基于多聚合和多尺度的YOLOV5水下目标检测算法  

YOLOV5 Underwater Object Detection Algorithm Based on Multi-aggregation and Multi-scale

作　　者：黄明发 黄文明 肖雁南 温雅媛 邓珍荣 HUANG Mingfa;HUANG Wenming;XIAO Yannan;WEN Yayuan;DENG Zhenrong(School of Computer and Information Security,Guilin University of Electronic Technology,Guilin 541004;College of Electronic Engineering,Guangxi Normal University,Guilin 541004)

机构地区：[1]桂林电子科技大学计算机与信息安全学院,桂林541004 [2]广西师范大学电子工程学院,桂林541004

出　　处：《计算机与数字工程》2024年第11期3197-3201,3309,共6页Computer & Digital Engineering

基　　金：国家自然科学基金项目(编号:6202780103);广西科技厅广西重点研发计划(编号:桂科AB22035052)资助。

摘　　要：针对海洋牧场对海参、海胆及扇贝等目标生物,检测实时性和准确性的更高要求,提出了一种基于多聚合和多尺度的YOLOV5水下目标检测算法。该算法引入CIoU Loss损失函数以及DIoU-NMS非极大值抑制方法,以改善数据密集场景下目标漏检的问题;在网络中增加小目标检测层,提升小目标的检测精度;引入了坐标注意力机制CoordAtt,提取不同层次特征图的坐标信息;改进了特征融合网络结构,以提升目标检测坐标定位准确性以及获取更丰富的特征信息。在实验分析中引入了目标检测常用的平均精度(AP)、平均精度均值(mAP)以及检测速度作为评估标准,通过消融实验,以及与其他一阶算法结果进行对比验证算法的性能,实验结果显示该算法取得了更好的检测效果。Aiming at the higher requirements for the real-time and accuracy of detection of object such as holothurian,echi-nus and scallop in ocean ranch,a YOLOV5 underwater object detection algorithm based on multi-aggregation and multi-scale is proposed.The algorithm introduces the CIoU Loss loss function and DIoU-NMS non-maximum suppression method to improve the problem of missed detection of object in data-intensive scenarios.In order to improve the detection accuracy of small objects,a small object detection layer is added to the network.Then,the coordinate attention mechanism CoordAtt is introduced to extract the coordinate information of different levels of feature maps.Finally,the structure of the feature fusion network is improved,and the ex-tracted coordinate information is fused with the feature maps of the corresponding scale in the feature fusion network.In the experi-mental analysis,the indicators commonly used in object detection are introduced as the evaluation criteria,including the average ac-curacy(AP),average accuracy(mAP)and detection speed.The performance of the algorithm is verified through ablation experi-ments and compared with the experimental results of other latest one-stage algorithms.The experimental results show that the algo-rithm proposed in this paper at an advanced level.

关 键 词：目标检测 坐标注意力 非极大值抑制 YOLOV5 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]