基于改进YOLOv5的遥感图像小目标检测算法  

Remote Sensing Image Small Object Detection Algorithm Based on Improved YOLOv5

作　　者：王欣 江涛[1] 魏玉梅 马珍 白金燕 WANG Xin;JIANG Tao;WEI Yumei;MA Zhen;BAI Jinyan(College of Mathematics and Computer Science,Yunnan Minzu University,Kunming 650500)

机构地区：[1]云南民族大学数学与计算机科学学院,昆明650500

出　　处：《计算机与数字工程》2024年第7期2050-2054,2060,共6页Computer & Digital Engineering

基　　金：国家自然科学基金项目“基于改进YOLOv5的遥感图像小目标检测”(编号:61363022)资助。

摘　　要：遥感图像中小目标占图像的比例极小,准确识别这些目标具有很大的挑战性。针对遥感图像领域小目标检测困难的问题,提出一种改进YOLOv5遥感图像小目标算法。首先,使用改进后的Mosaic-9对数据集进行了预处理,以此解决遥感图像小目标研究数据稀缺的问题。其次,在主干网络中添加CA注意机制模块增加对细节信息的感知力,提升对小目标的检测能力。最后,在特征融合网络引入高效的双向跨尺度连接加权特征融合BiFPN,解决目标特征较少且易丢失的问题。实验结果表明,改进的网络模型对遥感图像小目标检测平均精度(mAP)达到90.3%,优于常规目标检测模型,适合遥感场景下的小目标检测。It is very challenging to accurately identify small objects in remote sensing images.In order to solve the problem of small object detection in remote sensing images,this paper proposes an improved YOLOv5.Firstly,Mosaic-9 is used to preprocess the data set to solve the problem of data scarcity.Secondly,the CA attention mechanism module is added to the backbone to in⁃crease the perception of detailed information and improve the detection ability of small objects.Finally,BiFPN is introduced into the feature fusion network to solve the problem that the object features are few and easy to be lost.The experimental results show that mAP of the improved YOLOv5 for small object detection in remote sensing images reaches 90.3%,which is superior to the conven⁃tional object detection model and suitable for small object detection in remote sensing images.

关 键 词：YOLOv5 遥感图像 小目标检测 CA注意力机制 BiFPN 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置] TP391[自动化与计算机技术—控制科学与工程]