The fusion of infrared and visible images should emphasize the salient targets in the infrared image while preserving the textural details of the visible images.To meet these requirements,an autoencoder-based method f...The fusion of infrared and visible images should emphasize the salient targets in the infrared image while preserving the textural details of the visible images.To meet these requirements,an autoencoder-based method for infrared and visible image fusion is proposed.The encoder designed according to the optimization objective consists of a base encoder and a detail encoder,which is used to extract low-frequency and high-frequency information from the image.This extraction may lead to some information not being captured,so a compensation encoder is proposed to supplement the missing information.Multi-scale decomposition is also employed to extract image features more comprehensively.The decoder combines low-frequency,high-frequency and supplementary information to obtain multi-scale features.Subsequently,the attention strategy and fusion module are introduced to perform multi-scale fusion for image reconstruction.Experimental results on three datasets show that the fused images generated by this network effectively retain salient targets while being more consistent with human visual perception.展开更多
Globally,diabetic retinopathy(DR)is the primary cause of blindness,affecting millions of people worldwide.This widespread impact underscores the critical need for reliable and precise diagnostic techniques to ensure p...Globally,diabetic retinopathy(DR)is the primary cause of blindness,affecting millions of people worldwide.This widespread impact underscores the critical need for reliable and precise diagnostic techniques to ensure prompt diagnosis and effective treatment.Deep learning-based automated diagnosis for diabetic retinopathy can facilitate early detection and treatment.However,traditional deep learning models that focus on local views often learn feature representations that are less discriminative at the semantic level.On the other hand,models that focus on global semantic-level information might overlook critical,subtle local pathological features.To address this issue,we propose an adaptive multi-scale feature fusion network called(AMSFuse),which can adaptively combine multi-scale global and local features without compromising their individual representation.Specifically,our model incorporates global features for extracting high-level contextual information from retinal images.Concurrently,local features capture fine-grained details,such as microaneurysms,hemorrhages,and exudates,which are critical for DR diagnosis.These global and local features are adaptively fused using a fusion block,followed by an Integrated Attention Mechanism(IAM)that refines the fused features by emphasizing relevant regions,thereby enhancing classification accuracy for DR classification.Our model achieves 86.3%accuracy on the APTOS dataset and 96.6%RFMiD,both of which are comparable to state-of-the-art methods.展开更多
Detecting abnormal cervical cells is crucial for early identification and timely treatment of cervical cancer.However,this task is challenging due to the morphological similarities between abnormal and normal cells an...Detecting abnormal cervical cells is crucial for early identification and timely treatment of cervical cancer.However,this task is challenging due to the morphological similarities between abnormal and normal cells and the significant variations in cell size.Pathologists often refer to surrounding cells to identify abnormalities.To emulate this slide examination behavior,this study proposes a Multi-Scale Feature Fusion Network(MSFF-Net)for detecting cervical abnormal cells.MSFF-Net employs a Cross-Scale Pooling Model(CSPM)to effectively capture diverse features and contextual information,ranging from local details to the overall structure.Additionally,a Multi-Scale Fusion Attention(MSFA)module is introduced to mitigate the impact of cell size variations by adaptively fusing local and global information at different scales.To handle the complex environment of cervical cell images,such as cell adhesion and overlapping,the Inner-CIoU loss function is utilized to more precisely measure the overlap between bounding boxes,thereby improving detection accuracy in such scenarios.Experimental results on the Comparison detector dataset demonstrate that MSFF-Net achieves a mean average precision(mAP)of 63.2%,outperforming state-of-the-art methods while maintaining a relatively small number of parameters(26.8 M).This study highlights the effectiveness of multi-scale feature fusion in enhancing the detection of cervical abnormal cells,contributing to more accurate and efficient cervical cancer screening.展开更多
To solve the problems of redundant feature information,the insignificant difference in feature representation,and low recognition accuracy of the fine-grained image,based on the ResNeXt50 model,an MSFResNet network mo...To solve the problems of redundant feature information,the insignificant difference in feature representation,and low recognition accuracy of the fine-grained image,based on the ResNeXt50 model,an MSFResNet network model is proposed by fusing multi-scale feature information.Firstly,a multi-scale feature extraction module is designed to obtain multi-scale information on feature images by using different scales of convolution kernels.Meanwhile,the channel attention mechanism is used to increase the global information acquisition of the network.Secondly,the feature images processed by the multi-scale feature extraction module are fused with the deep feature images through short links to guide the full learning of the network,thus reducing the loss of texture details of the deep network feature images,and improving network generalization ability and recognition accuracy.Finally,the validity of the MSFResNet model is verified using public datasets and applied to wild mushroom identification.Experimental results show that compared with ResNeXt50 network model,the accuracy of the MSFResNet model is improved by 6.01%on the FGVC-Aircraft common dataset.It achieves 99.13%classification accuracy on the wild mushroom dataset,which is 0.47%higher than ResNeXt50.Furthermore,the experimental results of the thermal map show that the MSFResNet model significantly reduces the interference of background information,making the network focus on the location of the main body of wild mushroom,which can effectively improve the accuracy of wild mushroom identification.展开更多
With the rapid growth of socialmedia,the spread of fake news has become a growing problem,misleading the public and causing significant harm.As social media content is often composed of both images and text,the use of...With the rapid growth of socialmedia,the spread of fake news has become a growing problem,misleading the public and causing significant harm.As social media content is often composed of both images and text,the use of multimodal approaches for fake news detection has gained significant attention.To solve the problems existing in previous multi-modal fake news detection algorithms,such as insufficient feature extraction and insufficient use of semantic relations between modes,this paper proposes the MFFFND-Co(Multimodal Feature Fusion Fake News Detection with Co-Attention Block)model.First,the model deeply explores the textual content,image content,and frequency domain features.Then,it employs a Co-Attention mechanism for cross-modal fusion.Additionally,a semantic consistency detectionmodule is designed to quantify semantic deviations,thereby enhancing the performance of fake news detection.Experimentally verified on two commonly used datasets,Twitter and Weibo,the model achieved F1 scores of 90.0% and 94.0%,respectively,significantly outperforming the pre-modified MFFFND(Multimodal Feature Fusion Fake News Detection with Attention Block)model and surpassing other baseline models.This improves the accuracy of detecting fake information in artificial intelligence detection and engineering software detection.展开更多
Deep Learning has been widely used to model soft sensors in modern industrial processes with nonlinear variables and uncertainty.Due to the outstanding ability for high-level feature extraction,stacked autoencoder(SAE...Deep Learning has been widely used to model soft sensors in modern industrial processes with nonlinear variables and uncertainty.Due to the outstanding ability for high-level feature extraction,stacked autoencoder(SAE)has been widely used to improve the model accuracy of soft sensors.However,with the increase of network layers,SAE may encounter serious information loss issues,which affect the modeling performance of soft sensors.Besides,there are typically very few labeled samples in the data set,which brings challenges to traditional neural networks to solve.In this paper,a multi-scale feature fused stacked autoencoder(MFF-SAE)is suggested for feature representation related to hierarchical output,where stacked autoencoder,mutual information(MI)and multi-scale feature fusion(MFF)strategies are integrated.Based on correlation analysis between output and input variables,critical hidden variables are extracted from the original variables in each autoencoder's input layer,which are correspondingly given varying weights.Besides,an integration strategy based on multi-scale feature fusion is adopted to mitigate the impact of information loss with the deepening of the network layers.Then,the MFF-SAE method is designed and stacked to form deep networks.Two practical industrial processes are utilized to evaluate the performance of MFF-SAE.Results from simulations indicate that in comparison to other cutting-edge techniques,the proposed method may considerably enhance the accuracy of soft sensor modeling,where the suggested method reduces the root mean square error(RMSE)by 71.8%,17.1%and 64.7%,15.1%,respectively.展开更多
Convolutional neural networks (CNNs) are widely used in image classification tasks, but their increasing model size and computation make them challenging to implement on embedded systems with constrained hardware reso...Convolutional neural networks (CNNs) are widely used in image classification tasks, but their increasing model size and computation make them challenging to implement on embedded systems with constrained hardware resources. To address this issue, the MobileNetV1 network was developed, which employs depthwise convolution to reduce network complexity. MobileNetV1 employs a stride of 2 in several convolutional layers to decrease the spatial resolution of feature maps, thereby lowering computational costs. However, this stride setting can lead to a loss of spatial information, particularly affecting the detection and representation of smaller objects or finer details in images. To maintain the trade-off between complexity and model performance, a lightweight convolutional neural network with hierarchical multi-scale feature fusion based on the MobileNetV1 network is proposed. The network consists of two main subnetworks. The first subnetwork uses a depthwise dilated separable convolution (DDSC) layer to learn imaging features with fewer parameters, which results in a lightweight and computationally inexpensive network. Furthermore, depthwise dilated convolution in DDSC layer effectively expands the field of view of filters, allowing them to incorporate a larger context. The second subnetwork is a hierarchical multi-scale feature fusion (HMFF) module that uses parallel multi-resolution branches architecture to process the input feature map in order to extract the multi-scale feature information of the input image. Experimental results on the CIFAR-10, Malaria, and KvasirV1 datasets demonstrate that the proposed method is efficient, reducing the network parameters and computational cost by 65.02% and 39.78%, respectively, while maintaining the network performance compared to the MobileNetV1 baseline.展开更多
Solar cell defect detection is crucial for quality inspection in photovoltaic power generation modules.In the production process,defect samples occur infrequently and exhibit random shapes and sizes,which makes it cha...Solar cell defect detection is crucial for quality inspection in photovoltaic power generation modules.In the production process,defect samples occur infrequently and exhibit random shapes and sizes,which makes it challenging to collect defective samples.Additionally,the complex surface background of polysilicon cell wafers complicates the accurate identification and localization of defective regions.This paper proposes a novel Lightweight Multiscale Feature Fusion network(LMFF)to address these challenges.The network comprises a feature extraction network,a multi-scale feature fusion module(MFF),and a segmentation network.Specifically,a feature extraction network is proposed to obtain multi-scale feature outputs,and a multi-scale feature fusion module(MFF)is used to fuse multi-scale feature information effectively.In order to capture finer-grained multi-scale information from the fusion features,we propose a multi-scale attention module(MSA)in the segmentation network to enhance the network’s ability for small target detection.Moreover,depthwise separable convolutions are introduced to construct depthwise separable residual blocks(DSR)to reduce the model’s parameter number.Finally,to validate the proposed method’s defect segmentation and localization performance,we constructed three solar cell defect detection datasets:SolarCells,SolarCells-S,and PVEL-S.SolarCells and SolarCells-S are monocrystalline silicon datasets,and PVEL-S is a polycrystalline silicon dataset.Experimental results show that the IOU of our method on these three datasets can reach 68.5%,51.0%,and 92.7%,respectively,and the F1-Score can reach 81.3%,67.5%,and 96.2%,respectively,which surpasses other commonly usedmethods and verifies the effectiveness of our LMFF network.展开更多
The capacity to diagnose faults in rolling bearings is of significant practical importance to ensure the normal operation of the equipment.Frequency-domain features can effectively enhance the identification of fault ...The capacity to diagnose faults in rolling bearings is of significant practical importance to ensure the normal operation of the equipment.Frequency-domain features can effectively enhance the identification of fault modes.However,existing methods often suffer from insufficient frequency-domain representation in practical applications,which greatly affects diagnostic performance.Therefore,this paper proposes a rolling bearing fault diagnosismethod based on aMulti-Scale FusionNetwork(MSFN)using the Time-Division Fourier Transform(TDFT).The method constructs multi-scale channels to extract time-domain and frequency-domain features of the signal in parallel.A multi-level,multi-scale filter-based approach is designed to extract frequency-domain features in a segmented manner.A cross-attention mechanism is introduced to facilitate the fusion of the extracted time-frequency domain features.The performance of the proposed method is validated using the CWRU and Ottawa datasets.The results show that the average accuracy of MSFN under complex noisy signals is 97.75%and 94.41%.The average accuracy under variable load conditions is 98.68%.This demonstrates its significant application potential compared to existing methods.展开更多
Efficient road distress detection is crucial for transportation safety.To address the challenge of balancing detection accuracy,efficiency,and multi-scale feature fusion in existing methods,this paper proposes a light...Efficient road distress detection is crucial for transportation safety.To address the challenge of balancing detection accuracy,efficiency,and multi-scale feature fusion in existing methods,this paper proposes a lightweight model named MADF-YOLOv8.The model enhances multi-scale feature extraction capability by introducing the Multi-Scale Ghost Residual Convolution(MSGRConv)and the Multiscale Adaptive Feature Processing Module(MAFP).Furthermore,it constructs a Multi-scale Dynamic sampling Bidirectional Feature Pyramid Network(MD-BiFPN)and incorporates the C2f-Faster module to optimize feature fusion efficiency.Experiments on the RDD2022 dataset demonstrate that the proposed model achieves a mean Average Precision at 0.5 Intersection over Union(mAP@0.5)of 88.6%with only 2.312 million parameters.Its overall performance surpasses various mainstream detectors,achieving an exceptional balance between accuracy and efficiency.展开更多
Infrared images typically exhibit diverse backgrounds,each potentially containing noise and target-like interference elements.In complex backgrounds,infrared small targets are prone to be submerged by background noise...Infrared images typically exhibit diverse backgrounds,each potentially containing noise and target-like interference elements.In complex backgrounds,infrared small targets are prone to be submerged by background noise due to their low pixel proportion and limited available features,leading to detection failure.To address this problem,this paper proposes an Attention Shift-Invariant Cross-Evolutionary Feature Fusion Network(ASCFNet)tailored for the detection of infrared weak and small targets.The network architecture first designs a Multidimensional Lightweight Pixel-level Attention Module(MLPA),which alleviates the issue of small-target feature suppression during deep network propagation by combining channel reshaping,multi-scale parallel subnet architectures,and local cross-channel interactions.Then,a Multidimensional Shift-Invariant Recall Module(MSIR)is designed to ensure the network remains unaffected by minor input perturbations when processing infrared images,through focusing on the model’s shift invariance.Subsequently,a Cross-Evolutionary Feature Fusion structure(CEFF)is designed to allow flexible and efficient integration of multidimensional feature information from different network hierarchies,thereby achieving complementarity and enhancement among features.Experimental results on three public datasets,SIRST,NUDT-SIRST,and IRST640,demonstrate that our proposed network outperforms advanced algorithms in the field.Specifically,on the NUDT-SIRST dataset,the mAP50,mAP50-95,and metrics reached 99.26%,85.22%,and 99.31%,respectively.Visual evaluations of detection results in diverse scenarios indicate that our algorithm exhibits an increased detection rate and reduced false alarm rate.Our method balances accuracy and real-time performance,and achieves efficient and stable detection of infrared weak and small targets.展开更多
Image registration is an indispensable component in multi-source remote sensing image processing. In this paper, we put forward a remote sensing image registration method by including an improved multi-scale and multi...Image registration is an indispensable component in multi-source remote sensing image processing. In this paper, we put forward a remote sensing image registration method by including an improved multi-scale and multi-direction Harris algorithm and a novel compound feature. Multi-scale circle Gaussian combined invariant moments and multi-direction gray level co-occurrence matrix are extracted as features for image matching. The proposed algorithm is evaluated on numerous multi-source remote sensor images with noise and illumination changes. Extensive experimental studies prove that our proposed method is capable of receiving stable and even distribution of key points as well as obtaining robust and accurate correspondence matches. It is a promising scheme in multi-source remote sensing image registration.展开更多
Convolutional neural networks(CNNs)-based medical image segmentation technologies have been widely used in medical image segmentation because of their strong representation and generalization abilities.However,due to ...Convolutional neural networks(CNNs)-based medical image segmentation technologies have been widely used in medical image segmentation because of their strong representation and generalization abilities.However,due to the inability to effectively capture global information from images,CNNs can easily lead to loss of contours and textures in segmentation results.Notice that the transformer model can effectively capture the properties of long-range dependencies in the image,and furthermore,combining the CNN and the transformer can effectively extract local details and global contextual features of the image.Motivated by this,we propose a multi-branch and multi-scale attention network(M2ANet)for medical image segmentation,whose architecture consists of three components.Specifically,in the first component,we construct an adaptive multi-branch patch module for parallel extraction of image features to reduce information loss caused by downsampling.In the second component,we apply residual block to the well-known convolutional block attention module to enhance the network’s ability to recognize important features of images and alleviate the phenomenon of gradient vanishing.In the third component,we design a multi-scale feature fusion module,in which we adopt adaptive average pooling and position encoding to enhance contextual features,and then multi-head attention is introduced to further enrich feature representation.Finally,we validate the effectiveness and feasibility of the proposed M2ANet method through comparative experiments on four benchmark medical image segmentation datasets,particularly in the context of preserving contours and textures.展开更多
In order to improve the detection accuracy of small objects,a neighborhood fusion-based hierarchical parallel feature pyramid network(NFPN)is proposed.Unlike the layer-by-layer structure adopted in the feature pyramid...In order to improve the detection accuracy of small objects,a neighborhood fusion-based hierarchical parallel feature pyramid network(NFPN)is proposed.Unlike the layer-by-layer structure adopted in the feature pyramid network(FPN)and deconvolutional single shot detector(DSSD),where the bottom layer of the feature pyramid network relies on the top layer,NFPN builds the feature pyramid network with no connections between the upper and lower layers.That is,it only fuses shallow features on similar scales.NFPN is highly portable and can be embedded in many models to further boost performance.Extensive experiments on PASCAL VOC 2007,2012,and COCO datasets demonstrate that the NFPN-based SSD without intricate tricks can exceed the DSSD model in terms of detection accuracy and inference speed,especially for small objects,e.g.,4%to 5%higher mAP(mean average precision)than SSD,and 2%to 3%higher mAP than DSSD.On VOC 2007 test set,the NFPN-based SSD with 300×300 input reaches 79.4%mAP at 34.6 frame/s,and the mAP can raise to 82.9%after using the multi-scale testing strategy.展开更多
A novel face recognition method based on fusion of spatial and frequency features was presented to improve recognition accuracy. Dual-Tree Complex Wavelet Transform derives desirable facial features to cope with the v...A novel face recognition method based on fusion of spatial and frequency features was presented to improve recognition accuracy. Dual-Tree Complex Wavelet Transform derives desirable facial features to cope with the variation due to the illumination and facial expression changes. By adopting spectral regression and complex fusion technologies respectively, two improved neighborhood preserving discriminant analysis feature extraction methods were proposed to capture the face manifold structures and locality discriminatory information. Extensive experiments have been made to compare the recognition performance of the proposed method with some popular dimensionality reduction methods on ORL and Yale face databases. The results verify the effectiveness of the proposed method.展开更多
Semantic segmentation plays a foundational role in biomedical image analysis, providing precise information about cellular, tissue, and organ structures in both biological and medical imaging modalities. Traditional a...Semantic segmentation plays a foundational role in biomedical image analysis, providing precise information about cellular, tissue, and organ structures in both biological and medical imaging modalities. Traditional approaches often fail in the face of challenges such as low contrast, morphological variability, and densely packed structures. Recent advancements in deep learning have transformed segmentation capabilities through the integration of fine-scale detail preservation, coarse-scale contextual modeling, and multi-scale feature fusion. This work provides a comprehensive analysis of state-of-the-art deep learning models, including U-Net variants, attention-based frameworks, and Transformer-integrated networks, highlighting innovations that improve accuracy, generalizability, and computational efficiency. Key architectural components such as convolution operations, shallow and deep blocks, skip connections, and hybrid encoders are examined for their roles in enhancing spatial representation and semantic consistency. We further discuss the importance of hierarchical and instance-aware segmentation and annotation in interpreting complex biological scenes and multiplexed medical images. By bridging methodological developments with diverse application domains, this paper outlines current trends and future directions for semantic segmentation, emphasizing its critical role in facilitating annotation, diagnosis, and discovery in biomedical research.展开更多
The precise and automatic segmentation of prostate magnetic resonance imaging(MRI)images is vital for assisting doctors in diagnosing prostate diseases.In recent years,many advanced methods have been applied to prosta...The precise and automatic segmentation of prostate magnetic resonance imaging(MRI)images is vital for assisting doctors in diagnosing prostate diseases.In recent years,many advanced methods have been applied to prostate segmentation,but due to the variability caused by prostate diseases,automatic segmentation of the prostate presents significant challenges.In this paper,we propose an attention-guided multi-scale feature fusion network(AGMSF-Net)to segment prostate MRI images.We propose an attention mechanism for extracting multi-scale features,and introduce a 3D transformer module to enhance global feature representation by adding it during the transition phase from encoder to decoder.In the decoder stage,a feature fusion module is proposed to obtain global context information.We evaluate our model on MRI images of the prostate acquired from a local hospital.The relative volume difference(RVD)and dice similarity coefficient(DSC)between the results of automatic prostate segmentation and ground truth were 1.21%and 93.68%,respectively.To quantitatively evaluate prostate volume on MRI,which is of significant clinical significance,we propose a unique AGMSF-Net.The essential performance evaluation and validation experiments have demonstrated the effectiveness of our method in automatic prostate segmentation.展开更多
A system for classifying four basic table tennis strokes using wearable devices and deep learning networks is proposed in this study.The wearable device consisted of a six-axis sensor,Raspberry Pi 3,and a power bank.M...A system for classifying four basic table tennis strokes using wearable devices and deep learning networks is proposed in this study.The wearable device consisted of a six-axis sensor,Raspberry Pi 3,and a power bank.Multiple kernel sizes were used in convolutional neural network(CNN)to evaluate their performance for extracting features.Moreover,a multiscale CNN with two kernel sizes was used to perform feature fusion at different scales in a concatenated manner.The CNN achieved recognition of the four table tennis strokes.Experimental data were obtained from20 research participants who wore sensors on the back of their hands while performing the four table tennis strokes in a laboratory environment.The data were collected to verify the performance of the proposed models for wearable devices.Finally,the sensor and multi-scale CNN designed in this study achieved accuracy and F1 scores of 99.58%and 99.16%,respectively,for the four strokes.The accuracy for five-fold cross validation was 99.87%.This result also shows that the multi-scale convolutional neural network has better robustness after fivefold cross validation.展开更多
To explore the influence of the fusion of different features on recognition,this paper took the electromyography(EMG)signals of rectus femoris under different motions(walk,step,ramp,squat,and sitting)as samples,linear...To explore the influence of the fusion of different features on recognition,this paper took the electromyography(EMG)signals of rectus femoris under different motions(walk,step,ramp,squat,and sitting)as samples,linear features(time-domain features(variance(VAR)and root mean square(RMS)),frequency-domain features(mean frequency(MF)and mean power frequency(MPF)),and nonlinear features(empirical mode decomposition(EMD))of the samples were extracted.Two feature fusion algorithms,the series splicing method and complex vector method,were designed,which were verified by a double hidden layer(BP)error back propagation neural network.Results show that with the increase of the types and complexity of feature fusions,the recognition rate of the EMG signal to actions is gradually improved.When the EMG signal is used in the series splicing method,the recognition rate of time-domain+frequency-domain+empirical mode decomposition(TD+FD+EMD)splicing is the highest,and the average recognition rate is 92.32%.And this rate is raised to 96.1%by using the complex vector method,and the variance of the BP system is also reduced.展开更多
The YOLOv5 algorithm is widely used in edge computing systems for object detection.However,the limited computing resources of embedded devices and the large model size of existing deep learning based methods increase ...The YOLOv5 algorithm is widely used in edge computing systems for object detection.However,the limited computing resources of embedded devices and the large model size of existing deep learning based methods increase the difficulty of real-time object detection on edge devices.To address this issue,we propose a smaller,less computationally intensive,and more accurate algorithm for object detection.Multi-scale Feature Fusion-YOLO(MFF-YOLO)is built on top of the YOLOv5s framework,but it contains substantial improvements to YOLOv5s.First,we design the MFF module to improve the feature propagation path in the feature pyramid,which further integrates the semantic information from different paths of feature layers.Then,a large convolution-kernel module is used in the bottleneck.The structure enlarges the receptive field and preserves shallow semantic information,which overcomes the performance limitation arising from uneven propagation in Feature Pyramid Networks(FPN).In addition,a multi-branch downsampling method based on depthwise separable convolutions and a bottleneck structure with deformable convolutions are designed to reduce the complexity of the backbone network and minimize the real-time performance loss caused by the increased model complexity.The experimental results on PASCAL VOC and MS COCO datasets show that,compared with YOLOv5s,MFF-YOLO reduces the number of parameters by 7%and the number of FLoating point Operations Per second(FLOPs)by 11.8%.The mAP@0.5 has improved by 3.7%and 5.5%,and the mAP@0.5:0.95 has improved by 6.5%and 6.2%,respetively.Furthermore,compared with YOLOv7-tiny,PP-YOLO-tiny,and other mainstream methods,MFF-YOLO has achieved better results on multiple indicators.展开更多
基金Supported by the Henan Province Key Research and Development Project(231111211300)the Central Government of Henan Province Guides Local Science and Technology Development Funds(Z20231811005)+2 种基金Henan Province Key Research and Development Project(231111110100)Henan Provincial Outstanding Foreign Scientist Studio(GZS2024006)Henan Provincial Joint Fund for Scientific and Technological Research and Development Plan(Application and Overcoming Technical Barriers)(242103810028)。
文摘The fusion of infrared and visible images should emphasize the salient targets in the infrared image while preserving the textural details of the visible images.To meet these requirements,an autoencoder-based method for infrared and visible image fusion is proposed.The encoder designed according to the optimization objective consists of a base encoder and a detail encoder,which is used to extract low-frequency and high-frequency information from the image.This extraction may lead to some information not being captured,so a compensation encoder is proposed to supplement the missing information.Multi-scale decomposition is also employed to extract image features more comprehensively.The decoder combines low-frequency,high-frequency and supplementary information to obtain multi-scale features.Subsequently,the attention strategy and fusion module are introduced to perform multi-scale fusion for image reconstruction.Experimental results on three datasets show that the fused images generated by this network effectively retain salient targets while being more consistent with human visual perception.
基金supported by the National Natural Science Foundation of China(No.62376287)the International Science and Technology Innovation Joint Base of Machine Vision and Medical Image Processing in Hunan Province(2021CB1013)the Natural Science Foundation of Hunan Province(Nos.2022JJ30762,2023JJ70016).
文摘Globally,diabetic retinopathy(DR)is the primary cause of blindness,affecting millions of people worldwide.This widespread impact underscores the critical need for reliable and precise diagnostic techniques to ensure prompt diagnosis and effective treatment.Deep learning-based automated diagnosis for diabetic retinopathy can facilitate early detection and treatment.However,traditional deep learning models that focus on local views often learn feature representations that are less discriminative at the semantic level.On the other hand,models that focus on global semantic-level information might overlook critical,subtle local pathological features.To address this issue,we propose an adaptive multi-scale feature fusion network called(AMSFuse),which can adaptively combine multi-scale global and local features without compromising their individual representation.Specifically,our model incorporates global features for extracting high-level contextual information from retinal images.Concurrently,local features capture fine-grained details,such as microaneurysms,hemorrhages,and exudates,which are critical for DR diagnosis.These global and local features are adaptively fused using a fusion block,followed by an Integrated Attention Mechanism(IAM)that refines the fused features by emphasizing relevant regions,thereby enhancing classification accuracy for DR classification.Our model achieves 86.3%accuracy on the APTOS dataset and 96.6%RFMiD,both of which are comparable to state-of-the-art methods.
基金funded by the China Chongqing Municipal Science and Technology Bureau,grant numbers 2024TIAD-CYKJCXX0121,2024NSCQ-LZX0135Chongqing Municipal Commission of Housing and Urban-Rural Development,grant number CKZ2024-87+3 种基金the Chongqing University of Technology graduate education high-quality development project,grant number gzlsz202401the Chongqing University of Technology-Chongqing LINGLUE Technology Co.,Ltd.,Electronic Information(Artificial Intelligence)graduate joint training basethe Postgraduate Education and Teaching Reform Research Project in Chongqing,grant number yjg213116the Chongqing University of Technology-CISDI Chongqing Information Technology Co.,Ltd.,Computer Technology graduate joint training base.
文摘Detecting abnormal cervical cells is crucial for early identification and timely treatment of cervical cancer.However,this task is challenging due to the morphological similarities between abnormal and normal cells and the significant variations in cell size.Pathologists often refer to surrounding cells to identify abnormalities.To emulate this slide examination behavior,this study proposes a Multi-Scale Feature Fusion Network(MSFF-Net)for detecting cervical abnormal cells.MSFF-Net employs a Cross-Scale Pooling Model(CSPM)to effectively capture diverse features and contextual information,ranging from local details to the overall structure.Additionally,a Multi-Scale Fusion Attention(MSFA)module is introduced to mitigate the impact of cell size variations by adaptively fusing local and global information at different scales.To handle the complex environment of cervical cell images,such as cell adhesion and overlapping,the Inner-CIoU loss function is utilized to more precisely measure the overlap between bounding boxes,thereby improving detection accuracy in such scenarios.Experimental results on the Comparison detector dataset demonstrate that MSFF-Net achieves a mean average precision(mAP)of 63.2%,outperforming state-of-the-art methods while maintaining a relatively small number of parameters(26.8 M).This study highlights the effectiveness of multi-scale feature fusion in enhancing the detection of cervical abnormal cells,contributing to more accurate and efficient cervical cancer screening.
基金supported by National Natural Science Foundation of China(No.61862037)Lanzhou Jiaotong University Tianyou Innovation Team Project(No.TY202002)。
文摘To solve the problems of redundant feature information,the insignificant difference in feature representation,and low recognition accuracy of the fine-grained image,based on the ResNeXt50 model,an MSFResNet network model is proposed by fusing multi-scale feature information.Firstly,a multi-scale feature extraction module is designed to obtain multi-scale information on feature images by using different scales of convolution kernels.Meanwhile,the channel attention mechanism is used to increase the global information acquisition of the network.Secondly,the feature images processed by the multi-scale feature extraction module are fused with the deep feature images through short links to guide the full learning of the network,thus reducing the loss of texture details of the deep network feature images,and improving network generalization ability and recognition accuracy.Finally,the validity of the MSFResNet model is verified using public datasets and applied to wild mushroom identification.Experimental results show that compared with ResNeXt50 network model,the accuracy of the MSFResNet model is improved by 6.01%on the FGVC-Aircraft common dataset.It achieves 99.13%classification accuracy on the wild mushroom dataset,which is 0.47%higher than ResNeXt50.Furthermore,the experimental results of the thermal map show that the MSFResNet model significantly reduces the interference of background information,making the network focus on the location of the main body of wild mushroom,which can effectively improve the accuracy of wild mushroom identification.
基金supported by Communication University of China(HG23035)partly supported by the Fundamental Research Funds for the Central Universities(CUC230A013).
文摘With the rapid growth of socialmedia,the spread of fake news has become a growing problem,misleading the public and causing significant harm.As social media content is often composed of both images and text,the use of multimodal approaches for fake news detection has gained significant attention.To solve the problems existing in previous multi-modal fake news detection algorithms,such as insufficient feature extraction and insufficient use of semantic relations between modes,this paper proposes the MFFFND-Co(Multimodal Feature Fusion Fake News Detection with Co-Attention Block)model.First,the model deeply explores the textual content,image content,and frequency domain features.Then,it employs a Co-Attention mechanism for cross-modal fusion.Additionally,a semantic consistency detectionmodule is designed to quantify semantic deviations,thereby enhancing the performance of fake news detection.Experimentally verified on two commonly used datasets,Twitter and Weibo,the model achieved F1 scores of 90.0% and 94.0%,respectively,significantly outperforming the pre-modified MFFFND(Multimodal Feature Fusion Fake News Detection with Attention Block)model and surpassing other baseline models.This improves the accuracy of detecting fake information in artificial intelligence detection and engineering software detection.
基金supported by the National Key Research and Development Program of China(2023YFB3307800)National Natural Science Foundation of China(62394343,62373155)+2 种基金Major Science and Technology Project of Xinjiang(No.2022A01006-4)State Key Laboratory of Industrial Control Technology,China(Grant No.ICT2024A26)Fundamental Research Funds for the Central Universities.
文摘Deep Learning has been widely used to model soft sensors in modern industrial processes with nonlinear variables and uncertainty.Due to the outstanding ability for high-level feature extraction,stacked autoencoder(SAE)has been widely used to improve the model accuracy of soft sensors.However,with the increase of network layers,SAE may encounter serious information loss issues,which affect the modeling performance of soft sensors.Besides,there are typically very few labeled samples in the data set,which brings challenges to traditional neural networks to solve.In this paper,a multi-scale feature fused stacked autoencoder(MFF-SAE)is suggested for feature representation related to hierarchical output,where stacked autoencoder,mutual information(MI)and multi-scale feature fusion(MFF)strategies are integrated.Based on correlation analysis between output and input variables,critical hidden variables are extracted from the original variables in each autoencoder's input layer,which are correspondingly given varying weights.Besides,an integration strategy based on multi-scale feature fusion is adopted to mitigate the impact of information loss with the deepening of the network layers.Then,the MFF-SAE method is designed and stacked to form deep networks.Two practical industrial processes are utilized to evaluate the performance of MFF-SAE.Results from simulations indicate that in comparison to other cutting-edge techniques,the proposed method may considerably enhance the accuracy of soft sensor modeling,where the suggested method reduces the root mean square error(RMSE)by 71.8%,17.1%and 64.7%,15.1%,respectively.
文摘Convolutional neural networks (CNNs) are widely used in image classification tasks, but their increasing model size and computation make them challenging to implement on embedded systems with constrained hardware resources. To address this issue, the MobileNetV1 network was developed, which employs depthwise convolution to reduce network complexity. MobileNetV1 employs a stride of 2 in several convolutional layers to decrease the spatial resolution of feature maps, thereby lowering computational costs. However, this stride setting can lead to a loss of spatial information, particularly affecting the detection and representation of smaller objects or finer details in images. To maintain the trade-off between complexity and model performance, a lightweight convolutional neural network with hierarchical multi-scale feature fusion based on the MobileNetV1 network is proposed. The network consists of two main subnetworks. The first subnetwork uses a depthwise dilated separable convolution (DDSC) layer to learn imaging features with fewer parameters, which results in a lightweight and computationally inexpensive network. Furthermore, depthwise dilated convolution in DDSC layer effectively expands the field of view of filters, allowing them to incorporate a larger context. The second subnetwork is a hierarchical multi-scale feature fusion (HMFF) module that uses parallel multi-resolution branches architecture to process the input feature map in order to extract the multi-scale feature information of the input image. Experimental results on the CIFAR-10, Malaria, and KvasirV1 datasets demonstrate that the proposed method is efficient, reducing the network parameters and computational cost by 65.02% and 39.78%, respectively, while maintaining the network performance compared to the MobileNetV1 baseline.
基金supported in part by the National Natural Science Foundation of China under Grants 62463002,62062021 and 62473033in part by the Guiyang Scientific Plan Project[2023]48–11,in part by QKHZYD[2023]010 Guizhou Province Science and Technology Innovation Base Construction Project“Key Laboratory Construction of Intelligent Mountain Agricultural Equipment”.
文摘Solar cell defect detection is crucial for quality inspection in photovoltaic power generation modules.In the production process,defect samples occur infrequently and exhibit random shapes and sizes,which makes it challenging to collect defective samples.Additionally,the complex surface background of polysilicon cell wafers complicates the accurate identification and localization of defective regions.This paper proposes a novel Lightweight Multiscale Feature Fusion network(LMFF)to address these challenges.The network comprises a feature extraction network,a multi-scale feature fusion module(MFF),and a segmentation network.Specifically,a feature extraction network is proposed to obtain multi-scale feature outputs,and a multi-scale feature fusion module(MFF)is used to fuse multi-scale feature information effectively.In order to capture finer-grained multi-scale information from the fusion features,we propose a multi-scale attention module(MSA)in the segmentation network to enhance the network’s ability for small target detection.Moreover,depthwise separable convolutions are introduced to construct depthwise separable residual blocks(DSR)to reduce the model’s parameter number.Finally,to validate the proposed method’s defect segmentation and localization performance,we constructed three solar cell defect detection datasets:SolarCells,SolarCells-S,and PVEL-S.SolarCells and SolarCells-S are monocrystalline silicon datasets,and PVEL-S is a polycrystalline silicon dataset.Experimental results show that the IOU of our method on these three datasets can reach 68.5%,51.0%,and 92.7%,respectively,and the F1-Score can reach 81.3%,67.5%,and 96.2%,respectively,which surpasses other commonly usedmethods and verifies the effectiveness of our LMFF network.
基金fully supported by the Frontier Exploration Projects of Longmen Laboratory(No.LMQYTSKT034)Key Research and Development and Promotion of Special(Science and Technology)Project of Henan Province,China(No.252102210158)。
文摘The capacity to diagnose faults in rolling bearings is of significant practical importance to ensure the normal operation of the equipment.Frequency-domain features can effectively enhance the identification of fault modes.However,existing methods often suffer from insufficient frequency-domain representation in practical applications,which greatly affects diagnostic performance.Therefore,this paper proposes a rolling bearing fault diagnosismethod based on aMulti-Scale FusionNetwork(MSFN)using the Time-Division Fourier Transform(TDFT).The method constructs multi-scale channels to extract time-domain and frequency-domain features of the signal in parallel.A multi-level,multi-scale filter-based approach is designed to extract frequency-domain features in a segmented manner.A cross-attention mechanism is introduced to facilitate the fusion of the extracted time-frequency domain features.The performance of the proposed method is validated using the CWRU and Ottawa datasets.The results show that the average accuracy of MSFN under complex noisy signals is 97.75%and 94.41%.The average accuracy under variable load conditions is 98.68%.This demonstrates its significant application potential compared to existing methods.
基金supported by the National Key Research and Development Program of China under Grant No.2022YFA1602003,entitled"Intelligent Monitoring of Taishan Neutrino Detector".
文摘Efficient road distress detection is crucial for transportation safety.To address the challenge of balancing detection accuracy,efficiency,and multi-scale feature fusion in existing methods,this paper proposes a lightweight model named MADF-YOLOv8.The model enhances multi-scale feature extraction capability by introducing the Multi-Scale Ghost Residual Convolution(MSGRConv)and the Multiscale Adaptive Feature Processing Module(MAFP).Furthermore,it constructs a Multi-scale Dynamic sampling Bidirectional Feature Pyramid Network(MD-BiFPN)and incorporates the C2f-Faster module to optimize feature fusion efficiency.Experiments on the RDD2022 dataset demonstrate that the proposed model achieves a mean Average Precision at 0.5 Intersection over Union(mAP@0.5)of 88.6%with only 2.312 million parameters.Its overall performance surpasses various mainstream detectors,achieving an exceptional balance between accuracy and efficiency.
基金supported in part by the National Natural Science Foundation of China under Grant 62271302the Shanghai Municipal Natural Science Foundation under Grant 20ZR1423500.
文摘Infrared images typically exhibit diverse backgrounds,each potentially containing noise and target-like interference elements.In complex backgrounds,infrared small targets are prone to be submerged by background noise due to their low pixel proportion and limited available features,leading to detection failure.To address this problem,this paper proposes an Attention Shift-Invariant Cross-Evolutionary Feature Fusion Network(ASCFNet)tailored for the detection of infrared weak and small targets.The network architecture first designs a Multidimensional Lightweight Pixel-level Attention Module(MLPA),which alleviates the issue of small-target feature suppression during deep network propagation by combining channel reshaping,multi-scale parallel subnet architectures,and local cross-channel interactions.Then,a Multidimensional Shift-Invariant Recall Module(MSIR)is designed to ensure the network remains unaffected by minor input perturbations when processing infrared images,through focusing on the model’s shift invariance.Subsequently,a Cross-Evolutionary Feature Fusion structure(CEFF)is designed to allow flexible and efficient integration of multidimensional feature information from different network hierarchies,thereby achieving complementarity and enhancement among features.Experimental results on three public datasets,SIRST,NUDT-SIRST,and IRST640,demonstrate that our proposed network outperforms advanced algorithms in the field.Specifically,on the NUDT-SIRST dataset,the mAP50,mAP50-95,and metrics reached 99.26%,85.22%,and 99.31%,respectively.Visual evaluations of detection results in diverse scenarios indicate that our algorithm exhibits an increased detection rate and reduced false alarm rate.Our method balances accuracy and real-time performance,and achieves efficient and stable detection of infrared weak and small targets.
基金supported by National Nature Science Foundation of China (Nos. 61462046 and 61762052)Natural Science Foundation of Jiangxi Province (Nos. 20161BAB202049 and 20161BAB204172)+2 种基金the Bidding Project of the Key Laboratory of Watershed Ecology and Geographical Environment Monitoring, NASG (Nos. WE2016003, WE2016013 and WE2016015)the Science and Technology Research Projects of Jiangxi Province Education Department (Nos. GJJ160741, GJJ170632 and GJJ170633)the Art Planning Project of Jiangxi Province (Nos. YG2016250 and YG2017381)
文摘Image registration is an indispensable component in multi-source remote sensing image processing. In this paper, we put forward a remote sensing image registration method by including an improved multi-scale and multi-direction Harris algorithm and a novel compound feature. Multi-scale circle Gaussian combined invariant moments and multi-direction gray level co-occurrence matrix are extracted as features for image matching. The proposed algorithm is evaluated on numerous multi-source remote sensor images with noise and illumination changes. Extensive experimental studies prove that our proposed method is capable of receiving stable and even distribution of key points as well as obtaining robust and accurate correspondence matches. It is a promising scheme in multi-source remote sensing image registration.
基金supported by the Natural Science Foundation of the Anhui Higher Education Institutions of China(Grant Nos.2023AH040149 and 2024AH051915)the Anhui Provincial Natural Science Foundation(Grant No.2208085MF168)+1 种基金the Science and Technology Innovation Tackle Plan Project of Maanshan(Grant No.2024RGZN001)the Scientific Research Fund Project of Anhui Medical University(Grant No.2023xkj122).
文摘Convolutional neural networks(CNNs)-based medical image segmentation technologies have been widely used in medical image segmentation because of their strong representation and generalization abilities.However,due to the inability to effectively capture global information from images,CNNs can easily lead to loss of contours and textures in segmentation results.Notice that the transformer model can effectively capture the properties of long-range dependencies in the image,and furthermore,combining the CNN and the transformer can effectively extract local details and global contextual features of the image.Motivated by this,we propose a multi-branch and multi-scale attention network(M2ANet)for medical image segmentation,whose architecture consists of three components.Specifically,in the first component,we construct an adaptive multi-branch patch module for parallel extraction of image features to reduce information loss caused by downsampling.In the second component,we apply residual block to the well-known convolutional block attention module to enhance the network’s ability to recognize important features of images and alleviate the phenomenon of gradient vanishing.In the third component,we design a multi-scale feature fusion module,in which we adopt adaptive average pooling and position encoding to enhance contextual features,and then multi-head attention is introduced to further enrich feature representation.Finally,we validate the effectiveness and feasibility of the proposed M2ANet method through comparative experiments on four benchmark medical image segmentation datasets,particularly in the context of preserving contours and textures.
基金The National Natural Science Foundation of China(No.61603091)。
文摘In order to improve the detection accuracy of small objects,a neighborhood fusion-based hierarchical parallel feature pyramid network(NFPN)is proposed.Unlike the layer-by-layer structure adopted in the feature pyramid network(FPN)and deconvolutional single shot detector(DSSD),where the bottom layer of the feature pyramid network relies on the top layer,NFPN builds the feature pyramid network with no connections between the upper and lower layers.That is,it only fuses shallow features on similar scales.NFPN is highly portable and can be embedded in many models to further boost performance.Extensive experiments on PASCAL VOC 2007,2012,and COCO datasets demonstrate that the NFPN-based SSD without intricate tricks can exceed the DSSD model in terms of detection accuracy and inference speed,especially for small objects,e.g.,4%to 5%higher mAP(mean average precision)than SSD,and 2%to 3%higher mAP than DSSD.On VOC 2007 test set,the NFPN-based SSD with 300×300 input reaches 79.4%mAP at 34.6 frame/s,and the mAP can raise to 82.9%after using the multi-scale testing strategy.
基金National Natural Science Foundation of China(No.61004088)Key Basic Research Foundation of Shanghai Municipal Science and Technology Commission,China(No.09JC1408000)
文摘A novel face recognition method based on fusion of spatial and frequency features was presented to improve recognition accuracy. Dual-Tree Complex Wavelet Transform derives desirable facial features to cope with the variation due to the illumination and facial expression changes. By adopting spectral regression and complex fusion technologies respectively, two improved neighborhood preserving discriminant analysis feature extraction methods were proposed to capture the face manifold structures and locality discriminatory information. Extensive experiments have been made to compare the recognition performance of the proposed method with some popular dimensionality reduction methods on ORL and Yale face databases. The results verify the effectiveness of the proposed method.
基金Open Access funding provided by the National Institutes of Health(NIH)The funding for this project was provided by NCATS Intramural Fund.
文摘Semantic segmentation plays a foundational role in biomedical image analysis, providing precise information about cellular, tissue, and organ structures in both biological and medical imaging modalities. Traditional approaches often fail in the face of challenges such as low contrast, morphological variability, and densely packed structures. Recent advancements in deep learning have transformed segmentation capabilities through the integration of fine-scale detail preservation, coarse-scale contextual modeling, and multi-scale feature fusion. This work provides a comprehensive analysis of state-of-the-art deep learning models, including U-Net variants, attention-based frameworks, and Transformer-integrated networks, highlighting innovations that improve accuracy, generalizability, and computational efficiency. Key architectural components such as convolution operations, shallow and deep blocks, skip connections, and hybrid encoders are examined for their roles in enhancing spatial representation and semantic consistency. We further discuss the importance of hierarchical and instance-aware segmentation and annotation in interpreting complex biological scenes and multiplexed medical images. By bridging methodological developments with diverse application domains, this paper outlines current trends and future directions for semantic segmentation, emphasizing its critical role in facilitating annotation, diagnosis, and discovery in biomedical research.
基金This work was supported in part by the National Natural Science Foundation of China(Grant#:82260362)in part by the National Key R&D Program of China(Grant#:2021ZD0111000)+1 种基金in part by the Key R&D Project of Hainan Province(Grant#:ZDYF2021SHFZ243)in part by the Major Science and Technology Project of Haikou(Grant#:2020-009).
文摘The precise and automatic segmentation of prostate magnetic resonance imaging(MRI)images is vital for assisting doctors in diagnosing prostate diseases.In recent years,many advanced methods have been applied to prostate segmentation,but due to the variability caused by prostate diseases,automatic segmentation of the prostate presents significant challenges.In this paper,we propose an attention-guided multi-scale feature fusion network(AGMSF-Net)to segment prostate MRI images.We propose an attention mechanism for extracting multi-scale features,and introduce a 3D transformer module to enhance global feature representation by adding it during the transition phase from encoder to decoder.In the decoder stage,a feature fusion module is proposed to obtain global context information.We evaluate our model on MRI images of the prostate acquired from a local hospital.The relative volume difference(RVD)and dice similarity coefficient(DSC)between the results of automatic prostate segmentation and ground truth were 1.21%and 93.68%,respectively.To quantitatively evaluate prostate volume on MRI,which is of significant clinical significance,we propose a unique AGMSF-Net.The essential performance evaluation and validation experiments have demonstrated the effectiveness of our method in automatic prostate segmentation.
基金supporting of the Ministry of Science and Technology MOST(Grant No.MOST 108–2221-E-150–022-MY3,MOST 110–2634-F-019–002)the National Taiwan Ocean University,China.
文摘A system for classifying four basic table tennis strokes using wearable devices and deep learning networks is proposed in this study.The wearable device consisted of a six-axis sensor,Raspberry Pi 3,and a power bank.Multiple kernel sizes were used in convolutional neural network(CNN)to evaluate their performance for extracting features.Moreover,a multiscale CNN with two kernel sizes was used to perform feature fusion at different scales in a concatenated manner.The CNN achieved recognition of the four table tennis strokes.Experimental data were obtained from20 research participants who wore sensors on the back of their hands while performing the four table tennis strokes in a laboratory environment.The data were collected to verify the performance of the proposed models for wearable devices.Finally,the sensor and multi-scale CNN designed in this study achieved accuracy and F1 scores of 99.58%and 99.16%,respectively,for the four strokes.The accuracy for five-fold cross validation was 99.87%.This result also shows that the multi-scale convolutional neural network has better robustness after fivefold cross validation.
基金support by the Aerospace Research Project of China under Grant No.020202。
文摘To explore the influence of the fusion of different features on recognition,this paper took the electromyography(EMG)signals of rectus femoris under different motions(walk,step,ramp,squat,and sitting)as samples,linear features(time-domain features(variance(VAR)and root mean square(RMS)),frequency-domain features(mean frequency(MF)and mean power frequency(MPF)),and nonlinear features(empirical mode decomposition(EMD))of the samples were extracted.Two feature fusion algorithms,the series splicing method and complex vector method,were designed,which were verified by a double hidden layer(BP)error back propagation neural network.Results show that with the increase of the types and complexity of feature fusions,the recognition rate of the EMG signal to actions is gradually improved.When the EMG signal is used in the series splicing method,the recognition rate of time-domain+frequency-domain+empirical mode decomposition(TD+FD+EMD)splicing is the highest,and the average recognition rate is 92.32%.And this rate is raised to 96.1%by using the complex vector method,and the variance of the BP system is also reduced.
基金supported by the Natural Science Foundation of Shandong Province(Nos.ZR2023LZH017 and ZR2024MF066)the Natural Science Foundation of Hebei Province(No.F2022511001)+1 种基金the Key Funding from National Natural Science Foundation of China(No.92067206)the National Natural Science Foundation of China(No.62471493).
文摘The YOLOv5 algorithm is widely used in edge computing systems for object detection.However,the limited computing resources of embedded devices and the large model size of existing deep learning based methods increase the difficulty of real-time object detection on edge devices.To address this issue,we propose a smaller,less computationally intensive,and more accurate algorithm for object detection.Multi-scale Feature Fusion-YOLO(MFF-YOLO)is built on top of the YOLOv5s framework,but it contains substantial improvements to YOLOv5s.First,we design the MFF module to improve the feature propagation path in the feature pyramid,which further integrates the semantic information from different paths of feature layers.Then,a large convolution-kernel module is used in the bottleneck.The structure enlarges the receptive field and preserves shallow semantic information,which overcomes the performance limitation arising from uneven propagation in Feature Pyramid Networks(FPN).In addition,a multi-branch downsampling method based on depthwise separable convolutions and a bottleneck structure with deformable convolutions are designed to reduce the complexity of the backbone network and minimize the real-time performance loss caused by the increased model complexity.The experimental results on PASCAL VOC and MS COCO datasets show that,compared with YOLOv5s,MFF-YOLO reduces the number of parameters by 7%and the number of FLoating point Operations Per second(FLOPs)by 11.8%.The mAP@0.5 has improved by 3.7%and 5.5%,and the mAP@0.5:0.95 has improved by 6.5%and 6.2%,respetively.Furthermore,compared with YOLOv7-tiny,PP-YOLO-tiny,and other mainstream methods,MFF-YOLO has achieved better results on multiple indicators.
