基于YOLOv5算法的遥感飞机图像检测

Remote sensing aircraft image detection based on YOLOv5 algorithm

针对遥感飞机图像检测存在实验数据集缺少的问题,对目前公开的遥感飞机数据集进行筛选,并从GoogleEarth等开源途径获取飞机图像完成对数据的扩增,制作新数据集MTAR。针对飞机型号间差异性较小的问题,采用YOLOv5作为基准网络框架,在特征提取阶段引入CBAM注意力机制模块,在特征加强阶段使用Bifpn结构,加强对飞机局部特征的学习,增强识别能力。实验结果表明,该方法在MTAR数据集上的mAP值达到90.9%,具有较好的检测效果。

In view of the lack of experimental data set in remote sensing aircraft image detection,the current public remote sens⁃ing aircraft data set was screened,aircraft images were obtained from Google Earth and other open source channels to complete data am⁃plification,and a new data set MTAR was prepared.To solve the problem of small differences among aircraft models,YOLOv5 is adopt⁃ed as the reference network framework,CBAM attention mechanism module is introduced in the feature extraction stage,and Bifpn structure is used in the feature strengthening stage to strengthen the learning of local features of aircraft and enhance the recognition ability.The experimental results show that the mAP value of the proposed method reaches 90.9%on the MTAR dataset,which has a good detection effect.