UAV Detection Method Based on Improved YOLOv8

To address the challenge of Unmanned Aerial Vehicle (UAV) detection in computer vision—due to their high-speed motion and small size—we propose Hybrid Feature-Aware You Only Look Once (HFA-YOLO), an enhanced version of YOLOv8. The model integrates Mix Structure Block (MSB) and Triplet Attention (TA) to jointly improve multi-scale feature extraction and cross-dimensional information fusion, while adopting Wise Intersection over Union version 3 (WIoU v3) to enhance bounding box regression accuracy. MSB strengthens small target representation, TA improves inter-feature interaction, and WIoU v3 refines localization precision. All input images used in this method are in color format, which helps preserve texture and contextual information for better detection performance. Experimental results on two public datasets, both obtained from the Roboflow platform, demonstrate that HFA-YOLO significantly outperforms the original YOLOv8n in mAP, precision, and recall, validating the effectiveness of our improvements.