Improved YOLOv5 for remote sensing image small target detection
An improved YOLOv5 small target detection algorithm,YOLOv5-FRM,is proposed to address the issue of poor detection performance of mainstream algorithms in remote sensing image small target detection.Firstly,a Coordinate Attention(CA)module is added at the end of the original YOLOv5 backbone network to replace the original SPP module.Then,an improved multi-scale spatial purification module is proposed,which implements the addition of detection heads and integrates them into the neck network of the original YOLOv5.Finally,a Copy-reduce-paste data augmentation method is introduced to improve the training effectiveness of the model.The experimental results show that the improved algorithm effectively improves the detection accuracy of small targets in remote sensing images,and reduces the false detection rate and missed detection rate.
NETL
NSTL
万方数据
维普
