Small Target Vehicle Detection Algorithm for UAV Based on Improved YOLOv11

To address the challenge of insufficient detection accuracy for small targets (vehicles) in UAV-captured complex scenes, this paper proposes an improved YOLOv11 algorithm. The proposed method incorporates a multi-branch depthwise separable convolution module to enhance the C3K2 module, significantly enhancing the model's ability to detect targets of various sizes. Furthermore, a novel scalable pyramid atrous convolution module is designed to replace the SPPF module. This module utilizes dilated convolutions instead of pooling layers, effectively reducing the loss of detailed information and improving both operational flexibility and computational efficiency of feature extraction. Experimental results on the VisDrone2019 dataset demonstrate that the proposed algorithm achieves mAP@0.5 of 46.4%, surpassing mainstream models while maintaining the detection speed of 76 FPS, thus fulfilling real-time requirements. Compared with the benchmark model YOLOv11, the precision has increased by 4.7%. Ablation and comparative experiments further validate the effectiveness and high efficiency in small target detection tasks for UAV, while visual results demonstrate the robustness of the proposed algorithm in complex scenes.