基于改进YOLOv7的复杂道路目标检测算法OA

To address the challenges of low detection accuracy for dense targets in complex autonomous driving road scenarios,which often lead to missed detections and false alarms,an improved YOLOv7 target detection algorithm was proposed.The algorithm introduced a small-target detection layer to the basic model architecture,enhancing the receptive field and improving feature learning for small targets.The K-means++algorithm was employed for dataset clustering,ensuring that anchor boxes better align with actual targets.A two-level routing mechanism was integrated into the SPPCSPC module,enabling dynamic querying and sparse perception capabilities of the model.Additionally,the original CIoU loss function was replaced with the EIoU loss function to accelerate model convergence.Experimental re-sults demonstrate that the proposed method achieves an mAP@0.5 of 94.9%on the KITTI dataset,rep-resenting a 3.1 percentage point improvement over the original algorithm.On the SODA10M dataset,it attains an mAP@0.5 of 67.8%,a 5.6 percentage point increase.This approach effectively mitigates small-target detection challenges in complex autonomous driving road environments.