Improved YOLO Network for Free-Angle Remote Sensing Target Detection

Despite significant progress in object detection tasks, remote sensing image target is still challenging owing to complex backgrounds, large differences sizes, and uneven distribution of rotating objects. In this study, we consider model accuracy, inference speed, objects at any angle. We also propose a RepVGG-YOLO network using an improved RepVGG as the backbone feature extraction network, which performs initial from input considers training accuracy speed. use pyramid (FPN) path aggregation (PANet) reprocess output by network. The FPN PANet module integrates maps different layers, combines context information on multiple scales, accumulates features, strengthens extraction. Finally, maximize all four scales enhance small pixels. To solve angle problem object, loss function for classification circular smooth label technology, turning regression into problem, increasing conducted experiments two public datasets, DOTA HRSC2016. Our results show proposed method better than previous methods.