检索结果-南通市图书馆

Automatic channel pruning of U-Net Applied to Retinal Blood Vessels Segmentation

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

Automatic Channel Pruning of U-Net Applied to Retinal Blood ...

Effective Model Compression via Stage-wise pruning

第43届中国控制会议

作者： Xiajin Mei Zhenbin Zhu Mingxu Sun Yuliang Ma College of Automation Hangzhou Dianzi University Jinan Key Laboratory of Rehabilitation and Evaluation of Motor Dysfunction The People's Hospital of Huaiyin Shandong BetR Medical Technology Co. Ltd.

The U-Net model is widely utilized in medical image segmentation, yielding satisfactory results. However, retinal vascular image segmentation encounters an issue of model parameter redundancy. In our study, we propose a U-Net network pruning method utilizing a genetic optimization algorithm. The objective is to streamline the U-Net network structure by channel pruning, reducing the network's parameter count. Unlike conventional channel pruning methods, our algorithm prunes channels in each U-Net layer, avoiding selective pruning of "non-important" channels. Experimental evaluation on the DRIVE dataset produced a pruned model structure, later validated on STARE and CHASEB1 datasets to confirm accuracy. Analysis of experimental data indicates the pruned model is more lightweight than the original U-Net, with a 37.69% reduction in parameters and a 22.5% reduction in channels. Segmentation results show no significant differences. The proposed algorithm effectively trims U-Net network parameters without negative impacts on segmentation. Compared to manually pruned structures, our model demonstrates advantages in retinal image segmentation.

关键词： channel pruning Compression rate Genetic algorithm Retinal blood vessels segmentation U-Net

来源： cnki会议评论

在线全文

cnki会议

学校读者我要写书评

暂无评论

Machine Intelligence Research 2023年第6期20卷 937-951页

作者： Ming-Yang Zhang Xin-Yi Yu Lin-Lin Ou School of Information Engineering Zhejiang University of TechnologyHangzhou310000China

Automated machine learning(AutoML)pruning methods aim at searching for a pruning strategy automatically to reduce the computational complexity of deep convolutional neural networks(deep CNNs).However,some previous work found that the results of many Auto-ML pruning methods cannot even surpass the results of the uniformly pruning *** this paper,the ineffectiveness of Auto-ML pruning,which is caused by unfull and unfair training of the supernet,is shown.A deep supernet suffers from unfull training because it contains too many *** overcome the unfull training,a stage-wise pruning(SWP)method is proposed,which splits a deep supernet into several stage-wise supernets to reduce the candidate number and utilize inplace distillation to supervise the stage ***,a wide supernet is hit by unfair training since the sampling probability of each channel is ***,the fullnet and the tinynet are sampled in each training iteration to ensure that each channel can be ***,the proxy performance of the subnets trained with SWP is closer to the actual performance than that of most of the previous AutoML pruning ***,experiments show that SWP achieves the state-of-the-art in both CIFAR-10 and ImageNet under the mobile setting.

关键词： Automated machine learning(AutoML) channel pruning model compression distillation convolutional neural networks(CNN)

A Real-Time Small Target Vehicle Detection Algorithm with an Improved YOLOv5m Network Model

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

Computers, Materials & Continua 2024年第1期78卷 303-327页

作者： Yaoyao Du Xiangkui Jiang School of Automation Xi’an University of Posts and TelecommunicationsXi’an710121China

To address the challenges of high complexity,poor real-time performance,and low detection rates for small target vehicles in existing vehicle object detection algorithms,this paper proposes a real-time lightweight architecture based on You Only Look Once(YOLO)***,a lightweight upsampling operator called Content-Aware Reassembly of Features(CARAFE)is introduced in the feature fusion layer of the network to maximize the extraction of deep-level features for small target vehicles,reducing the missed detection rate and false detection ***,a new prediction layer for tiny targets is added,and the feature fusion network is redesigned to enhance the detection capability for small ***,this paper applies L1 regularization to train the improved network,followed by pruning and fine-tuning operations to remove redundant channels,reducing computational and parameter complexity and enhancing the detection efficiency of the *** is conducted on the VisDrone2019-DET *** experimental results show that the proposed algorithmreduces parameters and computation by 63.8% and 65.8%,*** average detection accuracy improves by 5.15%,and the detection speed reaches 47 images per second,satisfying real-time *** with existing approaches,including YOLOv5m and classical vehicle detection algorithms,our method achieves higher accuracy and faster speed for real-time detection of small target vehicles in edge computing.

关键词： Vehicle detection YOLOv5m small target channel pruning CARAFE

Pedestrian and Vehicle Detection Based on pruning YOLOv4 with Cloud-Edge Collaboration

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

Computer Modeling in Engineering & Sciences 2023年第11期137卷 2025-2047页

作者： Huabin Wang Ruichao Mo Yuping Chen Weiwei Lin Minxian Xu Bo Liu School of Computer Science and Engineering Huizhou UniversityHuizhou516007China School of Computer Science and Engineering South China University of TechnologyGuangzhou510006China School of Computer Science and Technology South China Normal UniversityGuangzhou510631China Department of New Network Technologies Peng Cheng LaboratoryShenzhen518066China Shenzhen Institute of Advanced Technology Chinese Academy of SciencesShenzhen518055China

Nowadays,the rapid development of edge computing has driven an increasing number of deep learning applications deployed at the edge of the network,such as pedestrian and vehicle detection,to provide efficient intelligent services to mobile ***,as the accuracy requirements continue to increase,the components of deep learning models for pedestrian and vehicle detection,such as YOLOv4,become more sophisticated and the computing resources required for model training are increasing dramatically,which in turn leads to significant challenges in achieving effective deployment on resource-constrained edge devices while ensuring the high accuracy *** addressing this challenge,a cloud-edge collaboration-based pedestrian and vehicle detection framework is proposed in this paper,which enables sufficient training of models by utilizing the abundant computing resources in the cloud,and then deploying the well-trained models on edge devices,thus reducing the computing resource requirements for model training on edge ***,to reduce the size of the model deployed on edge devices,an automatic pruning method combines the convolution layer and BN layer is proposed to compress the pedestrian and vehicle detection model *** results show that the framework proposed in this paper is able to deploy the pruned model on a real edge device,Jetson TX2,with 6.72 times higher ***,the channel pruning reduces the volume and the number of parameters to 96.77%for the model,and the computing amount is reduced to 81.37%.

关键词： Pedestrian and vehicle detection YOLOv4 channel pruning cloud-edge collaboration

Improved Lightweight YOLOv5 Based on GhostNet and Its Application in Target Board Detection for Ladle Nozzle Positioning

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

Improved Lightweight YOLOv5 Based on GhostNet and Its Applic...

A Lightweight Two-stream Spatial Temporal Convolutional Neural Network for Human Actions Recognition

第43届中国控制会议

作者： Fengfei Si Yiming Fang Hao Ju Dongliang Zhao Jingliang Xin Key Lab of Industrial Computer Control Engineering of Hebei Province Yanshan University Engineering Research Center of the Education Ministry for Intelligent Control System and Intelligent Equipment Yanshan University Tangshan Branch HBIS Company Limited

The detection of the target board of ladle nozzle positioning is the basis of automatic loading and unloading of ladle nozzle by ladle robot in continuous casting. Due to the complex situation of high temperature, dust, backlight and so on, the target board of ladle nozzle positioning is prone to false detection or low detection accuracy, and it is difficult to take into account the contradiction between rapidity and detection accuracy. Therefore, this paper proposes an improved lightweight algorithm GVP-YOLOv5(YOLOv5 with GhostNet, VoVGSCSP and pruning), which takes into account both detection accuracy and rapidity, to detect the target board for ladle nozzle positioning. Firstly, Ghost module is used to improve the backbone network of YOLOv5 as the backbone feature extraction network to reduce the number of parameters and calculation of the backbone. Then in the neck, the GSConv module is used to design a thin neck structure, which can ensure the multi-scale feature fusion capability and further reduce the complexity of the network. Finally, the model pruning technology is used to prune the whole model, and the channels with small scaling factor are removed to compress the model to the greatest extent, so as to meet the requirements of industrial field application deployment. The experimental results show that the lightweight GVP-YOLOv5 algorithm proposed in this paper can complete the real-time detection of the target board in complex industrial scenarios. Compared with the original YOLOv5s algorithm, the number of parameters in the model is reduced by 82.6%, and the size of the model is significantly reduced to 19.4%. At the same time, the detection accuracy is maintained at more than 95% on mAP@50:95 and the detection speed is effectively improved.

关键词： Target board detection Ladle nozzle positioning YOLOv5 lightweight GhostNet thin neck structure channel pruning

来源： cnki会议评论

在线全文

cnki会议

学校读者我要写书评

暂无评论

A Lightweight Two-stream Spatial Temporal Convolutional Neur...