YOLOv5-CCE:一种基于CA和EIoU的目标检测算法  

YOLOv5-CCE:an Object Detection Algorithm Based on CA and EIoU

作　　者：王军 黄博文 蔡景贵 WANG JUN;HUANG Bowen;CAI Jinggui(School of Computer Science and Technology,Shenyang University of Chemical Technology,Shenyang 110142,China;Liaoning Provincial Key Laboratory of Chemical Process Industry and Intelligent Technology,Shenyang 110142,China)

机构地区：[1]沈阳化工大学计算机科学与技术学院,沈阳110142 [2]辽宁省化工过程工业智能化技术重点实验室,沈阳110142

出　　处：《火力与指挥控制》2024年第9期90-96,103,共8页Fire Control & Command Control

基　　金：辽宁省自然科学基金(2022-MS-291);辽宁省教育厅科研基金(LJ2020024);中国高校产学研创新基金(2021LD06009);辽宁省教育厅科研基金资助项目(LJKMZ20220781)。

摘　　要：为了减少YOLOv5模型在复杂环境下的误检率和漏检率,提出一种基于CA(Coordinate Attention)和EIoU(Efficient Intersection over Union)的目标检测模型YOLOv5-CCE。首先向Neck网络中的部分C3_2模块中嵌入坐标注意力机制CA,增强模型对特征的提取能力;其次为提高回归精度,提出一种基于Focal EIoU Loss改进的Focal CEIoU Loss。实验结果表明,在PASCAL VOC 2007+2012数据集上,YOLOv5-CCE模型在参数量和计算量基本保持不变的情况下,相较于原模型mAP@0.5、mAP@0.5:0.95和准确率分别提升了1.4%、1.3%和3.7%,因此,YOLOv5-CCE模型可以更好地适应复杂环境下的目标检测任务。In order to reduce the false detection rate and missed detection rate of the YOLOv5 model in complex environments,a target detection model YOLOv5-CCE based on CA(Coordinate Attention)and EIoU(Efficient Intersection over Union)is proposed.Firstly,the coordinate attention mechanism CA is embedded in partial C3_2 module in the Neck network to enhance the feature extraction ability of the model;secondly,to improve the regression accuracy,an improved Focal CEIoU Loss based on Focal EIoU Loss is proposed.The experimental results show that on the PASCAL VOC 2007+2012 data set,the YOLOv5-CCE model maintains the parameters and calculations basically unchanged,compared with the original model of mAP@0.5 and mAP@0.5:0.95 and the accuracy rate respectively,its accuracy rate has increased by 1.4%,1.3%and 3.7%respectively.Therefore,the YOLOv5-CCE model can better adapt to the target detection task in complex envi⁃ronments.

关 键 词：YOLOv5算法 EIoU Focal Loss CA注意力机制 目标检测 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]