基于改进YOLOv5的视网膜黄斑病变分类检测算法  

Classification and Detection Algorithm of Macular Disease Based on Improved YOLOv5

作　　者：王楠楠 吴其洲[1] 王召巴[1] 金永[1] WANG Nannan;WU Qizhou;WANG Zhaoba;JIN Yong(School of Information and Communication Engineering,North University of China,Taiyuan 030051,China)

机构地区：[1]中北大学信息与通信工程学院,山西太原030051

出　　处：《测试技术学报》2025年第2期130-137,共8页Journal of Test and Measurement Technology

基　　金：山西省自然科学基金资助项目(202103021224202);山西省归国留学人员科研基金资助项目(20210038)。

摘　　要：视网膜黄斑病变是失明的重要原因之一,人工筛查容易出现误诊,而基于深度学习的自动诊断有助于早期的检测和治疗。提出一种基于改进YOLOv5的黄斑病变分类检测算法,针对黄斑病变图像细微特征融合不充分的问题,将YOLOv5颈部的PANet特征融合模块替换为加权双向特征金字塔网络,实现高效的多尺度特征融合,以获得更好的黄斑病变细节特征;针对小目标病变检测能力差的问题,在模型中引入SK注意力机制,通过自适应地调整感受野,增强对黄斑病变区域特征的捕获。对比实验证明所提算法可将小目标检测准确率由原来的91.9%提升至94.2%,全类平均准确率由原来的93.4%提升到96.6%,且在相同条件下,该算法的表现优于其他目标检测网络模型。Macular disease is one of the important causes of blindness,manual screening is prone to misdiagnosis,and automatic diagnosis based on deep learning helps early detection and treatment.A classification and detection algorithm for macular lesions based on improved YOLOv5 was proposed.To solve the problem of insufficient fusion of fine features in macular lesions images,a weighted bidirectional feature pyramid network replaced the PANet feature fusion module of YOLOv5 neck to achieve efficient multiscale feature fusion to obtain better-detailed features of macular lesions.To solve the problem of poor detection ability of small target lesions,the SK attention mechanism was introduced into the model to enhance the capture of regional features of macular lesions by adjusting the receptive field adaptively.Comparative experiments show that the proposed algorithm can improve the detection accuracy of small targets from 91.9%to 94.2%,and the average accuracy of the whole class from 93.4%to 96.6%.Moreover,under the same conditions,the algorithm performs better than other target detection network models.

关 键 词：目标检测 视网膜黄斑病变 加权双向特征金字塔网络 注意力机制 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]