Multi scale and direction target detecting in remote sensing images via modified YOLO-v4

Traditional target detection algorithms have difficulty to adapt complex environmental changes and have limited applicable scenarios. However, the deep learning-based target detection model can automatically learn with strong generalization capability. In this paper, we choose a single-stage deep learning-based target detection model for research based on the models real-time processing requirements and to improve the accuracy and robustness of target detection in remote sensing images, In addition, we improves the YOLOv4 network and present a new approach. Firstly, proposes a classification setting of the Non-maximum suppression (NMS) threshold to increase accuracy without affecting the speed. Secondly, we study the anchor frame allocation problem in YOLOv4 and proposes two allocation schemes. The proposed anchor frame scheme also improves the detection performance, and experimental results on Dota dataset validate their effectiveness.