Aircraft Targets Detection in Remote Sensing Images with Feature Optimization

Due to background interference, small size and dense arrangement in remote sensing images, the aircraft targets features are unconspicuous and easy to miss detection or misdetect. Aiming at this problem, a feature optimization algorithm based on YOLOv4 was proposed for aircraft targets detection in remote sensing images in this paper. In this algorithm, the feature fusion method of YOLOv4 was changed, and the multi-feature layers weighted fusion was used to reconstruct and optimize the different scale features. By expanding the scale range of the fusion features, the small targets detection ability is improved. Secondly, coordinate attention was used to effectively capture the relationship between feature-channels and the position information of feature-space. It can effectively reduce the interference of background and further enhance the feature information of the targets. The comparative trial results on the joint datasets of DIOR, RSOD and LEVIR show that the improved algorithm can effectively reduce the interference of complex background, significantly improve the detection effect of dense-small aircraft, and the performance is more superior.