This research centers on structural health monitoring of bridges,a critical transportation infrastructure.Owing to the cumulative action of heavy vehicle loads,environmental variations,and material aging,bridge compon...This research centers on structural health monitoring of bridges,a critical transportation infrastructure.Owing to the cumulative action of heavy vehicle loads,environmental variations,and material aging,bridge components are prone to cracks and other defects,severely compromising structural safety and service life.Traditional inspection methods relying on manual visual assessment or vehicle-mounted sensors suffer from low efficiency,strong subjectivity,and high costs,while conventional image processing techniques and early deep learning models(e.g.,UNet,Faster R-CNN)still performinadequately in complex environments(e.g.,varying illumination,noise,false cracks)due to poor perception of fine cracks andmulti-scale features,limiting practical application.To address these challenges,this paper proposes CACNN-Net(CBAM-Augmented CNN),a novel dual-encoder architecture that innovatively couples a CNN for local detail extraction with a CBAM-Transformer for global context modeling.A key contribution is the dedicated Feature FusionModule(FFM),which strategically integratesmulti-scale features and focuses attention on crack regions while suppressing irrelevant noise.Experiments on bridge crack datasets demonstrate that CACNNNet achieves a precision of 77.6%,a recall of 79.4%,and an mIoU of 62.7%.These results significantly outperform several typical models(e.g.,UNet-ResNet34,Deeplabv3),confirming their superior accuracy and robust generalization,providing a high-precision automated solution for bridge crack detection and a novel network design paradigm for structural surface defect identification in complex scenarios,while future research may integrate physical features like depth information to advance intelligent infrastructure maintenance and digital twin management.展开更多
Audio-visual scene classification(AVSC)poses a formidable challenge owing to the intricate spatial-temporal relationships exhibited by audio-visual signals,coupled with the complex spatial patterns of objects and text...Audio-visual scene classification(AVSC)poses a formidable challenge owing to the intricate spatial-temporal relationships exhibited by audio-visual signals,coupled with the complex spatial patterns of objects and textures found in visual images.The focus of recent studies has predominantly revolved around extracting features from diverse neural network structures,inadvertently neglecting the acquisition of semantically meaningful regions and crucial components within audio-visual data.The authors present a feature pyramid attention network(FPANet)for audio-visual scene understanding,which extracts semantically significant characteristics from audio-visual data.The authors’approach builds multi-scale hierarchical features of sound spectrograms and visual images using a feature pyramid representation and localises the semantically relevant regions with a feature pyramid attention module(FPAM).A dimension alignment(DA)strategy is employed to align feature maps from multiple layers,a pyramid spatial attention(PSA)to spatially locate essential regions,and a pyramid channel attention(PCA)to pinpoint significant temporal frames.Experiments on visual scene classification(VSC),audio scene classification(ASC),and AVSC tasks demonstrate that FPANet achieves performance on par with state-of-the-art(SOTA)approaches,with a 95.9 F1-score on the ADVANCE dataset and a relative improvement of 28.8%.Visualisation results show that FPANet can prioritise semantically meaningful areas in audio-visual signals.展开更多
With the rapid development of Internet of Things technology,the sharp increase in network devices and their inherent security vulnerabilities present a stark contrast,bringing unprecedented challenges to the field of ...With the rapid development of Internet of Things technology,the sharp increase in network devices and their inherent security vulnerabilities present a stark contrast,bringing unprecedented challenges to the field of network security,especially in identifying malicious attacks.However,due to the uneven distribution of network traffic data,particularly the imbalance between attack traffic and normal traffic,as well as the imbalance between minority class attacks and majority class attacks,traditional machine learning detection algorithms have significant limitations when dealing with sparse network traffic data.To effectively tackle this challenge,we have designed a lightweight intrusion detection model based on diffusion mechanisms,named Diff-IDS,with the core objective of enhancing the model’s efficiency in parsing complex network traffic features,thereby significantly improving its detection speed and training efficiency.The model begins by finely filtering network traffic features and converting them into grayscale images,while also employing image-flipping techniques for data augmentation.Subsequently,these preprocessed images are fed into a diffusion model based on the Unet architecture for training.Once the model is trained,we fix the weights of the Unet network and propose a feature enhancement algorithm based on feature masking to further boost the model’s expressiveness.Finally,we devise an end-to-end lightweight detection strategy to streamline the model,enabling efficient lightweight detection of imbalanced samples.Our method has been subjected to multiple experimental tests on renowned network intrusion detection benchmarks,including CICIDS 2017,KDD 99,and NSL-KDD.The experimental results indicate that Diff-IDS leads in terms of detection accuracy,training efficiency,and lightweight metrics compared to the current state-of-the-art models,demonstrating exceptional detection capabilities and robustness.展开更多
Feature fusion is an important technique in medical image classification that can improve diagnostic accuracy by integrating complementary information from multiple sources.Recently,Deep Learning(DL)has been widely us...Feature fusion is an important technique in medical image classification that can improve diagnostic accuracy by integrating complementary information from multiple sources.Recently,Deep Learning(DL)has been widely used in pulmonary disease diagnosis,such as pneumonia and tuberculosis.However,traditional feature fusion methods often suffer from feature disparity,information loss,redundancy,and increased complexity,hindering the further extension of DL algorithms.To solve this problem,we propose a Graph-Convolution Fusion Network with Self-Supervised Feature Alignment(Self-FAGCFN)to address the limitations of traditional feature fusion methods in deep learning-based medical image classification for respiratory diseases such as pneumonia and tuberculosis.The network integrates Convolutional Neural Networks(CNNs)for robust feature extraction from two-dimensional grid structures and Graph Convolutional Networks(GCNs)within a Graph Neural Network branch to capture features based on graph structure,focusing on significant node representations.Additionally,an Attention-Embedding Ensemble Block is included to capture critical features from GCN outputs.To ensure effective feature alignment between pre-and post-fusion stages,we introduce a feature alignment loss that minimizes disparities.Moreover,to address the limitations of proposed methods,such as inappropriate centroid discrepancies during feature alignment and class imbalance in the dataset,we develop a Feature-Centroid Fusion(FCF)strategy and a Multi-Level Feature-Centroid Update(MLFCU)algorithm,respectively.Extensive experiments on public datasets LungVision and Chest-Xray demonstrate that the Self-FAGCFN model significantly outperforms existing methods in diagnosing pneumonia and tuberculosis,highlighting its potential for practical medical applications.展开更多
Accurate ultra-short-term photovoltaic(PV)power forecasting is crucial for mitigating variations caused by PV power generation and ensuring the stable and efficient operation of power grids.To capture intricate tempor...Accurate ultra-short-term photovoltaic(PV)power forecasting is crucial for mitigating variations caused by PV power generation and ensuring the stable and efficient operation of power grids.To capture intricate temporal relationships and enhance the precision of multi-step time forecast,this paper introduces an innovative approach for ultra-short-term photovoltaic(PV)power prediction,leveraging an enhanced Temporal Convolutional Neural Network(TCN)architecture and feature modeling.First,this study introduces a method employing the Spearman coefficient for meteorological feature filtration.Integrated with three-dimensional PV panel modeling,key factors influencing PV power generation are identified and prioritized.Second,the analysis of the correlation coefficient between astronomical features and PV power prediction demonstrates the theoretical substantiation for the practicality and essentiality of incorporating astronomical features.Third,an enhanced TCN model is introduced,augmenting the original TCN structure with a projection head layer to enhance its capacity for learning and expressing nonlinear features.Meanwhile,a new rolling timing network mechanism is constructed to guarantee the segmentation prediction of future long-time output sequences.Multiple experiments demonstrate the superior performance of the proposed forecasting method compared to existing models.The accuracy of PV power prediction in the next 4 hours,devoid of meteorological conditions,increases by 20.5%.Furthermore,incorporating shortwave radiation for predictions over 4 hours,2 hours,and 1 hour enhances accuracy by 11.1%,9.1%,and 8.8%,respectively.展开更多
Recent advances in convolution neural network (CNN) have fostered the progress in object recognition and semantic segmentation, which in turn has improved the performance of hyperspectral image (HSI) classification. N...Recent advances in convolution neural network (CNN) have fostered the progress in object recognition and semantic segmentation, which in turn has improved the performance of hyperspectral image (HSI) classification. Nevertheless, the difficulty of high dimensional feature extraction and the shortage of small training samples seriously hinder the future development of HSI classification. In this paper, we propose a novel algorithm for HSI classification based on three-dimensional (3D) CNN and a feature pyramid network (FPN), called 3D-FPN. The framework contains a principle component analysis, a feature extraction structure and a logistic regression. Specifically, the FPN built with 3D convolutions not only retains the advantages of 3D convolution to fully extract the spectral-spatial feature maps, but also concentrates on more detailed information and performs multi-scale feature fusion. This method avoids the excessive complexity of the model and is suitable for small sample hyperspectral classification with varying categories and spatial resolutions. In order to test the performance of our proposed 3D-FPN method, rigorous experimental analysis was performed on three public hyperspectral data sets and hyperspectral data of GF-5 satellite. Quantitative and qualitative results indicated that our proposed method attained the best performance among other current state-of-the-art end-to-end deep learning-based methods.展开更多
Dementias such as Alzheimer disease(AD)and mild cognitive impairment(MCI)lead to problems with memory,language,and daily activities resulting from damage to neurons in the brain.Given the irreversibility of this neuro...Dementias such as Alzheimer disease(AD)and mild cognitive impairment(MCI)lead to problems with memory,language,and daily activities resulting from damage to neurons in the brain.Given the irreversibility of this neuronal damage,it is crucial to find a biomarker to distinguish individuals with these diseases from healthy people.In this study,we construct a brain function network based on electroencephalography data to study changes in AD and MCI patients.Using a graph-theoretical approach,we examine connectivity features and explore their contributions to dementia recognition at edge,node,and network levels.We find that connectivity is reduced in AD and MCI patients compared with healthy controls.We also find that the edge-level features give the best performance when machine learning models are used to recognize dementia.The results of feature selection identify the top 50 ranked edge-level features constituting an optimal subset,which is mainly connected with the frontal nodes.A threshold analysis reveals that the performance of edge-level features is more sensitive to the threshold for the connection strength than that of node-and network-level features.In addition,edge-level features with a threshold of 0 provide the most effective dementia recognition.The K-nearest neighbors(KNN)machine learning model achieves the highest accuracy of 0.978 with the optimal subset when the threshold is 0.Visualization of edge-level features suggests that there are more long connections linking the frontal region with the occipital and parietal regions in AD and MCI patients compared with healthy controls.Our codes are publicly available at https://github.com/Debbie-85/eeg-connectivity.展开更多
Pest detection techniques are helpful in reducing the frequency and scale of pest outbreaks;however,their application in the actual agricultural production process is still challenging owing to the problems of intersp...Pest detection techniques are helpful in reducing the frequency and scale of pest outbreaks;however,their application in the actual agricultural production process is still challenging owing to the problems of interspecies similarity,multi-scale,and background complexity of pests.To address these problems,this study proposes an FD-YOLO pest target detection model.The FD-YOLO model uses a Fully Connected Feature Pyramid Network(FC-FPN)instead of a PANet in the neck,which can adaptively fuse multi-scale information so that the model can retain small-scale target features in the deep layer,enhance large-scale target features in the shallow layer,and enhance the multiplexing of effective features.A dual self-attention module(DSA)is then embedded in the C3 module of the neck,which captures the dependencies between the information in both spatial and channel dimensions,effectively enhancing global features.We selected 16 types of pests that widely damage field crops in the IP102 pest dataset,which were used as our dataset after data supplementation and enhancement.The experimental results showed that FD-YOLO’s mAP@0.5 improved by 6.8%compared to YOLOv5,reaching 82.6%and 19.1%–5%better than other state-of-the-art models.This method provides an effective new approach for detecting similar or multiscale pests in field crops.展开更多
In recent years,audio pattern recognition has emerged as a key area of research,driven by its applications in human-computer interaction,robotics,and healthcare.Traditional methods,which rely heavily on handcrafted fe...In recent years,audio pattern recognition has emerged as a key area of research,driven by its applications in human-computer interaction,robotics,and healthcare.Traditional methods,which rely heavily on handcrafted features such asMel filters,often suffer frominformation loss and limited feature representation capabilities.To address these limitations,this study proposes an innovative end-to-end audio pattern recognition framework that directly processes raw audio signals,preserving original information and extracting effective classification features.The proposed framework utilizes a dual-branch architecture:a global refinement module that retains channel and temporal details and a multi-scale embedding module that captures high-level semantic information.Additionally,a guided fusion module integrates complementary features from both branches,ensuring a comprehensive representation of audio data.Specifically,the multi-scale audio context embedding module is designed to effectively extract spatiotemporal dependencies,while the global refinement module aggregates multi-scale channel and temporal cues for enhanced modeling.The guided fusion module leverages these features to achieve efficient integration of complementary information,resulting in improved classification accuracy.Experimental results demonstrate the model’s superior performance on multiple datasets,including ESC-50,UrbanSound8K,RAVDESS,and CREMA-D,with classification accuracies of 93.25%,90.91%,92.36%,and 70.50%,respectively.These results highlight the robustness and effectiveness of the proposed framework,which significantly outperforms existing approaches.By addressing critical challenges such as information loss and limited feature representation,thiswork provides newinsights and methodologies for advancing audio classification and multimodal interaction systems.展开更多
A wavelet-based local and global feature fusion network(LAGN)is proposed for low-light image enhancement,aiming to enhance image details and restore colors in dark areas.This study focuses on addressing three key issu...A wavelet-based local and global feature fusion network(LAGN)is proposed for low-light image enhancement,aiming to enhance image details and restore colors in dark areas.This study focuses on addressing three key issues in low-light image enhancement:Enhancing low-light images using LAGN to preserve image details and colors;extracting image edge information via wavelet transform to enhance image details;and extracting local and global features of images through convolutional neural networks and Transformer to improve image contrast.Comparisons with state-of-the-art methods on two datasets verify that LAGN achieves the best performance in terms of details,brightness,and contrast.展开更多
Residual neural network (ResNet) is a powerful neural network architecture that has proven to be excellent in extracting spatial and channel-wise information of images. ResNet employs a residual learning strategy that...Residual neural network (ResNet) is a powerful neural network architecture that has proven to be excellent in extracting spatial and channel-wise information of images. ResNet employs a residual learning strategy that maps inputs directly to outputs, making it less difficult to optimize. In this paper, we incorporate differential information into the original residual block to improve the representative ability of the ResNet, allowing the modified network to capture more complex and metaphysical features. The proposed DFNet preserves the features after each convolutional operation in the residual block, and combines the feature maps of different levels of abstraction through the differential information. To verify the effectiveness of DFNet on image recognition, we select six distinct classification datasets. The experimental results show that our proposed DFNet has better performance and generalization ability than other state-of-the-art variants of ResNet in terms of classification accuracy and other statistical analysis.展开更多
The accurate state of health(SOH)estimation of lithium-ion batteries is crucial for efficient,healthy,and safe operation of battery systems.Extracting meaningful aging information from highly stochastic and noisy data...The accurate state of health(SOH)estimation of lithium-ion batteries is crucial for efficient,healthy,and safe operation of battery systems.Extracting meaningful aging information from highly stochastic and noisy data segments while designing SOH estimation algorithms that efficiently handle the large-scale computational demands of cloud-based battery management systems presents a substantial challenge.In this work,we propose a quantum convolutional neural network(QCNN)model designed for accurate,robust,and generalizable SOH estimation with minimal data and parameter requirements and is compatible with quantum computing cloud platforms in the Noisy Intermediate-Scale Quantum.First,we utilize data from 4 datasets comprising 272 cells,covering 5 chemical compositions,4 rated parameters,and 73operating conditions.We design 5 voltage windows as small as 0.3 V for each cell from incremental capacity peaks for stochastic SOH estimation scenarios generation.We extract 3 effective health indicators(HIs)sequences and develop an automated feature fusion method using quantum rotation gate encoding,achieving an R2of 96%.Subsequently,we design a QCNN whose convolutional layer,constructed with variational quantum circuits,comprises merely 39 parameters.Additionally,we explore the impact of training set size,using strategies,and battery materials on the model’s accuracy.Finally,the QCNN with quantum convolutional layers reduces root mean squared error by 28% and achieves an R^(2)exceeding 96% compared to other three commonly used algorithms.This work demonstrates the effectiveness of quantum encoding for automated feature fusion of HIs extracted from limited discharge data.It highlights the potential of QCNN in improving the accuracy,robustness,and generalization of SOH estimation while dealing with stochastic and noisy data with few parameters and simple structure.It also suggests a new paradigm for leveraging quantum computational power in SOH estimation.展开更多
With the birth of Software-Defined Networking(SDN),integration of both SDN and traditional architectures becomes the development trend of computer networks.Network intrusion detection faces challenges in dealing with ...With the birth of Software-Defined Networking(SDN),integration of both SDN and traditional architectures becomes the development trend of computer networks.Network intrusion detection faces challenges in dealing with complex attacks in SDN environments,thus to address the network security issues from the viewpoint of Artificial Intelligence(AI),this paper introduces the Crayfish Optimization Algorithm(COA)to the field of intrusion detection for both SDN and traditional network architectures,and based on the characteristics of the original COA,an Improved Crayfish Optimization Algorithm(ICOA)is proposed by integrating strategies of elite reverse learning,Levy flight,crowding factor and parameter modification.The ICOA is then utilized for AI-integrated feature selection of intrusion detection for both SDN and traditional network architectures,to reduce the dimensionality of the data and improve the performance of network intrusion detection.Finally,the performance evaluation is performed by testing not only the NSL-KDD dataset and the UNSW-NB 15 dataset for traditional networks but also the InSDN dataset for SDN-based networks.Experimental results show that ICOA improves the accuracy by 0.532%and 2.928%respectively compared with GWO and COA in traditional networks.In SDN networks,the accuracy of ICOA is 0.25%and 0.3%higher than COA and PSO.These findings collectively indicate that AI-integrated feature selection based on the proposed ICOA can promote network intrusion detection for both SDN and traditional architectures.展开更多
An intelligent diagnosis method based on self-adaptiveWasserstein dual generative adversarial networks and feature fusion is proposed due to problems such as insufficient sample size and incomplete fault feature extra...An intelligent diagnosis method based on self-adaptiveWasserstein dual generative adversarial networks and feature fusion is proposed due to problems such as insufficient sample size and incomplete fault feature extraction,which are commonly faced by rolling bearings and lead to low diagnostic accuracy.Initially,dual models of the Wasserstein deep convolutional generative adversarial network incorporating gradient penalty(1D-2DWDCGAN)are constructed to augment the original dataset.A self-adaptive loss threshold control training strategy is introduced,and establishing a self-adaptive balancing mechanism for stable model training.Subsequently,a diagnostic model based on multidimensional feature fusion is designed,wherein complex features from various dimensions are extracted,merging the original signal waveform features,structured features,and time-frequency features into a deep composite feature representation that encompasses multiple dimensions and scales;thus,efficient and accurate small sample fault diagnosis is facilitated.Finally,an experiment between the bearing fault dataset of CaseWestern ReserveUniversity and the fault simulation experimental platformdataset of this research group shows that this method effectively supplements the dataset and remarkably improves the diagnostic accuracy.The diagnostic accuracy after data augmentation reached 99.94%and 99.87%in two different experimental environments,respectively.In addition,robustness analysis is conducted on the diagnostic accuracy of the proposed method under different noise backgrounds,verifying its good generalization performance.展开更多
Dear Editor,D2This letter presents a node feature similarity preserving graph convolutional framework P G.Graph neural networks(GNNs)have garnered significant attention for their efficacy in learning graph representat...Dear Editor,D2This letter presents a node feature similarity preserving graph convolutional framework P G.Graph neural networks(GNNs)have garnered significant attention for their efficacy in learning graph representations across diverse real-world applications.展开更多
Recent advances in deep learning have significantly improved image deblurring;however,existing approaches still suffer from limited global context modeling,inadequate detail restoration,and poor texture or edge percep...Recent advances in deep learning have significantly improved image deblurring;however,existing approaches still suffer from limited global context modeling,inadequate detail restoration,and poor texture or edge perception,especially under complex dynamic blur.To address these challenges,we propose the Multi-Resolution Fusion Network(MRFNet),a blind multi-scale deblurring framework that integrates progressive residual connectivity for hierarchical feature fusion.The network employs a three-stage design:(1)TransformerBlocks capture long-range dependencies and reconstruct coarse global structures;(2)Nonlinear Activation Free Blocks(NAFBlocks)enhance local detail representation and mid-level feature fusion;and(3)an optimized residual subnetwork based on gated feature modulation refines texture and edge details for high-fidelity restoration.Extensive experiments demonstrate that MRFNet achieves superior performance compared to state-of-the-art methods.On GoPro,it attains 32.52 dB Peak Signal-to-Noise Ratio(PSNR)and 0.071 Learned Perceptual Image Patch Similarity(LPIPS),outperforming MIMOWNet(32.50 dB,0.075).On HIDE,it achieves 30.25 dB PSNR and 0.945 Structural Similarity Index Measure(SSIM),representing gains of+0.26 dB and+0.015 SSIM over MIMO-UNet(29.99 dB,0.930).On RealBlur-J,it reaches 28.82 dB PSNR and 0.872 SSIM,surpassing MIMO-UNet by+1.19 dB and+0.035 SSIM(27.63 dB,0.837).These results validate the effectiveness of the proposed progressive residual fusion and hybrid attention mechanisms in balancing global context understanding and local detail recovery for blind image deblurring.展开更多
The combustion and detonation processes of energetic materials exhibit remarkable complexity and ultra-fast transient characteristics.While reactive molecular dynamics has been extensively employed to investigate the ...The combustion and detonation processes of energetic materials exhibit remarkable complexity and ultra-fast transient characteristics.While reactive molecular dynamics has been extensively employed to investigate the reaction dynamics of energetic materials,its utility is often constrained to capturing only fundamental reaction events and species information,thereby limiting mechanistic investigations of complex reaction pathways.To elucidate the topological features of energetic material reaction networks and identify critical reaction pathways with high fidelity,this study presents ReacNetwork-an advanced large-scale reaction network analysis methodology that synergistically integrates complex network theory with molecular simulation techniques.Specifically,we have developed a multi-dimensional feature screening protocol based on node centrality metrics and K-shell decomposition algorithms.Takingα-Hexahydro-1,3,5-trinitro-1,3,5-triazine(α-RDX)as the subject,we successfully constructed a comprehensive high-temperature thermal decomposition reaction network consisting of 1,134 distinct chemical species and 3,626 elementary reactions.Through systematic application of community detection algorithms and global topological feature extraction techniques,we achieved effective dimensionality reduction and successfully identified the dominant reaction pathway within theα-RDX thermal decomposition network.The computational results not only validate the well-established initial reaction mechanism dominated by N-NO2 homolytic bond cleavage,but also provide unprecedented visualization ofα-RDX framework ring-opening dynamics and subsequent radical chain propagation networks.展开更多
Accurate state of health(SOH)estimation is essential for the safe and reliable operation of lithium-ion batteries.However,existing methods face significant challenges,primarily because they rely on complete charge–di...Accurate state of health(SOH)estimation is essential for the safe and reliable operation of lithium-ion batteries.However,existing methods face significant challenges,primarily because they rely on complete charge–discharge cycles and fixed-form physical constraints,which limit adaptability to different chemistries and real-world conditions.To address these issues,this study proposes an approach that extracts features from segmented state of charge(SOC)intervals and integrates them into an enhanced physics-informed neural network(PINN).Specifically,voltage data within the 25%–75%SOC range during charging are used to derive statistical,time–frequency,and mechanism-based features that capture degradation trends.A hybrid PINN-Lasso-Transformer-BiLSTM architecture is developed,where Lasso regression enables sparse feature selection,and a nonlinear empirical degradation model is embedded as a learnable physical term within a dynamically scaled composite loss.This design adaptively balances data-driven accuracy with physical consistency,thereby enhancing estimation precision,robustness,and generalization.The results show that the proposed method outperforms conventional neural networks across four battery chemistries,achieving root mean square error and mean absolute error below 1%.Notably,features from partial charging segments exhibit higher robustness than those from full cycles.Furthermore,the model maintains strong performance under high temperatures and demonstrates excellent generalization capacity in transfer learning across chemistries,temperatures,and C-rates.This work establishes a scalable and interpretable solution for accurate SOH estimation under diverse practical operating conditions.展开更多
This paper introduces a fuzzy C-means-based pooling layer for convolutional neural networks that explicitly models local uncertainty and ambiguity.Conventional pooling operations,such as max and average,apply rigid ag...This paper introduces a fuzzy C-means-based pooling layer for convolutional neural networks that explicitly models local uncertainty and ambiguity.Conventional pooling operations,such as max and average,apply rigid aggregation and often discard fine-grained boundary information.In contrast,our method computes soft membershipswithin each receptive field and aggregates cluster-wise responses throughmembership-weighted pooling,thereby preserving informative structure while reducing dimensionality.Being differentiable,the proposed layer operates as standard two-dimensional pooling.We evaluate our approach across various CNN backbones and open datasets,including CIFAR-10/100,STL-10,LFW,and ImageNette,and further probe small training set restrictions on MNIST and Fashion-MNIST.In these settings,the proposed pooling consistently improves accuracy and weighted F1 over conventional baselines,with particularly strong gains when training data are scarce.Even with less than 1%of the training set,ourmethodmaintains reliable performance,indicating improved sample efficiency and robustness to noisy or ambiguous local patterns.Overall,integrating soft memberships into the pooling operator provides a practical and generalizable inductive bias that enhances robustness and generalization in modern CNN pipelines.展开更多
With the increase of semiconductor integration density,in order to cope with the increase of wafer defect complexity and types,especially the low recognition accuracy of overlapping mixed defects and unknown wafer def...With the increase of semiconductor integration density,in order to cope with the increase of wafer defect complexity and types,especially the low recognition accuracy of overlapping mixed defects and unknown wafer defects,this study proposes a lightweight model for wafer defect detection called LightWMNet.First,using a hierarchical attention Encoder-Decoder architecture,the features of wafer defect pattern(WDP)are channel recalibrated to generate high-resolution fine-grained features and low-resolution coarse-grained features.Secondly,the backbone network incorporates two novel attention modules—feedforward spatial attention(FFSa)and feedforward channel attention(FFCa)—to amplify responses in critical defect regions and suppress noise from stochastic discrete pixels.These mechanisms synergistically enhance feature discriminability without introducing significant parametric overhead.Finally,the Dice loss function and the cross entropy loss function are combined to jointly evaluate the segmentation and classification accuracy of the model.Experimental results on the public mixed wafer defect dataset MixedWM38 show that the pixel accuracy(PA),intersection over union(IoU)and Dice coefficient of the proposed network reach 98.26%,94.83%and 97.22%,respectively.Without significantly increasing the computational complexity and size of the model,compared with the existing state-of-the-art(SOTA)model,the classification accuracy of lightWMNet in single defect,three mixed defects and four mixed defects is improved by 0.5%,0.25%and 0.89%respectively.Furthermore,we used transfer learning for the first time to evaluate the model's generalisation ability for unseen defect categories.The results showed that LightWMNet still has a certain recognition ability even in untrained wafer defects.展开更多
基金supported by the National Natural Science Foundation of China(No.52308332)the General Scientific Research Project of the Education Department of Zhejiang Province(No.Y202455824).
文摘This research centers on structural health monitoring of bridges,a critical transportation infrastructure.Owing to the cumulative action of heavy vehicle loads,environmental variations,and material aging,bridge components are prone to cracks and other defects,severely compromising structural safety and service life.Traditional inspection methods relying on manual visual assessment or vehicle-mounted sensors suffer from low efficiency,strong subjectivity,and high costs,while conventional image processing techniques and early deep learning models(e.g.,UNet,Faster R-CNN)still performinadequately in complex environments(e.g.,varying illumination,noise,false cracks)due to poor perception of fine cracks andmulti-scale features,limiting practical application.To address these challenges,this paper proposes CACNN-Net(CBAM-Augmented CNN),a novel dual-encoder architecture that innovatively couples a CNN for local detail extraction with a CBAM-Transformer for global context modeling.A key contribution is the dedicated Feature FusionModule(FFM),which strategically integratesmulti-scale features and focuses attention on crack regions while suppressing irrelevant noise.Experiments on bridge crack datasets demonstrate that CACNNNet achieves a precision of 77.6%,a recall of 79.4%,and an mIoU of 62.7%.These results significantly outperform several typical models(e.g.,UNet-ResNet34,Deeplabv3),confirming their superior accuracy and robust generalization,providing a high-precision automated solution for bridge crack detection and a novel network design paradigm for structural surface defect identification in complex scenarios,while future research may integrate physical features like depth information to advance intelligent infrastructure maintenance and digital twin management.
基金Shenzhen Institute of Artificial Intelligence and Robotics for Society,Grant/Award Number:AC01202201003-02GuangDong Basic and Applied Basic Research Foundation,Grant/Award Number:2024A1515010252Longgang District Shenzhen's“Ten Action Plan”for Supporting Innovation Projects,Grant/Award Number:LGKCSDPT2024002。
文摘Audio-visual scene classification(AVSC)poses a formidable challenge owing to the intricate spatial-temporal relationships exhibited by audio-visual signals,coupled with the complex spatial patterns of objects and textures found in visual images.The focus of recent studies has predominantly revolved around extracting features from diverse neural network structures,inadvertently neglecting the acquisition of semantically meaningful regions and crucial components within audio-visual data.The authors present a feature pyramid attention network(FPANet)for audio-visual scene understanding,which extracts semantically significant characteristics from audio-visual data.The authors’approach builds multi-scale hierarchical features of sound spectrograms and visual images using a feature pyramid representation and localises the semantically relevant regions with a feature pyramid attention module(FPAM).A dimension alignment(DA)strategy is employed to align feature maps from multiple layers,a pyramid spatial attention(PSA)to spatially locate essential regions,and a pyramid channel attention(PCA)to pinpoint significant temporal frames.Experiments on visual scene classification(VSC),audio scene classification(ASC),and AVSC tasks demonstrate that FPANet achieves performance on par with state-of-the-art(SOTA)approaches,with a 95.9 F1-score on the ADVANCE dataset and a relative improvement of 28.8%.Visualisation results show that FPANet can prioritise semantically meaningful areas in audio-visual signals.
基金supported by the Key Research and Development Program of Hainan Province(Grant Nos.ZDYF2024GXJS014,ZDYF2023GXJS163)the National Natural Science Foundation of China(NSFC)(Grant Nos.62162022,62162024)Collaborative Innovation Project of Hainan University(XTCX2022XXB02).
文摘With the rapid development of Internet of Things technology,the sharp increase in network devices and their inherent security vulnerabilities present a stark contrast,bringing unprecedented challenges to the field of network security,especially in identifying malicious attacks.However,due to the uneven distribution of network traffic data,particularly the imbalance between attack traffic and normal traffic,as well as the imbalance between minority class attacks and majority class attacks,traditional machine learning detection algorithms have significant limitations when dealing with sparse network traffic data.To effectively tackle this challenge,we have designed a lightweight intrusion detection model based on diffusion mechanisms,named Diff-IDS,with the core objective of enhancing the model’s efficiency in parsing complex network traffic features,thereby significantly improving its detection speed and training efficiency.The model begins by finely filtering network traffic features and converting them into grayscale images,while also employing image-flipping techniques for data augmentation.Subsequently,these preprocessed images are fed into a diffusion model based on the Unet architecture for training.Once the model is trained,we fix the weights of the Unet network and propose a feature enhancement algorithm based on feature masking to further boost the model’s expressiveness.Finally,we devise an end-to-end lightweight detection strategy to streamline the model,enabling efficient lightweight detection of imbalanced samples.Our method has been subjected to multiple experimental tests on renowned network intrusion detection benchmarks,including CICIDS 2017,KDD 99,and NSL-KDD.The experimental results indicate that Diff-IDS leads in terms of detection accuracy,training efficiency,and lightweight metrics compared to the current state-of-the-art models,demonstrating exceptional detection capabilities and robustness.
基金supported by the National Natural Science Foundation of China(62276092,62303167)the Postdoctoral Fellowship Program(Grade C)of China Postdoctoral Science Foundation(GZC20230707)+3 种基金the Key Science and Technology Program of Henan Province,China(242102211051,242102211042,212102310084)Key Scientiffc Research Projects of Colleges and Universities in Henan Province,China(25A520009)the China Postdoctoral Science Foundation(2024M760808)the Henan Province medical science and technology research plan joint construction project(LHGJ2024069).
文摘Feature fusion is an important technique in medical image classification that can improve diagnostic accuracy by integrating complementary information from multiple sources.Recently,Deep Learning(DL)has been widely used in pulmonary disease diagnosis,such as pneumonia and tuberculosis.However,traditional feature fusion methods often suffer from feature disparity,information loss,redundancy,and increased complexity,hindering the further extension of DL algorithms.To solve this problem,we propose a Graph-Convolution Fusion Network with Self-Supervised Feature Alignment(Self-FAGCFN)to address the limitations of traditional feature fusion methods in deep learning-based medical image classification for respiratory diseases such as pneumonia and tuberculosis.The network integrates Convolutional Neural Networks(CNNs)for robust feature extraction from two-dimensional grid structures and Graph Convolutional Networks(GCNs)within a Graph Neural Network branch to capture features based on graph structure,focusing on significant node representations.Additionally,an Attention-Embedding Ensemble Block is included to capture critical features from GCN outputs.To ensure effective feature alignment between pre-and post-fusion stages,we introduce a feature alignment loss that minimizes disparities.Moreover,to address the limitations of proposed methods,such as inappropriate centroid discrepancies during feature alignment and class imbalance in the dataset,we develop a Feature-Centroid Fusion(FCF)strategy and a Multi-Level Feature-Centroid Update(MLFCU)algorithm,respectively.Extensive experiments on public datasets LungVision and Chest-Xray demonstrate that the Self-FAGCFN model significantly outperforms existing methods in diagnosing pneumonia and tuberculosis,highlighting its potential for practical medical applications.
基金supported by National Key Research and Development Program of China(Key Techniques of Adaptive Grid Integration and Active Synchronization for Extremely High Penetration Distributed Photovoltaic Power Generation,2022YFB2402900).
文摘Accurate ultra-short-term photovoltaic(PV)power forecasting is crucial for mitigating variations caused by PV power generation and ensuring the stable and efficient operation of power grids.To capture intricate temporal relationships and enhance the precision of multi-step time forecast,this paper introduces an innovative approach for ultra-short-term photovoltaic(PV)power prediction,leveraging an enhanced Temporal Convolutional Neural Network(TCN)architecture and feature modeling.First,this study introduces a method employing the Spearman coefficient for meteorological feature filtration.Integrated with three-dimensional PV panel modeling,key factors influencing PV power generation are identified and prioritized.Second,the analysis of the correlation coefficient between astronomical features and PV power prediction demonstrates the theoretical substantiation for the practicality and essentiality of incorporating astronomical features.Third,an enhanced TCN model is introduced,augmenting the original TCN structure with a projection head layer to enhance its capacity for learning and expressing nonlinear features.Meanwhile,a new rolling timing network mechanism is constructed to guarantee the segmentation prediction of future long-time output sequences.Multiple experiments demonstrate the superior performance of the proposed forecasting method compared to existing models.The accuracy of PV power prediction in the next 4 hours,devoid of meteorological conditions,increases by 20.5%.Furthermore,incorporating shortwave radiation for predictions over 4 hours,2 hours,and 1 hour enhances accuracy by 11.1%,9.1%,and 8.8%,respectively.
基金the National Natural Science Foundation of China(No.51975374)。
文摘Recent advances in convolution neural network (CNN) have fostered the progress in object recognition and semantic segmentation, which in turn has improved the performance of hyperspectral image (HSI) classification. Nevertheless, the difficulty of high dimensional feature extraction and the shortage of small training samples seriously hinder the future development of HSI classification. In this paper, we propose a novel algorithm for HSI classification based on three-dimensional (3D) CNN and a feature pyramid network (FPN), called 3D-FPN. The framework contains a principle component analysis, a feature extraction structure and a logistic regression. Specifically, the FPN built with 3D convolutions not only retains the advantages of 3D convolution to fully extract the spectral-spatial feature maps, but also concentrates on more detailed information and performs multi-scale feature fusion. This method avoids the excessive complexity of the model and is suitable for small sample hyperspectral classification with varying categories and spatial resolutions. In order to test the performance of our proposed 3D-FPN method, rigorous experimental analysis was performed on three public hyperspectral data sets and hyperspectral data of GF-5 satellite. Quantitative and qualitative results indicated that our proposed method attained the best performance among other current state-of-the-art end-to-end deep learning-based methods.
基金supported by the National Natural Science Foundation of China(Grant Nos.62071451,62331025,and U21A20447)the National Key Research and Development Project(Grant No.2021YFC3002204)the CAMS Innovation Fund for Medical Sciences(Grant No.2019-I2M-5-019).
文摘Dementias such as Alzheimer disease(AD)and mild cognitive impairment(MCI)lead to problems with memory,language,and daily activities resulting from damage to neurons in the brain.Given the irreversibility of this neuronal damage,it is crucial to find a biomarker to distinguish individuals with these diseases from healthy people.In this study,we construct a brain function network based on electroencephalography data to study changes in AD and MCI patients.Using a graph-theoretical approach,we examine connectivity features and explore their contributions to dementia recognition at edge,node,and network levels.We find that connectivity is reduced in AD and MCI patients compared with healthy controls.We also find that the edge-level features give the best performance when machine learning models are used to recognize dementia.The results of feature selection identify the top 50 ranked edge-level features constituting an optimal subset,which is mainly connected with the frontal nodes.A threshold analysis reveals that the performance of edge-level features is more sensitive to the threshold for the connection strength than that of node-and network-level features.In addition,edge-level features with a threshold of 0 provide the most effective dementia recognition.The K-nearest neighbors(KNN)machine learning model achieves the highest accuracy of 0.978 with the optimal subset when the threshold is 0.Visualization of edge-level features suggests that there are more long connections linking the frontal region with the occipital and parietal regions in AD and MCI patients compared with healthy controls.Our codes are publicly available at https://github.com/Debbie-85/eeg-connectivity.
基金funded by Liaoning Provincial Department of Education Project,Award number JYTMS20230418.
文摘Pest detection techniques are helpful in reducing the frequency and scale of pest outbreaks;however,their application in the actual agricultural production process is still challenging owing to the problems of interspecies similarity,multi-scale,and background complexity of pests.To address these problems,this study proposes an FD-YOLO pest target detection model.The FD-YOLO model uses a Fully Connected Feature Pyramid Network(FC-FPN)instead of a PANet in the neck,which can adaptively fuse multi-scale information so that the model can retain small-scale target features in the deep layer,enhance large-scale target features in the shallow layer,and enhance the multiplexing of effective features.A dual self-attention module(DSA)is then embedded in the C3 module of the neck,which captures the dependencies between the information in both spatial and channel dimensions,effectively enhancing global features.We selected 16 types of pests that widely damage field crops in the IP102 pest dataset,which were used as our dataset after data supplementation and enhancement.The experimental results showed that FD-YOLO’s mAP@0.5 improved by 6.8%compared to YOLOv5,reaching 82.6%and 19.1%–5%better than other state-of-the-art models.This method provides an effective new approach for detecting similar or multiscale pests in field crops.
基金supported by the National Natural Science Foundation of China(62106214)the Hebei Natural Science Foundation(D2024203008)the Provincial Key Laboratory Performance Subsidy Project(22567612H).
文摘In recent years,audio pattern recognition has emerged as a key area of research,driven by its applications in human-computer interaction,robotics,and healthcare.Traditional methods,which rely heavily on handcrafted features such asMel filters,often suffer frominformation loss and limited feature representation capabilities.To address these limitations,this study proposes an innovative end-to-end audio pattern recognition framework that directly processes raw audio signals,preserving original information and extracting effective classification features.The proposed framework utilizes a dual-branch architecture:a global refinement module that retains channel and temporal details and a multi-scale embedding module that captures high-level semantic information.Additionally,a guided fusion module integrates complementary features from both branches,ensuring a comprehensive representation of audio data.Specifically,the multi-scale audio context embedding module is designed to effectively extract spatiotemporal dependencies,while the global refinement module aggregates multi-scale channel and temporal cues for enhanced modeling.The guided fusion module leverages these features to achieve efficient integration of complementary information,resulting in improved classification accuracy.Experimental results demonstrate the model’s superior performance on multiple datasets,including ESC-50,UrbanSound8K,RAVDESS,and CREMA-D,with classification accuracies of 93.25%,90.91%,92.36%,and 70.50%,respectively.These results highlight the robustness and effectiveness of the proposed framework,which significantly outperforms existing approaches.By addressing critical challenges such as information loss and limited feature representation,thiswork provides newinsights and methodologies for advancing audio classification and multimodal interaction systems.
文摘A wavelet-based local and global feature fusion network(LAGN)is proposed for low-light image enhancement,aiming to enhance image details and restore colors in dark areas.This study focuses on addressing three key issues in low-light image enhancement:Enhancing low-light images using LAGN to preserve image details and colors;extracting image edge information via wavelet transform to enhance image details;and extracting local and global features of images through convolutional neural networks and Transformer to improve image contrast.Comparisons with state-of-the-art methods on two datasets verify that LAGN achieves the best performance in terms of details,brightness,and contrast.
基金supported by the Japan Society for the Promotion of Science(JSPS)KAKENHI under Grant JP22H03643Japan Science and Technology Agency(JST)Support for Pioneering Research Initiated by the Next Generation(SPRING)under Grant JPMJSP2145JST through the Establishment of University Fellowships towards the Creation of Science Technology Innovation under Grant JPMJFS2115.
文摘Residual neural network (ResNet) is a powerful neural network architecture that has proven to be excellent in extracting spatial and channel-wise information of images. ResNet employs a residual learning strategy that maps inputs directly to outputs, making it less difficult to optimize. In this paper, we incorporate differential information into the original residual block to improve the representative ability of the ResNet, allowing the modified network to capture more complex and metaphysical features. The proposed DFNet preserves the features after each convolutional operation in the residual block, and combines the feature maps of different levels of abstraction through the differential information. To verify the effectiveness of DFNet on image recognition, we select six distinct classification datasets. The experimental results show that our proposed DFNet has better performance and generalization ability than other state-of-the-art variants of ResNet in terms of classification accuracy and other statistical analysis.
基金funded by the Research on SOC/SOH Joint Estimation Technology of Electric Vehicle Battery System State Based on Online Parameter Identification Project(2019)the National Natural Science Foundation of China(Grant No.51877120)。
文摘The accurate state of health(SOH)estimation of lithium-ion batteries is crucial for efficient,healthy,and safe operation of battery systems.Extracting meaningful aging information from highly stochastic and noisy data segments while designing SOH estimation algorithms that efficiently handle the large-scale computational demands of cloud-based battery management systems presents a substantial challenge.In this work,we propose a quantum convolutional neural network(QCNN)model designed for accurate,robust,and generalizable SOH estimation with minimal data and parameter requirements and is compatible with quantum computing cloud platforms in the Noisy Intermediate-Scale Quantum.First,we utilize data from 4 datasets comprising 272 cells,covering 5 chemical compositions,4 rated parameters,and 73operating conditions.We design 5 voltage windows as small as 0.3 V for each cell from incremental capacity peaks for stochastic SOH estimation scenarios generation.We extract 3 effective health indicators(HIs)sequences and develop an automated feature fusion method using quantum rotation gate encoding,achieving an R2of 96%.Subsequently,we design a QCNN whose convolutional layer,constructed with variational quantum circuits,comprises merely 39 parameters.Additionally,we explore the impact of training set size,using strategies,and battery materials on the model’s accuracy.Finally,the QCNN with quantum convolutional layers reduces root mean squared error by 28% and achieves an R^(2)exceeding 96% compared to other three commonly used algorithms.This work demonstrates the effectiveness of quantum encoding for automated feature fusion of HIs extracted from limited discharge data.It highlights the potential of QCNN in improving the accuracy,robustness,and generalization of SOH estimation while dealing with stochastic and noisy data with few parameters and simple structure.It also suggests a new paradigm for leveraging quantum computational power in SOH estimation.
基金supported by the National Natural Science Foundation of China under Grant 61602162the Hubei Provincial Science and Technology Plan Project under Grant 2023BCB041.
文摘With the birth of Software-Defined Networking(SDN),integration of both SDN and traditional architectures becomes the development trend of computer networks.Network intrusion detection faces challenges in dealing with complex attacks in SDN environments,thus to address the network security issues from the viewpoint of Artificial Intelligence(AI),this paper introduces the Crayfish Optimization Algorithm(COA)to the field of intrusion detection for both SDN and traditional network architectures,and based on the characteristics of the original COA,an Improved Crayfish Optimization Algorithm(ICOA)is proposed by integrating strategies of elite reverse learning,Levy flight,crowding factor and parameter modification.The ICOA is then utilized for AI-integrated feature selection of intrusion detection for both SDN and traditional network architectures,to reduce the dimensionality of the data and improve the performance of network intrusion detection.Finally,the performance evaluation is performed by testing not only the NSL-KDD dataset and the UNSW-NB 15 dataset for traditional networks but also the InSDN dataset for SDN-based networks.Experimental results show that ICOA improves the accuracy by 0.532%and 2.928%respectively compared with GWO and COA in traditional networks.In SDN networks,the accuracy of ICOA is 0.25%and 0.3%higher than COA and PSO.These findings collectively indicate that AI-integrated feature selection based on the proposed ICOA can promote network intrusion detection for both SDN and traditional architectures.
基金supported by the National Natural Science Foundation of China(Grant Nos.12272259 and 52005148).
文摘An intelligent diagnosis method based on self-adaptiveWasserstein dual generative adversarial networks and feature fusion is proposed due to problems such as insufficient sample size and incomplete fault feature extraction,which are commonly faced by rolling bearings and lead to low diagnostic accuracy.Initially,dual models of the Wasserstein deep convolutional generative adversarial network incorporating gradient penalty(1D-2DWDCGAN)are constructed to augment the original dataset.A self-adaptive loss threshold control training strategy is introduced,and establishing a self-adaptive balancing mechanism for stable model training.Subsequently,a diagnostic model based on multidimensional feature fusion is designed,wherein complex features from various dimensions are extracted,merging the original signal waveform features,structured features,and time-frequency features into a deep composite feature representation that encompasses multiple dimensions and scales;thus,efficient and accurate small sample fault diagnosis is facilitated.Finally,an experiment between the bearing fault dataset of CaseWestern ReserveUniversity and the fault simulation experimental platformdataset of this research group shows that this method effectively supplements the dataset and remarkably improves the diagnostic accuracy.The diagnostic accuracy after data augmentation reached 99.94%and 99.87%in two different experimental environments,respectively.In addition,robustness analysis is conducted on the diagnostic accuracy of the proposed method under different noise backgrounds,verifying its good generalization performance.
基金supported by the National Natural Science Foundation of China(62402399)the New Chongqing Youth Innovation Talent Project(CSTB2024NSCQ-QCXMX0035)。
文摘Dear Editor,D2This letter presents a node feature similarity preserving graph convolutional framework P G.Graph neural networks(GNNs)have garnered significant attention for their efficacy in learning graph representations across diverse real-world applications.
基金funded by Qinghai University Postgraduate Research and Practice Innovation Program of Funder,grant number 2025-GMKY-42.
文摘Recent advances in deep learning have significantly improved image deblurring;however,existing approaches still suffer from limited global context modeling,inadequate detail restoration,and poor texture or edge perception,especially under complex dynamic blur.To address these challenges,we propose the Multi-Resolution Fusion Network(MRFNet),a blind multi-scale deblurring framework that integrates progressive residual connectivity for hierarchical feature fusion.The network employs a three-stage design:(1)TransformerBlocks capture long-range dependencies and reconstruct coarse global structures;(2)Nonlinear Activation Free Blocks(NAFBlocks)enhance local detail representation and mid-level feature fusion;and(3)an optimized residual subnetwork based on gated feature modulation refines texture and edge details for high-fidelity restoration.Extensive experiments demonstrate that MRFNet achieves superior performance compared to state-of-the-art methods.On GoPro,it attains 32.52 dB Peak Signal-to-Noise Ratio(PSNR)and 0.071 Learned Perceptual Image Patch Similarity(LPIPS),outperforming MIMOWNet(32.50 dB,0.075).On HIDE,it achieves 30.25 dB PSNR and 0.945 Structural Similarity Index Measure(SSIM),representing gains of+0.26 dB and+0.015 SSIM over MIMO-UNet(29.99 dB,0.930).On RealBlur-J,it reaches 28.82 dB PSNR and 0.872 SSIM,surpassing MIMO-UNet by+1.19 dB and+0.035 SSIM(27.63 dB,0.837).These results validate the effectiveness of the proposed progressive residual fusion and hybrid attention mechanisms in balancing global context understanding and local detail recovery for blind image deblurring.
基金support from the National Natural Science Foundation of China(Grant No.22275018).
文摘The combustion and detonation processes of energetic materials exhibit remarkable complexity and ultra-fast transient characteristics.While reactive molecular dynamics has been extensively employed to investigate the reaction dynamics of energetic materials,its utility is often constrained to capturing only fundamental reaction events and species information,thereby limiting mechanistic investigations of complex reaction pathways.To elucidate the topological features of energetic material reaction networks and identify critical reaction pathways with high fidelity,this study presents ReacNetwork-an advanced large-scale reaction network analysis methodology that synergistically integrates complex network theory with molecular simulation techniques.Specifically,we have developed a multi-dimensional feature screening protocol based on node centrality metrics and K-shell decomposition algorithms.Takingα-Hexahydro-1,3,5-trinitro-1,3,5-triazine(α-RDX)as the subject,we successfully constructed a comprehensive high-temperature thermal decomposition reaction network consisting of 1,134 distinct chemical species and 3,626 elementary reactions.Through systematic application of community detection algorithms and global topological feature extraction techniques,we achieved effective dimensionality reduction and successfully identified the dominant reaction pathway within theα-RDX thermal decomposition network.The computational results not only validate the well-established initial reaction mechanism dominated by N-NO2 homolytic bond cleavage,but also provide unprecedented visualization ofα-RDX framework ring-opening dynamics and subsequent radical chain propagation networks.
基金supported by the Shanghai Pilot Program for Basic Research(22T01400100-18)the National Natural Science Foundation of China(22278127 and 12447149)+1 种基金the Fundamental Research Funds for the Central Universities(2022ZFJH004)the Postdoctoral Fellowship Program of CPSF(GZB20250159).
文摘Accurate state of health(SOH)estimation is essential for the safe and reliable operation of lithium-ion batteries.However,existing methods face significant challenges,primarily because they rely on complete charge–discharge cycles and fixed-form physical constraints,which limit adaptability to different chemistries and real-world conditions.To address these issues,this study proposes an approach that extracts features from segmented state of charge(SOC)intervals and integrates them into an enhanced physics-informed neural network(PINN).Specifically,voltage data within the 25%–75%SOC range during charging are used to derive statistical,time–frequency,and mechanism-based features that capture degradation trends.A hybrid PINN-Lasso-Transformer-BiLSTM architecture is developed,where Lasso regression enables sparse feature selection,and a nonlinear empirical degradation model is embedded as a learnable physical term within a dynamically scaled composite loss.This design adaptively balances data-driven accuracy with physical consistency,thereby enhancing estimation precision,robustness,and generalization.The results show that the proposed method outperforms conventional neural networks across four battery chemistries,achieving root mean square error and mean absolute error below 1%.Notably,features from partial charging segments exhibit higher robustness than those from full cycles.Furthermore,the model maintains strong performance under high temperatures and demonstrates excellent generalization capacity in transfer learning across chemistries,temperatures,and C-rates.This work establishes a scalable and interpretable solution for accurate SOH estimation under diverse practical operating conditions.
文摘This paper introduces a fuzzy C-means-based pooling layer for convolutional neural networks that explicitly models local uncertainty and ambiguity.Conventional pooling operations,such as max and average,apply rigid aggregation and often discard fine-grained boundary information.In contrast,our method computes soft membershipswithin each receptive field and aggregates cluster-wise responses throughmembership-weighted pooling,thereby preserving informative structure while reducing dimensionality.Being differentiable,the proposed layer operates as standard two-dimensional pooling.We evaluate our approach across various CNN backbones and open datasets,including CIFAR-10/100,STL-10,LFW,and ImageNette,and further probe small training set restrictions on MNIST and Fashion-MNIST.In these settings,the proposed pooling consistently improves accuracy and weighted F1 over conventional baselines,with particularly strong gains when training data are scarce.Even with less than 1%of the training set,ourmethodmaintains reliable performance,indicating improved sample efficiency and robustness to noisy or ambiguous local patterns.Overall,integrating soft memberships into the pooling operator provides a practical and generalizable inductive bias that enhances robustness and generalization in modern CNN pipelines.
基金supported by the National Natural Science Foundation of China under Grant 61573183.
文摘With the increase of semiconductor integration density,in order to cope with the increase of wafer defect complexity and types,especially the low recognition accuracy of overlapping mixed defects and unknown wafer defects,this study proposes a lightweight model for wafer defect detection called LightWMNet.First,using a hierarchical attention Encoder-Decoder architecture,the features of wafer defect pattern(WDP)are channel recalibrated to generate high-resolution fine-grained features and low-resolution coarse-grained features.Secondly,the backbone network incorporates two novel attention modules—feedforward spatial attention(FFSa)and feedforward channel attention(FFCa)—to amplify responses in critical defect regions and suppress noise from stochastic discrete pixels.These mechanisms synergistically enhance feature discriminability without introducing significant parametric overhead.Finally,the Dice loss function and the cross entropy loss function are combined to jointly evaluate the segmentation and classification accuracy of the model.Experimental results on the public mixed wafer defect dataset MixedWM38 show that the pixel accuracy(PA),intersection over union(IoU)and Dice coefficient of the proposed network reach 98.26%,94.83%and 97.22%,respectively.Without significantly increasing the computational complexity and size of the model,compared with the existing state-of-the-art(SOTA)model,the classification accuracy of lightWMNet in single defect,three mixed defects and four mixed defects is improved by 0.5%,0.25%and 0.89%respectively.Furthermore,we used transfer learning for the first time to evaluate the model's generalisation ability for unseen defect categories.The results showed that LightWMNet still has a certain recognition ability even in untrained wafer defects.
