基于注意力机制优化的YOLOv7目标检测算法OA

To ensure that an autonomous-driving system can accurately and swiftly identify road targets,this paper integrates SE,CBAM,and ECA attention modules into the YOLOv7 backbone network and designs an optimized network capable of feature recalibration.The network model is trained on the publicly available KITTI dataset,and a data processing platform is established.The performance of the network model and its detection of Car,Pedestrian,and Cyclist classes are compared before and after optimization with the SE attention module.Comparative experimental results demonstrate that the optimization algorithm incorporating the SE attention module achieves a better balance between computational efficiency and accuracy.Ablation studies further confirm that the SE module provides the primary performance gain,and the fused network balances detection accuracy with real-time performance.After optimization,the mean Average Precision(mAP)of the network increases by 1.16%,the FPS drop remains within 7%,and detection accuracies for Car,Pedestrian,and Cyclist improve by 3.17%,3.88%,and 2.77%respectively,effectively reducing missed and false detection rates.These results validate the effectiveness and practical value of the attention-optimized YOLOv7 algorithm for object detection in complex environments,providing robust technical support for the safety and reliable perception of autonomous driving systems.