Research on Small Object Detection Methods from the Perspective of Unmanned Aerial Vehicles Based on the Improved YOLOv11

When target detection is performed from a drone’s perspective, it often faces complex backgrounds and dense clusters of small targets, resulting in low detection accuracy. In this paper, an improved HC-YOLOv11 model is proposed based on the YOLOv11 model. Firstly, the YOLOv11 model extracts features with three detection heads, respectively corresponding to large, medium, and small targets of different scales. However, the detection head for small targets in this study cannot meet the needs, so an additional fine-grained extra Detection head is added to utilize more local information and improve the model’s feature extraction ability. Secondly, a lightweight convolutional block attention module (CBAM) is used at the end of the backbone network, which enhances important channel and spatial features to improve target detection accuracy. Finally, the experimental results show that HC-YOLOv11, compared with the original model, has an mAP@[. 5: 95] improvement of $2.42 \%$ and an mAP@ 0.5 improvement of $3.54 \%$ on the VisDrone-2019-DET dataset. The experimental results prove the effectiveness of the proposed improvement method.