Vehicle Detection Model Based on Improved Yolo11

Aiming at the challenges of multi-scale target recognition, background noise interference and real-time requirements in vehicle detection in complex traffic scenes, this paper proposes a lightweight vehicle detection model YOLO-MF (YOLO-MsBlock&Focal Modulation) based on improved YOLO11. The performance of the model is improved by two core improvements. Firstly, the Focal Modulation mechanism is introduced into the backbone network to replace the traditional SPPF module, and multi-focus hierarchical context aggregation and global feature modulation are used to enhance the ability to capture semantic information of vehicle contours and suppress background noise. 2) The multi-scale feature fusion module C3k2_MSBlock is designed to optimize the multi-scale feature expression ability through the parallel multi-branch structure and cross-layer feature interaction strategy. Experiments on the UADETRAC dataset show that the MAP50 and MAP50-95 of YOLO MF reach 93.4 % and 73.8 %, respectively, which are 1.9 and 1.2 percentage points higher than those of the baseline model. The number of parameters only increases by <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{0. 0 4 M}$</tex> (the total number of parameters is 2.54 M), and the computational complexity (GFLOPs) is controlled at 6.8. The model achieves a balance between accuracy and efficiency, which is suitable for edge device deployment and provides efficient and reliable vehicle detection solutions for intelligent transportation systems.