Background The use of remote photoplethysmography(rPPG)to estimate blood volume pulse in a noncontact manner has been an active research topic in recent years.Existing methods are primarily based on a singlescale regi...Background The use of remote photoplethysmography(rPPG)to estimate blood volume pulse in a noncontact manner has been an active research topic in recent years.Existing methods are primarily based on a singlescale region of interest(ROI).However,some noise signals that are not easily separated in a single-scale space can be easily separated in a multi-scale space.Also,existing spatiotemporal networks mainly focus on local spatiotemporal information and do not emphasize temporal information,which is crucial in pulse extraction problems,resulting in insufficient spatiotemporal feature modelling.Methods Here,we propose a multi-scale facial video pulse extraction network based on separable spatiotemporal convolution(SSTC)and dimension separable attention(DSAT).First,to solve the problem of a single-scale ROI,we constructed a multi-scale feature space for initial signal separation.Second,SSTC and DSAT were designed for efficient spatiotemporal correlation modeling,which increased the information interaction between the long-span time and space dimensions;this placed more emphasis on temporal features.Results The signal-to-noise ratio(SNR)of the proposed network reached 9.58dB on the PURE dataset and 6.77dB on the UBFC-rPPG dataset,outperforming state-of-the-art algorithms.Conclusions The results showed that fusing multi-scale signals yielded better results than methods based on only single-scale signals.The proposed SSTC and dimension-separable attention mechanism will contribute to more accurate pulse signal extraction.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
This paper aims to develop a nonrigid registration method of preoperative and intraoperative thoracoabdominal CT images in computer-assisted interventional surgeries for accurate tumor localization and tissue visualiz...This paper aims to develop a nonrigid registration method of preoperative and intraoperative thoracoabdominal CT images in computer-assisted interventional surgeries for accurate tumor localization and tissue visualization enhancement.However,fine structure registration of complex thoracoabdominal organs and large deformation registration caused by respiratory motion is challenging.To deal with this problem,we propose a 3D multi-scale attention VoxelMorph(MAVoxelMorph)registration network.To alleviate the large deformation problem,a multi-scale axial attention mechanism is utilized by using a residual dilated pyramid pooling for multi-scale feature extraction,and position-aware axial attention for long-distance dependencies between pixels capture.To further improve the large deformation and fine structure registration results,a multi-scale context channel attention mechanism is employed utilizing content information via adjacent encoding layers.Our method was evaluated on four public lung datasets(DIR-Lab dataset,Creatis dataset,Learn2Reg dataset,OASIS dataset)and a local dataset.Results proved that the proposed method achieved better registration performance than current state-of-the-art methods,especially in handling the registration of large deformations and fine structures.It also proved to be fast in 3D image registration,using about 1.5 s,and faster than most methods.Qualitative and quantitative assessments proved that the proposed MA-VoxelMorph has the potential to realize precise and fast tumor localization in clinical interventional surgeries.展开更多
The application of image super-resolution(SR)has brought significant assistance in the medical field,aiding doctors to make more precise diagnoses.However,solely relying on a convolutional neural network(CNN)for image...The application of image super-resolution(SR)has brought significant assistance in the medical field,aiding doctors to make more precise diagnoses.However,solely relying on a convolutional neural network(CNN)for image SR may lead to issues such as blurry details and excessive smoothness.To address the limitations,we proposed an algorithm based on the generative adversarial network(GAN)framework.In the generator network,three different sizes of convolutions connected by a residual dense structure were used to extract detailed features,and an attention mechanism combined with dual channel and spatial information was applied to concentrate the computing power on crucial areas.In the discriminator network,using InstanceNorm to normalize tensors sped up the training process while retaining feature information.The experimental results demonstrate that our algorithm achieves higher peak signal-to-noise ratio(PSNR)and structural similarity index measure(SSIM)compared to other methods,resulting in an improved visual quality.展开更多
With the rapid expansion of drone applications,accurate detection of objects in aerial imagery has become crucial for intelligent transportation,urban management,and emergency rescue missions.However,existing methods ...With the rapid expansion of drone applications,accurate detection of objects in aerial imagery has become crucial for intelligent transportation,urban management,and emergency rescue missions.However,existing methods face numerous challenges in practical deployment,including scale variation handling,feature degradation,and complex backgrounds.To address these issues,we propose Edge-enhanced and Detail-Capturing You Only Look Once(EHDC-YOLO),a novel framework for object detection in Unmanned Aerial Vehicle(UAV)imagery.Based on the You Only Look Once version 11 nano(YOLOv11n)baseline,EHDC-YOLO systematically introduces several architectural enhancements:(1)a Multi-Scale Edge Enhancement(MSEE)module that leverages multi-scale pooling and edge information to enhance boundary feature extraction;(2)an Enhanced Feature Pyramid Network(EFPN)that integrates P2-level features with Cross Stage Partial(CSP)structures and OmniKernel convolutions for better fine-grained representation;and(3)Dynamic Head(DyHead)with multi-dimensional attention mechanisms for enhanced cross-scale modeling and perspective adaptability.Comprehensive experiments on the Vision meets Drones for Detection(VisDrone-DET)2019 dataset demonstrate that EHDC-YOLO achieves significant improvements,increasing mean Average Precision(mAP)@0.5 from 33.2%to 46.1%(an absolute improvement of 12.9 percentage points)and mAP@0.5:0.95 from 19.5%to 28.0%(an absolute improvement of 8.5 percentage points)compared with the YOLOv11n baseline,while maintaining a reasonable parameter count(2.81 M vs the baseline’s 2.58 M).Further ablation studies confirm the effectiveness of each proposed component,while visualization results highlight EHDC-YOLO’s superior performance in detecting objects and handling occlusions in complex drone scenarios.展开更多
Accurate diagnosis of apple leaf diseases is crucial for improving the quality of apple production and promoting the development of the apple industry. However, apple leaf diseases do not differ significantly from ima...Accurate diagnosis of apple leaf diseases is crucial for improving the quality of apple production and promoting the development of the apple industry. However, apple leaf diseases do not differ significantly from image texture and structural information. The difficulties in disease feature extraction in complex backgrounds slow the related research progress. To address the problems, this paper proposes an improved multi-scale inverse bottleneck residual network model based on a triplet parallel attention mechanism, which is built upon ResNet-50, while improving and combining the inception module and ResNext inverse bottleneck blocks, to recognize seven types of apple leaf(including six diseases of alternaria leaf spot, brown spot, grey spot, mosaic, rust, scab, and one healthy). First, the 3×3 convolutions in some of the residual modules are replaced by multi-scale residual convolutions, the convolution kernels of different sizes contained in each branch of the multi-scale convolution are applied to extract feature maps of different sizes, and the outputs of these branches are multi-scale fused by summing to enrich the output features of the images. Second, the global layer-wise dynamic coordinated inverse bottleneck structure is used to reduce the network feature loss. The inverse bottleneck structure makes the image information less lossy when transforming from different dimensional feature spaces. The fusion of multi-scale and layer-wise dynamic coordinated inverse bottlenecks makes the model effectively balances computational efficiency and feature representation capability, and more robust with a combination of horizontal and vertical features in the fine identification of apple leaf diseases. Finally, after each improved module, a triplet parallel attention module is integrated with cross-dimensional interactions among channels through rotations and residual transformations, which improves the parallel search efficiency of important features and the recognition rate of the network with relatively small computational costs while the dimensional dependencies are improved. To verify the validity of the model in this paper, we uniformly enhance apple leaf disease images screened from the public data sets of Plant Village, Baidu Flying Paddle, and the Internet. The final processed image count is 14,000. The ablation study, pre-processing comparison, and method comparison are conducted on the processed datasets. The experimental results demonstrate that the proposed method reaches 98.73% accuracy on the adopted datasets, which is 1.82% higher than the classical ResNet-50 model, and 0.29% better than the apple leaf disease datasets before preprocessing. It also achieves competitive results in apple leaf disease identification compared to some state-ofthe-art methods.展开更多
Whole brain functional connectivity(FC)patterns obtained from resting-state functional magnetic resonance imaging(rs-fMRI)have been widely used in the diagnosis of brain disorders such as autism spectrum disorder(ASD)...Whole brain functional connectivity(FC)patterns obtained from resting-state functional magnetic resonance imaging(rs-fMRI)have been widely used in the diagnosis of brain disorders such as autism spectrum disorder(ASD).Recently,an increasing number of studies have focused on employing deep learning techniques to analyze FC patterns for brain disease classification.However,the high dimensionality of the FC features and the interpretation of deep learning results are issues that need to be addressed in the FC-based brain disease classification.In this paper,we proposed a multi-scale attention-based deep neural network(MSA-DNN)model to classify FC patterns for the ASD diagnosis.The model was implemented by adding a flexible multi-scale attention(MSA)module to the auto-encoder based backbone DNN,which can extract multi-scale features of the FC patterns and change the level of attention for different FCs by continuous learning.Our model will reinforce the weights of important FC features while suppress the unimportant FCs to ensure the sparsity of the model weights and enhance the model interpretability.We performed systematic experiments on the large multi-sites ASD dataset with both ten-fold and leaveone-site-out cross-validations.Results showed that our model outperformed classical methods in brain disease classification and revealed robust intersite prediction performance.We also localized important FC features and brain regions associated with ASD classification.Overall,our study further promotes the biomarker detection and computer-aided classification for ASD diagnosis,and the proposed MSA module is flexible and easy to implement in other classification networks.展开更多
Tea leaf picking is a crucial stage in tea production that directly influences the quality and value of the tea.Traditional tea-picking machines may compromise the quality of the tea leaves.High-quality teas are often...Tea leaf picking is a crucial stage in tea production that directly influences the quality and value of the tea.Traditional tea-picking machines may compromise the quality of the tea leaves.High-quality teas are often handpicked and need more delicate operations in intelligent picking machines.Compared with traditional image processing techniques,deep learning models have stronger feature extraction capabilities,and better generalization and are more suitable for practical tea shoot harvesting.However,current research mostly focuses on shoot detection and cannot directly accomplish end-to-end shoot segmentation tasks.We propose a tea shoot instance segmentation model based on multi-scale mixed attention(Mask2FusionNet)using a dataset from the tea garden in Hangzhou.We further analyzed the characteristics of the tea shoot dataset,where the proportion of small to medium-sized targets is 89.9%.Our algorithm is compared with several mainstream object segmentation algorithms,and the results demonstrate that our model achieves an accuracy of 82%in recognizing the tea shoots,showing a better performance compared to other models.Through ablation experiments,we found that ResNet50,PointRend strategy,and the Feature Pyramid Network(FPN)architecture can improve performance by 1.6%,1.4%,and 2.4%,respectively.These experiments demonstrated that our proposed multi-scale and point selection strategy optimizes the feature extraction capability for overlapping small targets.The results indicate that the proposed Mask2FusionNet model can perform the shoot segmentation in unstructured environments,realizing the individual distinction of tea shoots,and complete extraction of the shoot edge contours with a segmentation accuracy of 82.0%.The research results can provide algorithmic support for the segmentation and intelligent harvesting of premium tea shoots at different scales.展开更多
Nuclearmagnetic resonance imaging of breasts often presents complex backgrounds.Breast tumors exhibit varying sizes,uneven intensity,and indistinct boundaries.These characteristics can lead to challenges such as low a...Nuclearmagnetic resonance imaging of breasts often presents complex backgrounds.Breast tumors exhibit varying sizes,uneven intensity,and indistinct boundaries.These characteristics can lead to challenges such as low accuracy and incorrect segmentation during tumor segmentation.Thus,we propose a two-stage breast tumor segmentation method leveraging multi-scale features and boundary attention mechanisms.Initially,the breast region of interest is extracted to isolate the breast area from surrounding tissues and organs.Subsequently,we devise a fusion network incorporatingmulti-scale features and boundary attentionmechanisms for breast tumor segmentation.We incorporate multi-scale parallel dilated convolution modules into the network,enhancing its capability to segment tumors of various sizes through multi-scale convolution and novel fusion techniques.Additionally,attention and boundary detection modules are included to augment the network’s capacity to locate tumors by capturing nonlocal dependencies in both spatial and channel domains.Furthermore,a hybrid loss function with boundary weight is employed to address sample class imbalance issues and enhance the network’s boundary maintenance capability through additional loss.Themethod was evaluated using breast data from 207 patients at RuijinHospital,resulting in a 6.64%increase in Dice similarity coefficient compared to the benchmarkU-Net.Experimental results demonstrate the superiority of the method over other segmentation techniques,with fewer model parameters.展开更多
The end-to-end separation algorithm with superior performance in the field of speech separation has not been effectively used in music separation.Moreover,since music signals are often dual channel data with a high sa...The end-to-end separation algorithm with superior performance in the field of speech separation has not been effectively used in music separation.Moreover,since music signals are often dual channel data with a high sampling rate,how to model longsequence data and make rational use of the relevant information between channels is also an urgent problem to be solved.In order to solve the above problems,the performance of the end-to-end music separation algorithm is enhanced by improving the network structure.Our main contributions include the following:(1)A more reasonable densely connected U-Net is designed to capture the long-term characteristics of music,such as main melody,tone and so on.(2)On this basis,the multi-head attention and dualpath transformer are introduced in the separation module.Channel attention units are applied recursively on the feature map of each layer of the network,enabling the network to perform long-sequence separation.Experimental results show that after the introduction of the channel attention,the performance of the proposed algorithm has a stable improvement compared with the baseline system.On the MUSDB18 dataset,the average score of the separated audio exceeds that of the current best-performing music separation algorithm based on the time-frequency domain(T-F domain).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
Human pose estimation has received much attention from the research community because of its wide range of applications.However,current research for pose estimation is usually complex and computationally intensive,esp...Human pose estimation has received much attention from the research community because of its wide range of applications.However,current research for pose estimation is usually complex and computationally intensive,especially the feature loss problems in the feature fusion process.To address the above problems,we propose a lightweight human pose estimation network based on multi-attention mechanism(LMANet).In our method,network parameters can be significantly reduced by lightweighting the bottleneck blocks with depth-wise separable convolution on the high-resolution networks.After that,we also introduce a multi-attention mechanism to improve the model prediction accuracy,and the channel attention module is added in the initial stage of the network to enhance the local cross-channel information interaction.More importantly,we inject spatial crossawareness module in the multi-scale feature fusion stage to reduce the spatial information loss during feature extraction.Extensive experiments on COCO2017 dataset and MPII dataset show that LMANet can guarantee a higher prediction accuracy with fewer network parameters and computational effort.Compared with the highresolution network HRNet,the number of parameters and the computational complexity of the network are reduced by 67%and 73%,respectively.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
Current aero-engine life prediction areas typically focus on single-scale degradation features,and the existing methods are not comprehensive enough to capture the relationship within time series data.To address this ...Current aero-engine life prediction areas typically focus on single-scale degradation features,and the existing methods are not comprehensive enough to capture the relationship within time series data.To address this problem,we propose a novel remaining useful life(RUL)estimation method based on the attention mechanism.Our approach designs a two-layer multi-scale feature extraction module that integrates degradation features at different scales.These features are then processed in parallel by a self-attention module and a three-layer long short-term memory(LSTM)network,which together capture long-term dependencies and adaptively weigh important feature.The integration of degradation patterns from both components into the attention module enhances the model’s ability to capture long-term dependencies.Visualizing the attention module’s weight matrices further improves model interpretability.Experimental results on the C-MAPSS dataset demonstrate that our approach outperforms the existing state-of-the-art methods.展开更多
In this paper, through a multi-scale separation of the aeromagnetic anomaly by wavelet transform technique, we reprocessed the aeromagnetic data collected 20 years ago in Beijing area and analyzed the aeromagnetic ano...In this paper, through a multi-scale separation of the aeromagnetic anomaly by wavelet transform technique, we reprocessed the aeromagnetic data collected 20 years ago in Beijing area and analyzed the aeromagnetic anomaly qualitatively, integrating geological structure features in the area. In particular, we studied the spatial distributions of the two main faults called Shunyi-Liangxiang fault and Banqiao-Babaoshan-Tongxian fault, which have cut and gone through the central Beijing area striking in NE and EW directions, respectively. The influences of these two faults on the earthquakes have also been discussed briefly.展开更多
Bit-field separation is an important part of gravity and magnetic data processing.In order to extract different levels of anomaly information better,this paper introduces the dual-tree complex wavelet multi-scale sepa...Bit-field separation is an important part of gravity and magnetic data processing.In order to extract different levels of anomaly information better,this paper introduces the dual-tree complex wavelet multi-scale separation to the processing of bit-field data firstly and uses the geological model of different buried depth to ve-rify its feasibility.Finally,the dual-tree complex wavelet is applied to the aeromagnetic anomaly in Jinchuan copper nickel mining area.The results show that the method can effectively separate the anomaly information of different scales and analyze the output results with relevant geological data.展开更多
Fine-grained visual classification(FGVC)is a very challenging task due to distinguishing subcategories under the same super-category.Recent works mainly localize discriminative image regions and capture subtle inter-c...Fine-grained visual classification(FGVC)is a very challenging task due to distinguishing subcategories under the same super-category.Recent works mainly localize discriminative image regions and capture subtle inter-class differences by utilizing attention-based methods.However,at the same layer,most attention-based works only consider large-scale attention blocks with the same size as feature maps,and they ignore small-scale attention blocks that are smaller than feature maps.To distinguish subcategories,it is important to exploit small local regions.In this work,a novel multi-scale attention network(MSANet)is proposed to capture large and small regions at the same layer in fine-grained visual classification.Specifically,a novel multi-scale attention layer(MSAL)is proposed,which generates multiple groups in each feature maps to capture different-scale discriminative regions.The groups based on large-scale regions can exploit global features and the groups based on the small-scale regions can extract local subtle features.Then,a simple feature fusion strategy is utilized to fully integrate global features and local subtle features to mine information that are more conducive to FGVC.Comprehensive experiments in Caltech-UCSD Birds-200-2011(CUB),FGVC-Aircraft(AIR)and Stanford Cars(Cars)datasets show that our method achieves the competitive performances,which demonstrate its effectiveness.展开更多
To solve the problem of difficulty in identifying apple diseases in the natural environment and the low application rate of deep learning recognition networks,a lightweight ResNet(LW-ResNet)model for apple disease rec...To solve the problem of difficulty in identifying apple diseases in the natural environment and the low application rate of deep learning recognition networks,a lightweight ResNet(LW-ResNet)model for apple disease recognition is proposed.Based on the deep residual network(ResNet18),the multi-scale feature extraction layer is constructed by group convolution to realize the compression model and improve the extraction ability of different sizes of lesion features.By improving the identity mapping structure to reduce information loss.By introducing the efficient channel attention module(ECANet)to suppress noise from a complex background.The experimental results show that the average precision,recall and F1-score of the LW-ResNet on the test set are 97.80%,97.92%and 97.85%,respectively.The parameter memory is 2.32 MB,which is 94%less than that of ResNet18.Compared with the classic lightweight networks SqueezeNet and MobileNetV2,LW-ResNet has obvious advantages in recognition performance,speed,parameter memory requirement and time complexity.The proposed model has the advantages of low computational cost,low storage cost,strong real-time performance,high identification accuracy,and strong practicability,which can meet the needs of real-time identification task of apple leaf disease on resource-constrained devices.展开更多
Recently,deep learning-based image inpainting methods have made great strides in reconstructing damaged regions.However,these methods often struggle to produce satisfactory results when dealing with missing images wit...Recently,deep learning-based image inpainting methods have made great strides in reconstructing damaged regions.However,these methods often struggle to produce satisfactory results when dealing with missing images with large holes,leading to distortions in the structure and blurring of textures.To address these problems,we combine the advantages of transformers and convolutions to propose an image inpainting method that incorporates edge priors and attention mechanisms.The proposed method aims to improve the results of inpainting large holes in images by enhancing the accuracy of structure restoration and the ability to recover texture details.This method divides the inpainting task into two phases:edge prediction and image inpainting.Specifically,in the edge prediction phase,a transformer architecture is designed to combine axial attention with standard self-attention.This design enhances the extraction capability of global structural features and location awareness.It also balances the complexity of self-attention operations,resulting in accurate prediction of the edge structure in the defective region.In the image inpainting phase,a multi-scale fusion attention module is introduced.This module makes full use of multi-level distant features and enhances local pixel continuity,thereby significantly improving the quality of image inpainting.To evaluate the performance of our method.comparative experiments are conducted on several datasets,including CelebA,Places2,and Facade.Quantitative experiments show that our method outperforms the other mainstream methods.Specifically,it improves Peak Signal-to-Noise Ratio(PSNR)and Structure Similarity Index Measure(SSIM)by 1.141~3.234 db and 0.083~0.235,respectively.Moreover,it reduces Learning Perceptual Image Patch Similarity(LPIPS)and Mean Absolute Error(MAE)by 0.0347~0.1753 and 0.0104~0.0402,respectively.Qualitative experiments reveal that our method excels at reconstructing images with complete structural information and clear texture details.Furthermore,our model exhibits impressive performance in terms of the number of parameters,memory cost,and testing time.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
基金Supported by the National Natural Science Foundation of China(61903336,61976190)the Natural Science Foundation of Zhejiang Province(LY21F030015)。
文摘Background The use of remote photoplethysmography(rPPG)to estimate blood volume pulse in a noncontact manner has been an active research topic in recent years.Existing methods are primarily based on a singlescale region of interest(ROI).However,some noise signals that are not easily separated in a single-scale space can be easily separated in a multi-scale space.Also,existing spatiotemporal networks mainly focus on local spatiotemporal information and do not emphasize temporal information,which is crucial in pulse extraction problems,resulting in insufficient spatiotemporal feature modelling.Methods Here,we propose a multi-scale facial video pulse extraction network based on separable spatiotemporal convolution(SSTC)and dimension separable attention(DSAT).First,to solve the problem of a single-scale ROI,we constructed a multi-scale feature space for initial signal separation.Second,SSTC and DSAT were designed for efficient spatiotemporal correlation modeling,which increased the information interaction between the long-span time and space dimensions;this placed more emphasis on temporal features.Results The signal-to-noise ratio(SNR)of the proposed network reached 9.58dB on the PURE dataset and 6.77dB on the UBFC-rPPG dataset,outperforming state-of-the-art algorithms.Conclusions The results showed that fusing multi-scale signals yielded better results than methods based on only single-scale signals.The proposed SSTC and dimension-separable attention mechanism will contribute to more accurate pulse signal extraction.
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supported in part by the National Natural Science Foundation of China[62301374]Hubei Provincial Natural Science Foundation of China[2022CFB804]+2 种基金Hubei Provincial Education Research Project[B2022057]the Youths Science Foundation of Wuhan Institute of Technology[K202240]the 15th Graduate Education Innovation Fund of Wuhan Institute of Technology[CX2023295].
文摘This paper aims to develop a nonrigid registration method of preoperative and intraoperative thoracoabdominal CT images in computer-assisted interventional surgeries for accurate tumor localization and tissue visualization enhancement.However,fine structure registration of complex thoracoabdominal organs and large deformation registration caused by respiratory motion is challenging.To deal with this problem,we propose a 3D multi-scale attention VoxelMorph(MAVoxelMorph)registration network.To alleviate the large deformation problem,a multi-scale axial attention mechanism is utilized by using a residual dilated pyramid pooling for multi-scale feature extraction,and position-aware axial attention for long-distance dependencies between pixels capture.To further improve the large deformation and fine structure registration results,a multi-scale context channel attention mechanism is employed utilizing content information via adjacent encoding layers.Our method was evaluated on four public lung datasets(DIR-Lab dataset,Creatis dataset,Learn2Reg dataset,OASIS dataset)and a local dataset.Results proved that the proposed method achieved better registration performance than current state-of-the-art methods,especially in handling the registration of large deformations and fine structures.It also proved to be fast in 3D image registration,using about 1.5 s,and faster than most methods.Qualitative and quantitative assessments proved that the proposed MA-VoxelMorph has the potential to realize precise and fast tumor localization in clinical interventional surgeries.
文摘The application of image super-resolution(SR)has brought significant assistance in the medical field,aiding doctors to make more precise diagnoses.However,solely relying on a convolutional neural network(CNN)for image SR may lead to issues such as blurry details and excessive smoothness.To address the limitations,we proposed an algorithm based on the generative adversarial network(GAN)framework.In the generator network,three different sizes of convolutions connected by a residual dense structure were used to extract detailed features,and an attention mechanism combined with dual channel and spatial information was applied to concentrate the computing power on crucial areas.In the discriminator network,using InstanceNorm to normalize tensors sped up the training process while retaining feature information.The experimental results demonstrate that our algorithm achieves higher peak signal-to-noise ratio(PSNR)and structural similarity index measure(SSIM)compared to other methods,resulting in an improved visual quality.
文摘With the rapid expansion of drone applications,accurate detection of objects in aerial imagery has become crucial for intelligent transportation,urban management,and emergency rescue missions.However,existing methods face numerous challenges in practical deployment,including scale variation handling,feature degradation,and complex backgrounds.To address these issues,we propose Edge-enhanced and Detail-Capturing You Only Look Once(EHDC-YOLO),a novel framework for object detection in Unmanned Aerial Vehicle(UAV)imagery.Based on the You Only Look Once version 11 nano(YOLOv11n)baseline,EHDC-YOLO systematically introduces several architectural enhancements:(1)a Multi-Scale Edge Enhancement(MSEE)module that leverages multi-scale pooling and edge information to enhance boundary feature extraction;(2)an Enhanced Feature Pyramid Network(EFPN)that integrates P2-level features with Cross Stage Partial(CSP)structures and OmniKernel convolutions for better fine-grained representation;and(3)Dynamic Head(DyHead)with multi-dimensional attention mechanisms for enhanced cross-scale modeling and perspective adaptability.Comprehensive experiments on the Vision meets Drones for Detection(VisDrone-DET)2019 dataset demonstrate that EHDC-YOLO achieves significant improvements,increasing mean Average Precision(mAP)@0.5 from 33.2%to 46.1%(an absolute improvement of 12.9 percentage points)and mAP@0.5:0.95 from 19.5%to 28.0%(an absolute improvement of 8.5 percentage points)compared with the YOLOv11n baseline,while maintaining a reasonable parameter count(2.81 M vs the baseline’s 2.58 M).Further ablation studies confirm the effectiveness of each proposed component,while visualization results highlight EHDC-YOLO’s superior performance in detecting objects and handling occlusions in complex drone scenarios.
基金supported in part by the General Program Hunan Provincial Natural Science Foundation of 2022,China(2022JJ31022)the Undergraduate Education Reform Project of Hunan Province,China(HNJG-20210532)the National Natural Science Foundation of China(62276276)。
文摘Accurate diagnosis of apple leaf diseases is crucial for improving the quality of apple production and promoting the development of the apple industry. However, apple leaf diseases do not differ significantly from image texture and structural information. The difficulties in disease feature extraction in complex backgrounds slow the related research progress. To address the problems, this paper proposes an improved multi-scale inverse bottleneck residual network model based on a triplet parallel attention mechanism, which is built upon ResNet-50, while improving and combining the inception module and ResNext inverse bottleneck blocks, to recognize seven types of apple leaf(including six diseases of alternaria leaf spot, brown spot, grey spot, mosaic, rust, scab, and one healthy). First, the 3×3 convolutions in some of the residual modules are replaced by multi-scale residual convolutions, the convolution kernels of different sizes contained in each branch of the multi-scale convolution are applied to extract feature maps of different sizes, and the outputs of these branches are multi-scale fused by summing to enrich the output features of the images. Second, the global layer-wise dynamic coordinated inverse bottleneck structure is used to reduce the network feature loss. The inverse bottleneck structure makes the image information less lossy when transforming from different dimensional feature spaces. The fusion of multi-scale and layer-wise dynamic coordinated inverse bottlenecks makes the model effectively balances computational efficiency and feature representation capability, and more robust with a combination of horizontal and vertical features in the fine identification of apple leaf diseases. Finally, after each improved module, a triplet parallel attention module is integrated with cross-dimensional interactions among channels through rotations and residual transformations, which improves the parallel search efficiency of important features and the recognition rate of the network with relatively small computational costs while the dimensional dependencies are improved. To verify the validity of the model in this paper, we uniformly enhance apple leaf disease images screened from the public data sets of Plant Village, Baidu Flying Paddle, and the Internet. The final processed image count is 14,000. The ablation study, pre-processing comparison, and method comparison are conducted on the processed datasets. The experimental results demonstrate that the proposed method reaches 98.73% accuracy on the adopted datasets, which is 1.82% higher than the classical ResNet-50 model, and 0.29% better than the apple leaf disease datasets before preprocessing. It also achieves competitive results in apple leaf disease identification compared to some state-ofthe-art methods.
基金This work was supported by the National Natural Science Foundation of China(No.61906006).
文摘Whole brain functional connectivity(FC)patterns obtained from resting-state functional magnetic resonance imaging(rs-fMRI)have been widely used in the diagnosis of brain disorders such as autism spectrum disorder(ASD).Recently,an increasing number of studies have focused on employing deep learning techniques to analyze FC patterns for brain disease classification.However,the high dimensionality of the FC features and the interpretation of deep learning results are issues that need to be addressed in the FC-based brain disease classification.In this paper,we proposed a multi-scale attention-based deep neural network(MSA-DNN)model to classify FC patterns for the ASD diagnosis.The model was implemented by adding a flexible multi-scale attention(MSA)module to the auto-encoder based backbone DNN,which can extract multi-scale features of the FC patterns and change the level of attention for different FCs by continuous learning.Our model will reinforce the weights of important FC features while suppress the unimportant FCs to ensure the sparsity of the model weights and enhance the model interpretability.We performed systematic experiments on the large multi-sites ASD dataset with both ten-fold and leaveone-site-out cross-validations.Results showed that our model outperformed classical methods in brain disease classification and revealed robust intersite prediction performance.We also localized important FC features and brain regions associated with ASD classification.Overall,our study further promotes the biomarker detection and computer-aided classification for ASD diagnosis,and the proposed MSA module is flexible and easy to implement in other classification networks.
基金This research was supported by the National Natural Science Foundation of China No.62276086the National Key R&D Program of China No.2022YFD2000100Zhejiang Provincial Natural Science Foundation of China under Grant No.LTGN23D010002.
文摘Tea leaf picking is a crucial stage in tea production that directly influences the quality and value of the tea.Traditional tea-picking machines may compromise the quality of the tea leaves.High-quality teas are often handpicked and need more delicate operations in intelligent picking machines.Compared with traditional image processing techniques,deep learning models have stronger feature extraction capabilities,and better generalization and are more suitable for practical tea shoot harvesting.However,current research mostly focuses on shoot detection and cannot directly accomplish end-to-end shoot segmentation tasks.We propose a tea shoot instance segmentation model based on multi-scale mixed attention(Mask2FusionNet)using a dataset from the tea garden in Hangzhou.We further analyzed the characteristics of the tea shoot dataset,where the proportion of small to medium-sized targets is 89.9%.Our algorithm is compared with several mainstream object segmentation algorithms,and the results demonstrate that our model achieves an accuracy of 82%in recognizing the tea shoots,showing a better performance compared to other models.Through ablation experiments,we found that ResNet50,PointRend strategy,and the Feature Pyramid Network(FPN)architecture can improve performance by 1.6%,1.4%,and 2.4%,respectively.These experiments demonstrated that our proposed multi-scale and point selection strategy optimizes the feature extraction capability for overlapping small targets.The results indicate that the proposed Mask2FusionNet model can perform the shoot segmentation in unstructured environments,realizing the individual distinction of tea shoots,and complete extraction of the shoot edge contours with a segmentation accuracy of 82.0%.The research results can provide algorithmic support for the segmentation and intelligent harvesting of premium tea shoots at different scales.
基金funded by the National Natural Foundation of China under Grant No.61172167the Science Fund Project of Heilongjiang Province(LH2020F035).
文摘Nuclearmagnetic resonance imaging of breasts often presents complex backgrounds.Breast tumors exhibit varying sizes,uneven intensity,and indistinct boundaries.These characteristics can lead to challenges such as low accuracy and incorrect segmentation during tumor segmentation.Thus,we propose a two-stage breast tumor segmentation method leveraging multi-scale features and boundary attention mechanisms.Initially,the breast region of interest is extracted to isolate the breast area from surrounding tissues and organs.Subsequently,we devise a fusion network incorporatingmulti-scale features and boundary attentionmechanisms for breast tumor segmentation.We incorporate multi-scale parallel dilated convolution modules into the network,enhancing its capability to segment tumors of various sizes through multi-scale convolution and novel fusion techniques.Additionally,attention and boundary detection modules are included to augment the network’s capacity to locate tumors by capturing nonlocal dependencies in both spatial and channel domains.Furthermore,a hybrid loss function with boundary weight is employed to address sample class imbalance issues and enhance the network’s boundary maintenance capability through additional loss.Themethod was evaluated using breast data from 207 patients at RuijinHospital,resulting in a 6.64%increase in Dice similarity coefficient compared to the benchmarkU-Net.Experimental results demonstrate the superiority of the method over other segmentation techniques,with fewer model parameters.
基金National Natural Science Foundation of China,Grant/Award Number:62071039Beijing Natural Science Foundation,Grant/Award Number:L223033。
文摘The end-to-end separation algorithm with superior performance in the field of speech separation has not been effectively used in music separation.Moreover,since music signals are often dual channel data with a high sampling rate,how to model longsequence data and make rational use of the relevant information between channels is also an urgent problem to be solved.In order to solve the above problems,the performance of the end-to-end music separation algorithm is enhanced by improving the network structure.Our main contributions include the following:(1)A more reasonable densely connected U-Net is designed to capture the long-term characteristics of music,such as main melody,tone and so on.(2)On this basis,the multi-head attention and dualpath transformer are introduced in the separation module.Channel attention units are applied recursively on the feature map of each layer of the network,enabling the network to perform long-sequence separation.Experimental results show that after the introduction of the channel attention,the performance of the proposed algorithm has a stable improvement compared with the baseline system.On the MUSDB18 dataset,the average score of the separated audio exceeds that of the current best-performing music separation algorithm based on the time-frequency domain(T-F domain).
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the National Natural Science Foundation of China(Nos.61775139,62072126,61772164,and 61872242)。
文摘Human pose estimation has received much attention from the research community because of its wide range of applications.However,current research for pose estimation is usually complex and computationally intensive,especially the feature loss problems in the feature fusion process.To address the above problems,we propose a lightweight human pose estimation network based on multi-attention mechanism(LMANet).In our method,network parameters can be significantly reduced by lightweighting the bottleneck blocks with depth-wise separable convolution on the high-resolution networks.After that,we also introduce a multi-attention mechanism to improve the model prediction accuracy,and the channel attention module is added in the initial stage of the network to enhance the local cross-channel information interaction.More importantly,we inject spatial crossawareness module in the multi-scale feature fusion stage to reduce the spatial information loss during feature extraction.Extensive experiments on COCO2017 dataset and MPII dataset show that LMANet can guarantee a higher prediction accuracy with fewer network parameters and computational effort.Compared with the highresolution network HRNet,the number of parameters and the computational complexity of the network are reduced by 67%and 73%,respectively.
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supported by the National Key Research and Development Program of China (2023YFB4302403)the Research and Practical Innovation Program of NUAA (xcxjh20230735)。
文摘Current aero-engine life prediction areas typically focus on single-scale degradation features,and the existing methods are not comprehensive enough to capture the relationship within time series data.To address this problem,we propose a novel remaining useful life(RUL)estimation method based on the attention mechanism.Our approach designs a two-layer multi-scale feature extraction module that integrates degradation features at different scales.These features are then processed in parallel by a self-attention module and a three-layer long short-term memory(LSTM)network,which together capture long-term dependencies and adaptively weigh important feature.The integration of degradation patterns from both components into the attention module enhances the model’s ability to capture long-term dependencies.Visualizing the attention module’s weight matrices further improves model interpretability.Experimental results on the C-MAPSS dataset demonstrate that our approach outperforms the existing state-of-the-art methods.
基金National Development and Reform Commission Project ″Experimental Detection of Urban Active Faults″ (2004-1138).
文摘In this paper, through a multi-scale separation of the aeromagnetic anomaly by wavelet transform technique, we reprocessed the aeromagnetic data collected 20 years ago in Beijing area and analyzed the aeromagnetic anomaly qualitatively, integrating geological structure features in the area. In particular, we studied the spatial distributions of the two main faults called Shunyi-Liangxiang fault and Banqiao-Babaoshan-Tongxian fault, which have cut and gone through the central Beijing area striking in NE and EW directions, respectively. The influences of these two faults on the earthquakes have also been discussed briefly.
基金the National Key R&D Program of China(No.2016YFC0600505).
文摘Bit-field separation is an important part of gravity and magnetic data processing.In order to extract different levels of anomaly information better,this paper introduces the dual-tree complex wavelet multi-scale separation to the processing of bit-field data firstly and uses the geological model of different buried depth to ve-rify its feasibility.Finally,the dual-tree complex wavelet is applied to the aeromagnetic anomaly in Jinchuan copper nickel mining area.The results show that the method can effectively separate the anomaly information of different scales and analyze the output results with relevant geological data.
基金jointly supported by the National Science and Technology Major Project(2022ZD0117103)the National Natural Science Foundations of China(62272364)+2 种基金the provincial Key Research and Development Program of Shaanxi(2024GH-ZDXM-47)the Research Project on Higher Education Teaching Reform of Shaanxi Province(23JG003)the Natural Science Basic Research Program of Shaanxi(2024JC-YBQN0639).
文摘Fine-grained visual classification(FGVC)is a very challenging task due to distinguishing subcategories under the same super-category.Recent works mainly localize discriminative image regions and capture subtle inter-class differences by utilizing attention-based methods.However,at the same layer,most attention-based works only consider large-scale attention blocks with the same size as feature maps,and they ignore small-scale attention blocks that are smaller than feature maps.To distinguish subcategories,it is important to exploit small local regions.In this work,a novel multi-scale attention network(MSANet)is proposed to capture large and small regions at the same layer in fine-grained visual classification.Specifically,a novel multi-scale attention layer(MSAL)is proposed,which generates multiple groups in each feature maps to capture different-scale discriminative regions.The groups based on large-scale regions can exploit global features and the groups based on the small-scale regions can extract local subtle features.Then,a simple feature fusion strategy is utilized to fully integrate global features and local subtle features to mine information that are more conducive to FGVC.Comprehensive experiments in Caltech-UCSD Birds-200-2011(CUB),FGVC-Aircraft(AIR)and Stanford Cars(Cars)datasets show that our method achieves the competitive performances,which demonstrate its effectiveness.
基金funded by the Science and Technology Development Program of Jilin Province(20190301024NY)the Precision Agriculture and Big Data Engineering Research Center of Jilin Province(2020C005).
文摘To solve the problem of difficulty in identifying apple diseases in the natural environment and the low application rate of deep learning recognition networks,a lightweight ResNet(LW-ResNet)model for apple disease recognition is proposed.Based on the deep residual network(ResNet18),the multi-scale feature extraction layer is constructed by group convolution to realize the compression model and improve the extraction ability of different sizes of lesion features.By improving the identity mapping structure to reduce information loss.By introducing the efficient channel attention module(ECANet)to suppress noise from a complex background.The experimental results show that the average precision,recall and F1-score of the LW-ResNet on the test set are 97.80%,97.92%and 97.85%,respectively.The parameter memory is 2.32 MB,which is 94%less than that of ResNet18.Compared with the classic lightweight networks SqueezeNet and MobileNetV2,LW-ResNet has obvious advantages in recognition performance,speed,parameter memory requirement and time complexity.The proposed model has the advantages of low computational cost,low storage cost,strong real-time performance,high identification accuracy,and strong practicability,which can meet the needs of real-time identification task of apple leaf disease on resource-constrained devices.
基金supported in part by the National Natural Science Foundation of China under Grant 62062061/in part by the Major Project Cultivation Fund of Xizang Minzu University under Grant 324112300447.
文摘Recently,deep learning-based image inpainting methods have made great strides in reconstructing damaged regions.However,these methods often struggle to produce satisfactory results when dealing with missing images with large holes,leading to distortions in the structure and blurring of textures.To address these problems,we combine the advantages of transformers and convolutions to propose an image inpainting method that incorporates edge priors and attention mechanisms.The proposed method aims to improve the results of inpainting large holes in images by enhancing the accuracy of structure restoration and the ability to recover texture details.This method divides the inpainting task into two phases:edge prediction and image inpainting.Specifically,in the edge prediction phase,a transformer architecture is designed to combine axial attention with standard self-attention.This design enhances the extraction capability of global structural features and location awareness.It also balances the complexity of self-attention operations,resulting in accurate prediction of the edge structure in the defective region.In the image inpainting phase,a multi-scale fusion attention module is introduced.This module makes full use of multi-level distant features and enhances local pixel continuity,thereby significantly improving the quality of image inpainting.To evaluate the performance of our method.comparative experiments are conducted on several datasets,including CelebA,Places2,and Facade.Quantitative experiments show that our method outperforms the other mainstream methods.Specifically,it improves Peak Signal-to-Noise Ratio(PSNR)and Structure Similarity Index Measure(SSIM)by 1.141~3.234 db and 0.083~0.235,respectively.Moreover,it reduces Learning Perceptual Image Patch Similarity(LPIPS)and Mean Absolute Error(MAE)by 0.0347~0.1753 and 0.0104~0.0402,respectively.Qualitative experiments reveal that our method excels at reconstructing images with complete structural information and clear texture details.Furthermore,our model exhibits impressive performance in terms of the number of parameters,memory cost,and testing time.
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.
