior far-field recognition performance on the Google Speech Commands V2 dataset.Notably,the model maintains stable performance even in low-SNR environments such as–10 dB,and generalizes well to unseen SNR conditions during training,validating its robustness to novel noise scenarios.Furthermore,the proposed model exhibits significantly fewer parameters,making it highly suitable for deployment on resource-limited devices.Overall,the model achieves a favorable balance between performance and parameter efficiency,demonstrating strong potential for practical applications.展开更多
Autonomous vehicles rely heavily on accurate and efficient scene segmentation for safe navigation and efficient operations.Traditional Bird’s Eye View(BEV)methods on semantic scene segmentation,which leverage multimo...Autonomous vehicles rely heavily on accurate and efficient scene segmentation for safe navigation and efficient operations.Traditional Bird’s Eye View(BEV)methods on semantic scene segmentation,which leverage multimodal sensor fusion,often struggle with noisy data and demand high-performance GPUs,leading to sensor misalignment and performance degradation.This paper introduces an Enhanced Channel Attention BEV(ECABEV),a novel approach designed to address the challenges under insufficient GPU memory conditions.ECABEV integrates camera and radar data through a de-noise enhanced channel attention mechanism,which utilizes global average and max pooling to effectively filter out noise while preserving discriminative features.Furthermore,an improved fusion approach is proposed to efficiently merge categorical data across modalities.To reduce computational overhead,a bilinear interpolation layer normalizationmethod is devised to ensure spatial feature fidelity.Moreover,a scalable crossentropy loss function is further designed to handle the imbalanced classes with less computational efficiency sacrifice.Extensive experiments on the nuScenes dataset demonstrate that ECABEV achieves state-of-the-art performance with an IoU of 39.961,using a lightweight ViT-B/14 backbone and lower resolution(224×224).Our approach highlights its cost-effectiveness and practical applicability,even on low-end devices.The code is publicly available at:https://github.com/YYF-CQU/ECABEV.git.展开更多
Tomato is a major economic crop worldwide,and diseases on tomato leaves can significantly reduce both yield and quality.Traditional manual inspection is inefficient and highly subjective,making it difficult to meet th...Tomato is a major economic crop worldwide,and diseases on tomato leaves can significantly reduce both yield and quality.Traditional manual inspection is inefficient and highly subjective,making it difficult to meet the requirements of early disease identification in complex natural environments.To address this issue,this study proposes an improved YOLO11-based model,YOLO-SPDNet(Scale Sequence Fusion,Position-Channel Attention,and Dual Enhancement Network).The model integrates the SEAM(Self-Ensembling Attention Mechanism)semantic enhancement module,the MLCA(Mixed Local Channel Attention)lightweight attention mechanism,and the SPA(Scale-Position-Detail Awareness)module composed of SSFF(Scale Sequence Feature Fusion),TFE(Triple Feature Encoding),and CPAM(Channel and Position Attention Mechanism).These enhancements strengthen fine-grained lesion detection while maintaining model lightweightness.Experimental results show that YOLO-SPDNet achieves an accuracy of 91.8%,a recall of 86.5%,and an mAP@0.5 of 90.6%on the test set,with a computational complexity of 12.5 GFLOPs.Furthermore,the model reaches a real-time inference speed of 987 FPS,making it suitable for deployment on mobile agricultural terminals and online monitoring systems.Comparative analysis and ablation studies further validate the reliability and practical applicability of the proposed model in complex natural scenes.展开更多
Stereo matching is a pivotal task in computer vision,enabling precise depth estimation from stereo image pairs,yet it encounters challenges in regions with reflections,repetitive textures,or fine structures.In this pa...Stereo matching is a pivotal task in computer vision,enabling precise depth estimation from stereo image pairs,yet it encounters challenges in regions with reflections,repetitive textures,or fine structures.In this paper,we propose a Semantic-Guided Parallax Attention Stereo Matching Network(SGPASMnet)that can be trained in unsupervised manner,building upon the Parallax Attention Stereo Matching Network(PASMnet).Our approach leverages unsupervised learning to address the scarcity of ground truth disparity in stereo matching datasets,facilitating robust training across diverse scene-specific datasets and enhancing generalization.SGPASMnet incorporates two novel components:a Cross-Scale Feature Interaction(CSFI)block and semantic feature augmentation using a pre-trained semantic segmentation model,SegFormer,seamlessly embedded into the parallax attention mechanism.The CSFI block enables effective fusion ofmulti-scale features,integrating coarse and fine details to enhance disparity estimation accuracy.Semantic features,extracted by SegFormer,enrich the parallax attention mechanism by providing high-level scene context,significantly improving performance in ambiguous regions.Our model unifies these enhancements within a cohesive architecture,comprising semantic feature extraction,an hourglass network,a semantic-guided cascaded parallax attentionmodule,outputmodule,and a disparity refinement network.Evaluations on the KITTI2015 dataset demonstrate that our unsupervised method achieves a lower error rate compared to the original PASMnet,highlighting the effectiveness of our enhancements in handling complex scenes.By harnessing unsupervised learning without ground truth disparity needed,SGPASMnet offers a scalable and robust solution for accurate stereo matching,with superior generalization across varied real-world applications.展开更多
Wind turbine blade defect detection faces persistent challenges in separating small,low-contrast surface faults from complex backgrounds while maintaining reliability under variable illumination and viewpoints.Conven-...Wind turbine blade defect detection faces persistent challenges in separating small,low-contrast surface faults from complex backgrounds while maintaining reliability under variable illumination and viewpoints.Conven-tional image-processing pipelines struggle with scalability and robustness,and recent deep learning methods remain sensitive to class imbalance and acquisition variability.This paper introduces TurbineBladeDetNet,a convolutional architecture combining dual-attention mechanisms with multi-path feature extraction for detecting five distinct blade fault types.Our approach employs both channel-wise and spatial attention modules alongside an Albumentations-driven augmentation strategy to handle dataset imbalance and capture condition variability.The model achieves 97.14%accuracy,98.65%precision,and 98.68%recall,yielding a 98.66%F1-score with 0.0110 s inference time.Class-specific analysis shows uniformly high sensitivity and specificity;lightning damage reaches 99.80%for sensitivity,precision,and F1-score,and crack achieves perfect precision and specificity with a 98.94%F1-score.Comparative evaluation against recent wind-turbine inspection approaches indicates higher performance in both accuracy and F1-score.The resulting balance of sensitivity and specificity limits both missed defects and false alarms,supporting reliable deployment in routine unmanned aerial vehicle(UAV)inspection.展开更多
Lithology identificationwhile drilling technology can obtain rock information in real-time.However,traditional lithology identificationmodels often face limitations in feature extraction and adaptability to complex ge...Lithology identificationwhile drilling technology can obtain rock information in real-time.However,traditional lithology identificationmodels often face limitations in feature extraction and adaptability to complex geological conditions,limiting their accuracy in challenging environments.To address these challenges,a deep learning model for lithology identificationwhile drilling is proposed.The proposed model introduces a dual attention mechanism in the long short-term memory(LSTM)network,effectively enhancing the ability to capture spatial and channel dimension information.Subsequently,the crayfishoptimization algorithm(COA)is applied to optimize the model network structure,thereby enhancing its lithology identificationcapability.Laboratory test results demonstrate that the proposed model achieves 97.15%accuracy on the testing set,significantlyoutperforming the traditional support vector machine(SVM)method(81.77%).Field tests under actual drilling conditions demonstrate an average accuracy of 91.96%for the proposed model,representing a 14.31%improvement over the LSTM model alone.The proposed model demonstrates robust adaptability and generalization ability across diverse operational scenarios.This research offers reliable technical support for lithology identification while drilling.展开更多
Graph Federated Learning(GFL)has shown great potential in privacy protection and distributed intelligence through distributed collaborative training of graph-structured data without sharing raw information.However,exi...Graph Federated Learning(GFL)has shown great potential in privacy protection and distributed intelligence through distributed collaborative training of graph-structured data without sharing raw information.However,existing GFL approaches often lack the capability for comprehensive feature extraction and adaptive optimization,particularly in non-independent and identically distributed(NON-IID)scenarios where balancing global structural understanding and local node-level detail remains a challenge.To this end,this paper proposes a novel framework called GFL-SAR(Graph Federated Collaborative Learning Framework Based on Structural Amplification and Attention Refinement),which enhances the representation learning capability of graph data through a dual-branch collaborative design.Specifically,we propose the Structural Insight Amplifier(SIA),which utilizes an improved Graph Convolutional Network(GCN)to strengthen structural awareness and improve modeling of topological patterns.In parallel,we propose the Attentive Relational Refiner(ARR),which employs an enhanced Graph Attention Network(GAT)to perform fine-grained modeling of node relationships and neighborhood features,thereby improving the expressiveness of local interactions and preserving critical contextual information.GFL-SAR effectively integrates multi-scale features from every branch via feature fusion and federated optimization,thereby addressing existing GFL limitations in structural modeling and feature representation.Experiments on standard benchmark datasets including Cora,Citeseer,Polblogs,and Cora_ML demonstrate that GFL-SAR achieves superior performance in classification accuracy,convergence speed,and robustness compared to existing methods,confirming its effectiveness and generalizability in GFL tasks.展开更多
Accurate wind speed prediction is crucial for stabilizing power grids with high wind energy penetration.This study presents a novel machine learning model that integrates clustering,deep learning,and transfer learning...Accurate wind speed prediction is crucial for stabilizing power grids with high wind energy penetration.This study presents a novel machine learning model that integrates clustering,deep learning,and transfer learning to mitigate accuracy degradation in 24-h forecasting.Initially,an optimized DB-SCAN(Density-Based Spatial Clustering of Applications with Noise)algorithm clusters wind fields based on wind direction,probability density,and spectral features,enhancing physical interpretability and reducing training complexity.Subsequently,a ResNet(Residual Network)extracts multi-scale patterns from decomposed wind signals,while transfer learning adapts the backbone network across clusters,cutting training time by over 90%.Finally,a CBAM(Convolutional Block Attention Module)attention mechanism is employed to prioritize features for LSTM-based prediction.Tested on the 2015 Jena wind speed dataset,the model demonstrates superior accuracy and robustness compared to state-of-the-art baselines.Key innovations include:(a)Physics-informed clustering for interpretable wind regime classification;(b)Transfer learning with deep feature extraction,preserving accuracy while minimizing training time;and(c)On the 2016 Jena wind speed dataset,the model achieves MAPE(Mean Absolute Percentage Error)values of 16.82%and 18.02%for the Weibull-shaped and Gaussian-shaped wind speed clusters,respectively,demonstrating the model’s robust generalization capacity.This framework offers an efficient and effective solution for long-term wind forecasting.展开更多
Clock synchronization has important applications in multi-agent collaboration(such as drone light shows,intelligent transportation systems,and game AI),group decision-making,and emergency rescue operations.Synchroniza...Clock synchronization has important applications in multi-agent collaboration(such as drone light shows,intelligent transportation systems,and game AI),group decision-making,and emergency rescue operations.Synchronization method based on pulse-coupled oscillators(PCOs)provides an effective solution for clock synchronization in wireless networks.However,the existing clock synchronization algorithms in multi-agent ad hoc networks are difficult to meet the requirements of high precision and high stability of synchronization clock in group cooperation.Hence,this paper constructs a network model,named DAUNet(unsupervised neural network based on dual attention),to enhance clock synchronization accuracy in multi-agent wireless ad hoc networks.Specifically,we design an unsupervised distributed neural network framework as the backbone,building upon classical PCO-based synchronization methods.This framework resolves issues such as prolonged time synchronization message exchange between nodes,difficulties in centralized node coordination,and challenges in distributed training.Furthermore,we introduce a dual-attention mechanism as the core module of DAUNet.By integrating a Multi-Head Attention module and a Gated Attention module,the model significantly improves information extraction capabilities while reducing computational complexity,effectively mitigating synchronization inaccuracies and instability in multi-agent ad hoc networks.To evaluate the effectiveness of the proposed model,comparative experiments and ablation studies were conducted against classical methods and existing deep learning models.The research results show that,compared with the deep learning networks based on DASA and LSTM,DAUNet can reduce the mean normalized phase difference(NPD)by 1 to 2 orders of magnitude.Compared with the attention models based on additive attention and self-attention mechanisms,the performance of DAUNet has improved by more than ten times.This study demonstrates DAUNet’s potential in advancing multi-agent ad hoc networking technologies.展开更多
Defect detection in printed circuit boards(PCB)remains challenging due to the difficulty of identifying small-scale defects,the inefficiency of conventional approaches,and the interference from complex backgrounds.To ...Defect detection in printed circuit boards(PCB)remains challenging due to the difficulty of identifying small-scale defects,the inefficiency of conventional approaches,and the interference from complex backgrounds.To address these issues,this paper proposes SIM-Net,an enhanced detection framework derived from YOLOv11.The model integrates SPDConv to preserve fine-grained features for small object detection,introduces a novel convolutional partial attention module(C2PAM)to suppress redundant background information and highlight salient regions,and employs a multi-scale fusion network(MFN)with a multi-grain contextual module(MGCT)to strengthen contextual representation and accelerate inference.Experimental evaluations demonstrate that SIM-Net achieves 92.4%mAP,92%accuracy,and 89.4%recall with an inference speed of 75.1 FPS,outperforming existing state-of-the-art methods.These results confirm the robustness and real-time applicability of SIM-Net for PCB defect inspection.展开更多
In the context of modern software development characterized by increasing complexity and compressed development cycles,traditional static vulnerability detection methods face prominent challenges including high false ...In the context of modern software development characterized by increasing complexity and compressed development cycles,traditional static vulnerability detection methods face prominent challenges including high false positive rates and missed detections of complex logic due to their over-reliance on rule templates.This paper proposes a Syntax-Aware Hierarchical Attention Network(SAHAN)model,which achieves high-precision vulnerability detection through grammar-rule-driven multi-granularity code slicing and hierarchical semantic fusion mechanisms.The SAHAN model first generates Syntax Independent Units(SIUs),which slices the code based on Abstract Syntax Tree(AST)and predefined grammar rules,retaining vulnerability-sensitive contexts.Following this,through a hierarchical attention mechanism,the local syntax-aware layer encodes fine-grained patterns within SIUs,while the global semantic correlation layer captures vulnerability chains across SIUs,achieving synergistic modeling of syntax and semantics.Experiments show that on benchmark datasets like QEMU,SAHAN significantly improves detection performance by 4.8%to 13.1%on average compared to baseline models such as Devign and VulDeePecker.展开更多
Marine forecasting is critical for navigation safety and disaster prevention.However,traditional ocean numerical forecasting models are often limited by substantial errors and inadequate capture of temporal-spatial fe...Marine forecasting is critical for navigation safety and disaster prevention.However,traditional ocean numerical forecasting models are often limited by substantial errors and inadequate capture of temporal-spatial features.To address the limitations,the paper proposes a TimeXer-based numerical forecast correction model optimized by an exogenous-variable attention mechanism.The model treats target forecast values as internal variables,and incorporates historical temporal-spatial data and seven-day numerical forecast results from traditional models as external variables based on the embedding strategy of TimeXer.Using a self-attention structure,the model captures correlations between exogenous variables and target sequences,explores intrinsic multi-dimensional relationships,and subsequently corrects endogenous variables with the mined exogenous features.The model’s performance is evaluated using metrics including MSE(Mean Squared Error),MAE(Mean Absolute Error),RMSE(Root Mean Square Error),MAPE(Mean Absolute Percentage Error),MSPE(Mean Square Percentage Error),and computational time,with TimeXer and PatchTST models serving as benchmarks.Experiment results show that the proposed model achieves lower errors and higher correction accuracy for both one-day and seven-day forecasts.展开更多
Accurate prediction of remaining useful life serves as a reliable basis for maintenance strategies,effectively reducing both the frequency of failures and associated costs.As a core component of PHM,RUL prediction pla...Accurate prediction of remaining useful life serves as a reliable basis for maintenance strategies,effectively reducing both the frequency of failures and associated costs.As a core component of PHM,RUL prediction plays a crucial role in preventing equipment failures and optimizing maintenance decision-making.However,deep learning models often falter when processing raw,noisy temporal signals,fail to quantify prediction uncertainty,and face challenges in effectively capturing the nonlinear dynamics of equipment degradation.To address these issues,this study proposes a novel deep learning framework.First,a newbidirectional long short-termmemory network integrated with an attention mechanism is designed to enhance temporal feature extraction with improved noise robustness.Second,a probabilistic prediction framework based on kernel density estimation is constructed,incorporating residual connections and stochastic regularization to achieve precise RUL estimation.Finally,extensive experiments on the C-MAPSS dataset demonstrate that our method achieves competitive performance in terms of RMSE and Score metrics compared to state-of-the-artmodels.More importantly,the probabilistic output provides a quantifiablemeasure of prediction confidence,which is crucial for risk-informed maintenance planning,enabling managers to optimize maintenance strategies based on a quantifiable understanding of failure risk.展开更多
Foreign body classification on coal conveyor belts is a critical component of intelligent coal mining systems.Previous approaches have primarily utilized convolutional neural networks(CNNs)to effectively integrate spa...Foreign body classification on coal conveyor belts is a critical component of intelligent coal mining systems.Previous approaches have primarily utilized convolutional neural networks(CNNs)to effectively integrate spatial and semantic information.However,the performance of CNN-based methods remains limited in classification accuracy,primarily due to insufficient exploration of local image characteristics.Unlike CNNs,Vision Transformer(ViT)captures discriminative features by modeling relationships between local image patches.However,such methods typically require a large number of training samples to perform effectively.In the context of foreign body classification on coal conveyor belts,the limited availability of training samples hinders the full exploitation of Vision Transformer’s(ViT)capabilities.To address this issue,we propose an efficient approach,termed Key Part-level Attention Vision Transformer(KPA-ViT),which incorporates key local information into the transformer architecture to enrich the training information.It comprises three main components:a key-point detection module,a key local mining module,and an attention module.To extract key local regions,a key-point detection strategy is first employed to identify the positions of key points.Subsequently,the key local mining module extracts the relevant local features based on these detected points.Finally,an attention module composed of self-attention and cross-attention blocks is introduced to integrate global and key part-level information,thereby enhancing the model’s ability to learn discriminative features.Compared to recent transformer-based frameworks—such as ViT,Swin-Transformer,and EfficientViT—the proposed KPA-ViT achieves performance improvements of 9.3%,6.6%,and 2.8%,respectively,on the CUMT-BelT dataset,demonstrating its effectiveness.展开更多
基金supported by the National Natural Science Foundation of China under Grant No.12204062the Natural Science Foundation of Shandong Province under Grant No.ZR2022MF330。
文摘To enhance speech emotion recognition capability,this study constructs a speech emotion recognition model integrating the adaptive acoustic mixup(AAM)and improved coordinate and shuffle attention(ICASA)methods.The AAM method optimizes data augmentation by combining a sample selection strategy and dynamic interpolation coefficients,thus enabling information fusion of speech data with different emotions at the acoustic level.The ICASA method enhances feature extraction capability through dynamic fusion of the improved coordinate attention(ICA)and shuffle attention(SA)techniques.The ICA technique reduces computational overhead by employing depth-separable convolution and an h-swish activation function and captures long-range dependencies of multi-scale time-frequency features using the attention weights.The SA technique promotes feature interaction through channel shuffling,which helps the model learn richer and more discriminative emotional features.Experimental results demonstrate that,compared to the baseline model,the proposed model improves the weighted accuracy by 5.42%and 4.54%,and the unweighted accuracy by 3.37%and 3.85%on the IEMOCAP and RAVDESS datasets,respectively.These improvements were confirmed to be statistically significant by independent samples t-tests,further supporting the practical reliability and applicability of the proposed model in real-world emotion-aware speech systems.
基金supported by Institute of Information&Communications Technology Planning&Evaluation(IITP)under the Metaverse Support Program to Nurture the Best Talents(IITP-2024-RS-2023-00254529)grant funded by the Korea government(MSIT).
文摘Brain tumors require precise segmentation for diagnosis and treatment plans due to their complex morphology and heterogeneous characteristics.While MRI-based automatic brain tumor segmentation technology reduces the burden on medical staff and provides quantitative information,existing methodologies and recent models still struggle to accurately capture and classify the fine boundaries and diverse morphologies of tumors.In order to address these challenges and maximize the performance of brain tumor segmentation,this research introduces a novel SwinUNETR-based model by integrating a new decoder block,the Hierarchical Channel-wise Attention Decoder(HCAD),into a powerful SwinUNETR encoder.The HCAD decoder block utilizes hierarchical features and channelspecific attention mechanisms to further fuse information at different scales transmitted from the encoder and preserve spatial details throughout the reconstruction phase.Rigorous evaluations on the recent BraTS GLI datasets demonstrate that the proposed SwinHCAD model achieved superior and improved segmentation accuracy on both the Dice score and HD95 metrics across all tumor subregions(WT,TC,and ET)compared to baseline models.In particular,the rationale and contribution of the model design were clarified through ablation studies to verify the effectiveness of the proposed HCAD decoder block.The results of this study are expected to greatly contribute to enhancing the efficiency of clinical diagnosis and treatment planning by increasing the precision of automated brain tumor segmentation.
文摘Reliable traffic flow prediction is crucial for mitigating urban congestion.This paper proposes Attentionbased spatiotemporal Interactive Dynamic Graph Convolutional Network(AIDGCN),a novel architecture integrating Interactive Dynamic Graph Convolution Network(IDGCN)with Temporal Multi-Head Trend-Aware Attention.Its core innovation lies in IDGCN,which uniquely splits sequences into symmetric intervals for interactive feature sharing via dynamic graphs,and a novel attention mechanism incorporating convolutional operations to capture essential local traffic trends—addressing a critical gap in standard attention for continuous data.For 15-and 60-min forecasting on METR-LA,AIDGCN achieves MAEs of 0.75%and 0.39%,and RMSEs of 1.32%and 0.14%,respectively.In the 60-min long-term forecasting of the PEMS-BAY dataset,the AIDGCN out-performs the MRA-BGCN method by 6.28%,4.93%,and 7.17%in terms of MAE,RMSE,and MAPE,respectively.Experimental results demonstrate the superiority of our pro-posed model over state-of-the-art methods.
文摘随着光伏发电在全球能源体系中占比不断提升,超短期光伏发电量预测对电力系统调度与安全运行至关重要。然而,光伏发电量受多因素影响,具有显著随机性与波动性。为此,提出了一种基于TCN-BiLSTM-Attention模型的超短期光伏发电量预测方法。首先通过皮尔逊相关分析筛选关键特征,并利用孤立森林算法检测异常值,结合线性插值法和标准化完成数据预处理。随后,通过时间卷积网络(Temporal Convolutional Network,TCN)提取时序特征,再利用双向长短期记忆网络(Bidirectional Long Short-Term Memory,BiLSTM)网络捕获前后向时间依赖关系,并在输出端引入注意力机制聚焦关键时间步特征。最后,在Desert Knowledge Australia Solar Centre(DKASC)数据集上的对比实验表明,与传统LSTM、BiLSTM模型相比,提出的TCN-BiLSTM-Attention模型在预测精度、稳定性等方面均表现出一定优势。
文摘Salient object detection(SOD)models struggle to simultaneously preserve global structure,maintain sharp object boundaries,and sustain computational efficiency in complex scenes.In this study,we propose SPSALNet,a task-driven two-stage(macro–micro)architecture that restructures the SOD process around superpixel representations.In the proposed approach,a“split-and-enhance”principle,introduced to our knowledge for the first time in the SOD literature,hierarchically classifies superpixels and then applies targeted refinement only to ambiguous or error-prone regions.At the macro stage,the image is partitioned into content-adaptive superpixel regions,and each superpixel is represented by a high-dimensional region-level feature vector.These representations define a regional decomposition problem in which superpixels are assigned to three classes:background,object interior,and transition regions.Superpixel tokens interact with a global feature vector from a deep network backbone through a cross-attention module and are projected into an enriched embedding space that jointly encodes local topology and global context.At the micro stage,the model employs a U-Net-based refinement process that allocates computational resources only to ambiguous transition regions.The image and distance–similarity maps derived from superpixels are processed through a dual-encoder pathway.Subsequently,channel-aware fusion blocks adaptively combine information from these two sources,producing sharper and more stable object boundaries.Experimental results show that SPSALNet achieves high accuracy with lower computational cost compared to recent competing methods.On the PASCAL-S and DUT-OMRON datasets,SPSALNet exhibits a clear performance advantage across all key metrics,and it ranks first on accuracy-oriented measures on HKU-IS.On the challenging DUT-OMRON benchmark,SPSALNet reaches a MAE of 0.034.Across all datasets,it preserves object boundaries and regional structure in a stable and competitive manner.
文摘In daily life,keyword spotting plays an important role in human-computer interaction.However,noise often interferes with the extraction of time-frequency information,and achieving both computational efficiency and recognition accuracy on resource-constrained devices such as mobile terminals remains a major challenge.To address this,we propose a novel time-frequency dual-branch parallel residual network,which integrates a Dual-Branch Broadcast Residual module and a Time-Frequency Coordinate Attention module.The time-domain and frequency-domain branches are designed in parallel to independently extract temporal and spectral features,effectively avoiding the potential information loss caused by serial stacking,while enhancing information flow and multi-scale feature fusion.In terms of training strategy,a curriculum learning approach is introduced to progressively improve model robustness fromeasy to difficult tasks.Experimental results demonstrate that the proposed method consistently outperforms existing lightweight models under various signal-to-noise ratio(SNR)conditions,achieving superior far-field recognition performance on the Google Speech Commands V2 dataset.Notably,the model maintains stable performance even in low-SNR environments such as–10 dB,and generalizes well to unseen SNR conditions during training,validating its robustness to novel noise scenarios.Furthermore,the proposed model exhibits significantly fewer parameters,making it highly suitable for deployment on resource-limited devices.Overall,the model achieves a favorable balance between performance and parameter efficiency,demonstrating strong potential for practical applications.
基金funded by the National Natural Science Foundation of China,grant number 62262045the Fundamental Research Funds for the Central Universities,grant number 2023CDJYGRH-YB11the Open Funding of SUGON Industrial Control and Security Center,grant number CUIT-SICSC-2025-03.
文摘Autonomous vehicles rely heavily on accurate and efficient scene segmentation for safe navigation and efficient operations.Traditional Bird’s Eye View(BEV)methods on semantic scene segmentation,which leverage multimodal sensor fusion,often struggle with noisy data and demand high-performance GPUs,leading to sensor misalignment and performance degradation.This paper introduces an Enhanced Channel Attention BEV(ECABEV),a novel approach designed to address the challenges under insufficient GPU memory conditions.ECABEV integrates camera and radar data through a de-noise enhanced channel attention mechanism,which utilizes global average and max pooling to effectively filter out noise while preserving discriminative features.Furthermore,an improved fusion approach is proposed to efficiently merge categorical data across modalities.To reduce computational overhead,a bilinear interpolation layer normalizationmethod is devised to ensure spatial feature fidelity.Moreover,a scalable crossentropy loss function is further designed to handle the imbalanced classes with less computational efficiency sacrifice.Extensive experiments on the nuScenes dataset demonstrate that ECABEV achieves state-of-the-art performance with an IoU of 39.961,using a lightweight ViT-B/14 backbone and lower resolution(224×224).Our approach highlights its cost-effectiveness and practical applicability,even on low-end devices.The code is publicly available at:https://github.com/YYF-CQU/ECABEV.git.
基金Tianmin Tianyuan Boutique Vegetable Industry Technology Service Station(Grant No.2024120011003081)Development of Environmental Monitoring and Traceability System for Wuqing Agricultural Production Areas(Grant No.2024120011001866)。
文摘Tomato is a major economic crop worldwide,and diseases on tomato leaves can significantly reduce both yield and quality.Traditional manual inspection is inefficient and highly subjective,making it difficult to meet the requirements of early disease identification in complex natural environments.To address this issue,this study proposes an improved YOLO11-based model,YOLO-SPDNet(Scale Sequence Fusion,Position-Channel Attention,and Dual Enhancement Network).The model integrates the SEAM(Self-Ensembling Attention Mechanism)semantic enhancement module,the MLCA(Mixed Local Channel Attention)lightweight attention mechanism,and the SPA(Scale-Position-Detail Awareness)module composed of SSFF(Scale Sequence Feature Fusion),TFE(Triple Feature Encoding),and CPAM(Channel and Position Attention Mechanism).These enhancements strengthen fine-grained lesion detection while maintaining model lightweightness.Experimental results show that YOLO-SPDNet achieves an accuracy of 91.8%,a recall of 86.5%,and an mAP@0.5 of 90.6%on the test set,with a computational complexity of 12.5 GFLOPs.Furthermore,the model reaches a real-time inference speed of 987 FPS,making it suitable for deployment on mobile agricultural terminals and online monitoring systems.Comparative analysis and ablation studies further validate the reliability and practical applicability of the proposed model in complex natural scenes.
基金supported by the National Natural Science Foundation of China,No.62301497the Science and Technology Research Program of Henan,No.252102211024the Key Research and Development Program of Henan,No.231111212000.
文摘Stereo matching is a pivotal task in computer vision,enabling precise depth estimation from stereo image pairs,yet it encounters challenges in regions with reflections,repetitive textures,or fine structures.In this paper,we propose a Semantic-Guided Parallax Attention Stereo Matching Network(SGPASMnet)that can be trained in unsupervised manner,building upon the Parallax Attention Stereo Matching Network(PASMnet).Our approach leverages unsupervised learning to address the scarcity of ground truth disparity in stereo matching datasets,facilitating robust training across diverse scene-specific datasets and enhancing generalization.SGPASMnet incorporates two novel components:a Cross-Scale Feature Interaction(CSFI)block and semantic feature augmentation using a pre-trained semantic segmentation model,SegFormer,seamlessly embedded into the parallax attention mechanism.The CSFI block enables effective fusion ofmulti-scale features,integrating coarse and fine details to enhance disparity estimation accuracy.Semantic features,extracted by SegFormer,enrich the parallax attention mechanism by providing high-level scene context,significantly improving performance in ambiguous regions.Our model unifies these enhancements within a cohesive architecture,comprising semantic feature extraction,an hourglass network,a semantic-guided cascaded parallax attentionmodule,outputmodule,and a disparity refinement network.Evaluations on the KITTI2015 dataset demonstrate that our unsupervised method achieves a lower error rate compared to the original PASMnet,highlighting the effectiveness of our enhancements in handling complex scenes.By harnessing unsupervised learning without ground truth disparity needed,SGPASMnet offers a scalable and robust solution for accurate stereo matching,with superior generalization across varied real-world applications.
文摘Wind turbine blade defect detection faces persistent challenges in separating small,low-contrast surface faults from complex backgrounds while maintaining reliability under variable illumination and viewpoints.Conven-tional image-processing pipelines struggle with scalability and robustness,and recent deep learning methods remain sensitive to class imbalance and acquisition variability.This paper introduces TurbineBladeDetNet,a convolutional architecture combining dual-attention mechanisms with multi-path feature extraction for detecting five distinct blade fault types.Our approach employs both channel-wise and spatial attention modules alongside an Albumentations-driven augmentation strategy to handle dataset imbalance and capture condition variability.The model achieves 97.14%accuracy,98.65%precision,and 98.68%recall,yielding a 98.66%F1-score with 0.0110 s inference time.Class-specific analysis shows uniformly high sensitivity and specificity;lightning damage reaches 99.80%for sensitivity,precision,and F1-score,and crack achieves perfect precision and specificity with a 98.94%F1-score.Comparative evaluation against recent wind-turbine inspection approaches indicates higher performance in both accuracy and F1-score.The resulting balance of sensitivity and specificity limits both missed defects and false alarms,supporting reliable deployment in routine unmanned aerial vehicle(UAV)inspection.
基金supported by the National Key Research and Development Program for Young Scientists,Chin(Grant No.2021YFC2900400)the Sichuan-Chongqing Science and Technology Innovation Cooperation Program Project,China(Grant No.2024TIAD-CYKJCXX0269)the National Natural Science Foundation of China,China(Grant No.52304123).
文摘Lithology identificationwhile drilling technology can obtain rock information in real-time.However,traditional lithology identificationmodels often face limitations in feature extraction and adaptability to complex geological conditions,limiting their accuracy in challenging environments.To address these challenges,a deep learning model for lithology identificationwhile drilling is proposed.The proposed model introduces a dual attention mechanism in the long short-term memory(LSTM)network,effectively enhancing the ability to capture spatial and channel dimension information.Subsequently,the crayfishoptimization algorithm(COA)is applied to optimize the model network structure,thereby enhancing its lithology identificationcapability.Laboratory test results demonstrate that the proposed model achieves 97.15%accuracy on the testing set,significantlyoutperforming the traditional support vector machine(SVM)method(81.77%).Field tests under actual drilling conditions demonstrate an average accuracy of 91.96%for the proposed model,representing a 14.31%improvement over the LSTM model alone.The proposed model demonstrates robust adaptability and generalization ability across diverse operational scenarios.This research offers reliable technical support for lithology identification while drilling.
基金supported by National Natural Science Foundation of China(62466045)Inner Mongolia Natural Science Foundation Project(2021LHMS06003)Inner Mongolia University Basic Research Business Fee Project(114).
文摘Graph Federated Learning(GFL)has shown great potential in privacy protection and distributed intelligence through distributed collaborative training of graph-structured data without sharing raw information.However,existing GFL approaches often lack the capability for comprehensive feature extraction and adaptive optimization,particularly in non-independent and identically distributed(NON-IID)scenarios where balancing global structural understanding and local node-level detail remains a challenge.To this end,this paper proposes a novel framework called GFL-SAR(Graph Federated Collaborative Learning Framework Based on Structural Amplification and Attention Refinement),which enhances the representation learning capability of graph data through a dual-branch collaborative design.Specifically,we propose the Structural Insight Amplifier(SIA),which utilizes an improved Graph Convolutional Network(GCN)to strengthen structural awareness and improve modeling of topological patterns.In parallel,we propose the Attentive Relational Refiner(ARR),which employs an enhanced Graph Attention Network(GAT)to perform fine-grained modeling of node relationships and neighborhood features,thereby improving the expressiveness of local interactions and preserving critical contextual information.GFL-SAR effectively integrates multi-scale features from every branch via feature fusion and federated optimization,thereby addressing existing GFL limitations in structural modeling and feature representation.Experiments on standard benchmark datasets including Cora,Citeseer,Polblogs,and Cora_ML demonstrate that GFL-SAR achieves superior performance in classification accuracy,convergence speed,and robustness compared to existing methods,confirming its effectiveness and generalizability in GFL tasks.
基金funded by Science and Technology Research and Development Program Project of China Railway Group Limited(No.2023-Major-02)National Natural Science Foundation of China(Grant No.52378200)Sichuan Science and Technology Program(Grant No.2024NSFSC0017).
文摘Accurate wind speed prediction is crucial for stabilizing power grids with high wind energy penetration.This study presents a novel machine learning model that integrates clustering,deep learning,and transfer learning to mitigate accuracy degradation in 24-h forecasting.Initially,an optimized DB-SCAN(Density-Based Spatial Clustering of Applications with Noise)algorithm clusters wind fields based on wind direction,probability density,and spectral features,enhancing physical interpretability and reducing training complexity.Subsequently,a ResNet(Residual Network)extracts multi-scale patterns from decomposed wind signals,while transfer learning adapts the backbone network across clusters,cutting training time by over 90%.Finally,a CBAM(Convolutional Block Attention Module)attention mechanism is employed to prioritize features for LSTM-based prediction.Tested on the 2015 Jena wind speed dataset,the model demonstrates superior accuracy and robustness compared to state-of-the-art baselines.Key innovations include:(a)Physics-informed clustering for interpretable wind regime classification;(b)Transfer learning with deep feature extraction,preserving accuracy while minimizing training time;and(c)On the 2016 Jena wind speed dataset,the model achieves MAPE(Mean Absolute Percentage Error)values of 16.82%and 18.02%for the Weibull-shaped and Gaussian-shaped wind speed clusters,respectively,demonstrating the model’s robust generalization capacity.This framework offers an efficient and effective solution for long-term wind forecasting.
文摘Clock synchronization has important applications in multi-agent collaboration(such as drone light shows,intelligent transportation systems,and game AI),group decision-making,and emergency rescue operations.Synchronization method based on pulse-coupled oscillators(PCOs)provides an effective solution for clock synchronization in wireless networks.However,the existing clock synchronization algorithms in multi-agent ad hoc networks are difficult to meet the requirements of high precision and high stability of synchronization clock in group cooperation.Hence,this paper constructs a network model,named DAUNet(unsupervised neural network based on dual attention),to enhance clock synchronization accuracy in multi-agent wireless ad hoc networks.Specifically,we design an unsupervised distributed neural network framework as the backbone,building upon classical PCO-based synchronization methods.This framework resolves issues such as prolonged time synchronization message exchange between nodes,difficulties in centralized node coordination,and challenges in distributed training.Furthermore,we introduce a dual-attention mechanism as the core module of DAUNet.By integrating a Multi-Head Attention module and a Gated Attention module,the model significantly improves information extraction capabilities while reducing computational complexity,effectively mitigating synchronization inaccuracies and instability in multi-agent ad hoc networks.To evaluate the effectiveness of the proposed model,comparative experiments and ablation studies were conducted against classical methods and existing deep learning models.The research results show that,compared with the deep learning networks based on DASA and LSTM,DAUNet can reduce the mean normalized phase difference(NPD)by 1 to 2 orders of magnitude.Compared with the attention models based on additive attention and self-attention mechanisms,the performance of DAUNet has improved by more than ten times.This study demonstrates DAUNet’s potential in advancing multi-agent ad hoc networking technologies.
文摘Defect detection in printed circuit boards(PCB)remains challenging due to the difficulty of identifying small-scale defects,the inefficiency of conventional approaches,and the interference from complex backgrounds.To address these issues,this paper proposes SIM-Net,an enhanced detection framework derived from YOLOv11.The model integrates SPDConv to preserve fine-grained features for small object detection,introduces a novel convolutional partial attention module(C2PAM)to suppress redundant background information and highlight salient regions,and employs a multi-scale fusion network(MFN)with a multi-grain contextual module(MGCT)to strengthen contextual representation and accelerate inference.Experimental evaluations demonstrate that SIM-Net achieves 92.4%mAP,92%accuracy,and 89.4%recall with an inference speed of 75.1 FPS,outperforming existing state-of-the-art methods.These results confirm the robustness and real-time applicability of SIM-Net for PCB defect inspection.
基金supported by the research start-up funds for invited doctor of Lanzhou University of Technology under Grant 14/062402。
文摘In the context of modern software development characterized by increasing complexity and compressed development cycles,traditional static vulnerability detection methods face prominent challenges including high false positive rates and missed detections of complex logic due to their over-reliance on rule templates.This paper proposes a Syntax-Aware Hierarchical Attention Network(SAHAN)model,which achieves high-precision vulnerability detection through grammar-rule-driven multi-granularity code slicing and hierarchical semantic fusion mechanisms.The SAHAN model first generates Syntax Independent Units(SIUs),which slices the code based on Abstract Syntax Tree(AST)and predefined grammar rules,retaining vulnerability-sensitive contexts.Following this,through a hierarchical attention mechanism,the local syntax-aware layer encodes fine-grained patterns within SIUs,while the global semantic correlation layer captures vulnerability chains across SIUs,achieving synergistic modeling of syntax and semantics.Experiments show that on benchmark datasets like QEMU,SAHAN significantly improves detection performance by 4.8%to 13.1%on average compared to baseline models such as Devign and VulDeePecker.
基金supported by the National Key Research and Development Program Project(2023YFC3107804)Planning Fund Project of Humanities and Social Sciences Research of the Ministry of Education(24YJA880097)the Graduate Education Reform Project in North China University of Technology(217051360025XN095-17)。
文摘Marine forecasting is critical for navigation safety and disaster prevention.However,traditional ocean numerical forecasting models are often limited by substantial errors and inadequate capture of temporal-spatial features.To address the limitations,the paper proposes a TimeXer-based numerical forecast correction model optimized by an exogenous-variable attention mechanism.The model treats target forecast values as internal variables,and incorporates historical temporal-spatial data and seven-day numerical forecast results from traditional models as external variables based on the embedding strategy of TimeXer.Using a self-attention structure,the model captures correlations between exogenous variables and target sequences,explores intrinsic multi-dimensional relationships,and subsequently corrects endogenous variables with the mined exogenous features.The model’s performance is evaluated using metrics including MSE(Mean Squared Error),MAE(Mean Absolute Error),RMSE(Root Mean Square Error),MAPE(Mean Absolute Percentage Error),MSPE(Mean Square Percentage Error),and computational time,with TimeXer and PatchTST models serving as benchmarks.Experiment results show that the proposed model achieves lower errors and higher correction accuracy for both one-day and seven-day forecasts.
基金funded by scientific research projects under Grant JY2024B011.
文摘Accurate prediction of remaining useful life serves as a reliable basis for maintenance strategies,effectively reducing both the frequency of failures and associated costs.As a core component of PHM,RUL prediction plays a crucial role in preventing equipment failures and optimizing maintenance decision-making.However,deep learning models often falter when processing raw,noisy temporal signals,fail to quantify prediction uncertainty,and face challenges in effectively capturing the nonlinear dynamics of equipment degradation.To address these issues,this study proposes a novel deep learning framework.First,a newbidirectional long short-termmemory network integrated with an attention mechanism is designed to enhance temporal feature extraction with improved noise robustness.Second,a probabilistic prediction framework based on kernel density estimation is constructed,incorporating residual connections and stochastic regularization to achieve precise RUL estimation.Finally,extensive experiments on the C-MAPSS dataset demonstrate that our method achieves competitive performance in terms of RMSE and Score metrics compared to state-of-the-artmodels.More importantly,the probabilistic output provides a quantifiablemeasure of prediction confidence,which is crucial for risk-informed maintenance planning,enabling managers to optimize maintenance strategies based on a quantifiable understanding of failure risk.
基金funded by the National Key Research and Development Program of China(grant number 2023YFC2907600)the National Natural Science Foundation of China(grant number 52504132)Tiandi Science and Technology Co.,Ltd.Science and Technology Innovation Venture Capital Special Project(grant number 2023-TD-ZD011-004).
文摘Foreign body classification on coal conveyor belts is a critical component of intelligent coal mining systems.Previous approaches have primarily utilized convolutional neural networks(CNNs)to effectively integrate spatial and semantic information.However,the performance of CNN-based methods remains limited in classification accuracy,primarily due to insufficient exploration of local image characteristics.Unlike CNNs,Vision Transformer(ViT)captures discriminative features by modeling relationships between local image patches.However,such methods typically require a large number of training samples to perform effectively.In the context of foreign body classification on coal conveyor belts,the limited availability of training samples hinders the full exploitation of Vision Transformer’s(ViT)capabilities.To address this issue,we propose an efficient approach,termed Key Part-level Attention Vision Transformer(KPA-ViT),which incorporates key local information into the transformer architecture to enrich the training information.It comprises three main components:a key-point detection module,a key local mining module,and an attention module.To extract key local regions,a key-point detection strategy is first employed to identify the positions of key points.Subsequently,the key local mining module extracts the relevant local features based on these detected points.Finally,an attention module composed of self-attention and cross-attention blocks is introduced to integrate global and key part-level information,thereby enhancing the model’s ability to learn discriminative features.Compared to recent transformer-based frameworks—such as ViT,Swin-Transformer,and EfficientViT—the proposed KPA-ViT achieves performance improvements of 9.3%,6.6%,and 2.8%,respectively,on the CUMT-BelT dataset,demonstrating its effectiveness.
