基于改进YOLOv11s的无人机小目标检测算法

打开文本图片集

关键 词：无人机航拍检测;小目标识别；YOLOvl1;HMD-YOLO 中图分类号：TP391.4文献标识码：A doi：10.37188/CJLCD.2025-0193 CSTR：32172.14.CJLCD.2025-0193

Abstract：Small object detection in UAV aerial images often suffers from challenges such as tiny target sizes，complex backgrounds，and limited computational resources.Existing UAV object detection models generally show insufficient accuracy and struggle to achieve a good balance between detection precision and efficiency. To address these challenges，this paper proposes an improved small object detection algorithm based on YOLOvlls，namely HMD-YOLO.Firstly，we design the HR-MSCA （High-Resolution MultiScale Convolutional Attention）module，which improves small object detection through a joint design of resolution enhancement and multi-scale convolutional attntion.Secondly，a lightweight and eficient dynamic upsampler，Litesample，is employed to replace the original upsampling module in the neck of the model.Additionally，the Wise-IoU loss function is introduced to improve the accuracy of bounding box regression and overall model performance.Finally，a Dynamic Detection Head（DynamicHead） is incorporated to further enhance the model's capability in detecting small objects.Experimental results on the VisDrone2Ol9 ：dataset demonstrate that the improved model achieves 49. 98% （204号 mAP@0.5 and 30.73% mAP@0.95，showing improvements of 12.15% and 8.22% respectively over YOLO vlls. The effectiveness of the proposed approach is further validated through generalization experiments on the TinyPerson dataset，where the model also achieves a significant performance gain.

Key words： drone aerial surveying；smallobject recognition;YOLOvll;HMD-YOLO

1引言

随着人工智能与航空航天技术的深度融合，低空自主感知系统正逐步成为智能无人系统的重要组成部分。（剩余16927字）