摘要
目前视网膜静脉阻塞检测几乎都依靠人工识别,存在检测精度低、速度慢等问题。针对这一状况,在YOLOv5的基础上提出了一种基于空间、通道注意力增强(Channel and Spatial Attention Enhancement,CASE)的CSAE-YOLO网络。首先将空间、通道注意力机制与E-ELAN结合,组成新的CSAE-ELAN主干网络。CSAE-ELAN具有高效的通道间依赖关系和感受野更广的空间注意力,提升了提取特征信息的能力;其次使用改进的空间金字塔池化,加强对不同尺度病灶的检测能力。实验结果表明:改进后网络的mAP为87%,检测速度为30帧/s,远优于其他算法,具有精度高、参数规模小、识别速度快的特点,为RVO检测提供了一种新的AI辅助诊断方法。
At present,the detection of retinal vein occlusion(RVO)almost depends on manual recognition,and it has problems such as low detection accuracy and slow detection speed.In response to this situation,on the basis of YOLOv5,a CSAE-YOLO network based on channel and spatial attention enhancement(CASE)is proposed.Firstly,the channel and spatial attention mechanisms are combined with E-ELAN to form a new CSAE-ELAN backbone network.CSAE-ELAN has efficient inter channel dependency and wider spatial attention of receptive field,which improves the ability to extract feature information.Secondly,the improved spatial pyramid pooling is used to en-hance the detection ability of lesions at different scales.The experimental results show that the mAP of the improved network is 87%,and the detection speed is 30 frame/s,which is far superior to other algorithms.It has the characteristics of high accuracy,small param-eter scale,and fast recognition speed.It provides a new AI aided diagnosis method for RVO detection.
作者
吕辉
李沛洋
董帆
刘晓青
LÜHui;LI Peiyang;DONG Fan;LIU Xiaoqing(College of Electrical Engineering,Henan Polytechnic University,Jiaozuo Henan 454000,China;School of Mechanical and Electrical Engineering,Zhoukou Normal University,Zhoukou Henan 466001,China)
出处
《电子器件》
2025年第2期372-379,共8页
Chinese Journal of Electron Devices
基金
河南省自然科学基金项目(242300420283)
河南省高校基本科研业务费专项资金资助项目(NSFRF240819)。
