改进YOLOv5s遥感图像识别算法研究

Research on improved YOLOv5s remote sensing image recognition algorithm

针对遥感图像小目标检测精度不高、目标密集不易检测等问题,提出了一种改进YOLOv5s的遥感图像检测算法.在数据增强模块中通过增加垂直旋转方法解决了图像角度不同造成差异的问题;通过在YOLOv5s模型的Backbone模块中引入注意力机制使模型网络更加关注重点信息,以提高目标检测的精度.该算法在TGRS-HRRSD-Dataset数据集上进行实验,召回率达到99.8%,map@0.5达到98.8%,精确率达到75.6%,各项指标有明显提高.

Aiming at the problems of low detection accuracy of small targets in remote sensing images and difficult detection of dense targets,an improved YOLOv5s remote sensing image detection algorithm was proposed.In the data enhancement module,the vertical rotation method was added to solve the difference caused by different image angles.The attention mechanism was introduced into the Backbone module of YOLOv5s model to make the model pay more attention to key information,so as to improve the accuracy of object detection.The algorithm was tested on the TGRS-HRRSD-Dataset.The recall rate reached 99.8%,map@0.5 reached 98.8%,and the accuracy rate reached 75.6%.All indexes were improved significantly.