基于PSN-YOLOv7的光学遥感图像目标检测  

Optical remote sensing image object detection based on PSN-YOLOv7

作　　者：陈鑫钰 陈媛媛[1] 廉永健 耿艳兵 张楠 刘圣仁 张玲郡 CHEN Xinyu;CHEN Yuanyuan;LIAN Yongjian;GENG Yanbing;ZHANG Nan;LIU Shengren;ZHANG Lingjun(School of Information and Communication Engineering,North University of China,Taiyuan 030051,China;School of Computer Science and Technology,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 [2]中北大学计算机科学与技术学院,山西太原030051

出　　处：《现代电子技术》2024年第21期119-124,共6页Modern Electronics Technique

基　　金：山西省自然科学基金项目(202103021224192);山西省自然科学基金项目(202203021221103)。

摘　　要：针对光学遥感图像因目标尺度差异大,小目标特征信息不足导致的检测精度低、误检、漏检问题,文中提出一种基于PSN-YOLOv7的光学遥感图像目标检测方法。首先,增加一个160×160的小目标检测层,增强网络的多尺度目标检测能力;其次,提出改进的ELAN-P和ELAN-WS模块,在减少计算量和参数量的同时,增强全局理解能力减少目标特征信息的损失,有效改善误检;最后,提出MNI损失函数改进YOLOv7的CIoU位置损失函数,抑制小目标造成网络损失不平衡的影响,从而减少目标漏检的情况。在DIOR公开遥感图像数据集上进行训练与测试,实验对比结果表明,所提方法检测精度mAP@0.5为92.3%,比YOLOv7网络提高了3.2%,且计算量和参数量与原YOLOv7网络相当,提升检测精度的同时保证了网络的性能平衡。In view of the low detection accuracy,false detection and missing detection in optical remote sensing images caused by large difference in object scale and insufficient feature information of small objects,a method of optical remote sensing image object detection based on PSN-YOLOv7 is proposed.A 160×160 small object detection layer is added to enhance the multi-scale object detection capability of the network.The improved ELAN-P(ELAN is the abbreviation of efficient layer aggregation network)and ELAN-WS modules are put forward to strengthen the global understanding ability,reduce the loss of feature information and improve detection effectively while reducing the calculation burden and parameters.The MNI(mixed loss function of NWD and IoU)loss function is proposed to improve the CIoU(complete IoU)position loss function of YOLOv7 to suppress the influence of network loss imbalance caused by small objects,so as to reduce the missing detection of objects.Training and testing were carried out on public remote sensing image dataset DIOR.The results of comparison experiments show that the detection accuracy mAP@0.5 of the proposed method is 92.3%,which is 3.2%higher than that of YOLOv7 network,and its calculation burden and parameters are equivalent to those of the original YOLOv7 network.It can be seen from the above that the proposed algorithm improves the detection accuracy and ensures the performance balance of the network.

关 键 词：光学遥感图像 目标检测 损失函数 多尺度 深度学习 特征信息 

分 类 号：TN911.73-34[电子电信—通信与信息系统] TP751.1[电子电信—信息与通信工程]