Rotating object detection method in aerial images based on anchor-free decoupling head
Aerial image objects have the characteristics of small area ratio,dense arrangement,and arbitrary inclination angle.In order to meet the requirements of accurate detection of aerial image objects,the feature extraction network is improved,and the ellipse center sampling method is used to optimize the label sampling strategy to solve the problem of insufficient sampling.Finally,an anchor-free decoupling object detection head is used to separate the bounding box regression task from the object classification task to improve detection accuracy.Experiments show that the proposed method achieves 75.2%and 89.1%mAP on the DOTA and HRSC2016 datasets,respectively,which meets the requirements of accurate detection.
anchor-freedeep learningellipse center samplingdecoupling detection
NETL
NSTL
万方数据
维普
