FRN-YOLO: A Feature Re-fusion Network for Remote Sensing Target Detection

Abstract

With the development of artificial intelligence technology, remote sensing target detection has gradually become a hot issue in the field of computer vision, which can be widely used in navigation, exploration, disaster warning, etc., and it has important research significance and application value for remote sensing target detection. However, the scale difference of remote sensing targets makes detection very difficult. Therefore, we propose a feature re-fusion network based on YOLO-FRN-YOLO. Based on the original three detection layers of YOLO, by re-fusing the features of the three output layers of the backbone, each feature layer can be deeply combined with The semantic information before sampling or after sampling, and the depth of the detection layer after feature re-fusion retains the semantic information of targets of different scales, and improves the detection ability of targets of different scales. The results show that on the RSOD datasets, the average precision of our method exceeds YOLOv3, and it is also better than other advanced networks.