基于改进YOLOv5的遥感图像检测  

Remote Sensing Image Detection Based on the Improved YOLOv5

作　　者：王志林 于瓅[1] WANG Zhilin;YU Li(School of Computer Science and Engineering,Anhui University of Science and Technology,Huainan Anhui 232001,China)

机构地区：[1]安徽理工大学计算机科学与工程学院,安徽淮南232001

出　　处：《重庆科技学院学报（自然科学版）》2024年第2期62-67,共6页Journal of Chongqing University of Science and Technology：Natural Sciences Edition

基　　金：安徽省重点研究与开发计划项目“基于多源数据融合的交通事件主动防控技术研究与示范应用”(202104D07020010)。

摘　　要：针对现有遥感图像目标检测中背景复杂和尺度变化大等问题,基于YOLOv5模型提出了一种改进的遥感图像目标检测算法。首先,利用Mosaic数据增强方法重构数据集,以改善模型的训练效果和鲁棒性;其次,在YOLOv5s的Backbone中添加SE注意力机制,使改进后模型能够更精准地捕捉目标特征信息;最后,采用BiFPN替代原模型中的FPN+PAN结构,使模型能够进行不同尺度的特征融合,以减少检测过程中浅层信息的丢失。实验结果表明,相较于原模型,改进后模型的平均精度均值、准确率和召回率都有所提升;相较于原模型,改进后模型具有更强的特征提取能力及更快的检测效率。Aiming at the problems of background complexity and scale change in existing remote sensing image tar-get detection,an improved remote sensing image target detection algorithm based on YOLOv5 model was proposed.Firstly,the Mosaic image is used to enhance the reconstructed dataset to improve the training effect and robustness of the model.Secondly,the SE attention mechanism is added to the backbone network of YOLOv5s,so that the im-proved model can capture the target feature information more accurately.Finally,BiFPN is used to replace the FPN+PAN structure in the original model,so that the model can carry out feature fusion at different scales,and reduce the shallow information loss in the detection process.The experimental results show that compared with the original model,the average precision,accuracy rate and recalling rate of the improved model are promoted and the im-proved model has stronger feature extraction ability and higher detection efficiency for remote sensing image target detection,which verifies the effectiveness of the improved method.

关 键 词：遥感图像检测 YOLOv5算法 注意力机制 加权双向特征金字塔 目标检测 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置]