Research on small object detection method based on YOLOv8:Application of high-resolution detection layer and Mosaic data augmentation

Abstract

For the core problem that the number of small targets in remote sensing images is continually increasing and traditional object detection methods struggle to meet its detection requirements, a small-target detection method based on YOLOv8 is proposed, and its detection performance is enhanced through two key improvements. First, to address the difficulty of detecting small targets, a high-resolution detection layer P2 is added to the Neck part of the YOLOv8 model to enhance the model's feature extraction capability for small targets. Secondly, the Mosaic data augmentation technique is introduced, which stitches multiple images into one image to expand the diversity of training data, thereby improving the model's generalization to small targets. Experiments on the NWPU VHR-10 dataset show that the improved YOLOv8 model significantly outperforms the original model in small-target detection.Specifically, the Recall rate of the improved model reaches 85.6%, which is 9.18% higher than that of the baseline model. At the same time, mAP@0.5 edged up to 0.724, while the more stringent mAP@0.5:0.95 index increased significantly from 0.491 to 0.514.These results validate the effectiveness of the proposed approach in improving small-target detection performance.