基于YOLOv5的黑色素瘤图像检测仿真  

Simulation of Melanoma Image Detection Based on YOLOv5

作　　者：刘勇志 万方[1] 雷光波[1] 徐丽[1] LIU Yong-zhi;WAN Fang;LEI Guang-bo;XU Li(School of Computer Science,Hubei University of Technology,Wuhan Hubei 430068,China)

机构地区：[1]湖北工业大学计算机学院,湖北武汉430068

出　　处：《计算机仿真》2024年第4期214-218,516,共6页Computer Simulation

基　　金：湖北省教育厅指导性项目(B2021070)。

摘　　要：针对黑色素瘤疾病在临床上存在检测准确率低以及人为主观性太强等问题,提出一种改进的YOLOv5目标检测模型BiC-YOLOv5。首先设计了一种双向特征提取网络BiFPN-L3替换原模型中的特征提取网络FPN,针对不同分辨率下的特征,使用多尺度特征融合的方式提取特征;其次,在骨干网络中融合CBAM注意力模块,设计了一种C3CBAM模块从通道与空间两个层面捕获特征信息以提升检测精度;最后,使用DIOU_loss损失函数,进一步提高模型的检测精度。通过仿真对比实现,BiC-YOLOv5的mAP值达到95.2%,相较原YOLOv5模型,精确度提高了5.2%,召回率提高了4.9%,mAP值提高了5.8%,可以有效的协助临床医学对黑色素瘤进行诊断。To address the problems of low detection accuracy and too much human subjectivity in melanoma disease in clinical practice,an improved YOLOv5 target detection model BiC-YOLOv5 is proposed.Firstly,a bidirectional feature extraction network BiFPN-L3 was designed to replace the feature extraction network FPN in the original model,and a multi-scale feature fusion was used to extract features at different resolutions.Second,a CBAM attention module was fused in the backbone network,and a C3CBAM module was designed to capture feature information from both channel and space levels to improve detection accuracy;Finally,the DIOU_loss loss function was used to further improve the detection accuracy of the model.Through simulation comparison,the mAP value of BiC-YOLOv5 reached 95.2%,which is an improvement of 5.2%in accuracy,4.9%in recall,and 5.8%in mAP value compared to the original YOLOv5 model.This can effectively assist clinical medicine in diagnosing melanoma.

关 键 词：特征提取网络 注意力机制 黑色素瘤 皮肤镜图像 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]