FGHDet: Delving Into Fine-grained Features With Head Selection for UAV Object Detection

Detecting small objects in UAV imagery is a challenging and crucial task in computer vision. Most current methods struggle to tackle the challenges of the small object: fine-grained feature mining, multiple-layer feature fusion, and a mismatch in scale between anchors and feature maps. To alleviate the aforementioned issues, we present the FGHDet, focusing on delving into fine-grained features in low-level features with a head selection mechanism. Firstly, our approach introduces a Detail-preserving Semantic Information Enhancement Module to retain fine-grained information while excavating coarse-grained semantic details relevant to fine-grained information. Then, we devise a Coarse-to-fine Feature Guidance Module that leverages coarse-grained semantic information and fine-grained information to co-guide feature enhancement, further improving the model's classification ability. Finally, we introduce a multi-scale detection strategy based on anchor-head matching, ensuring scale-level matching between anchors and feature maps to prevent over-fitting due to overly fine anchor divisions. Extensive experiments on VisDrone, CARPK, and Drone-vs-Bird datasets demonstrate the effectiveness of FGHDet. It achieves notable mAP improvements (IoU range 0.5:0.95) of 4.9, 4.1, and 2.2, respectively, showcasing strong competitiveness against state-of-the-art methods. The code is available at https://github.com/b-yanchao/UAVDetection.git.