改进YOLOv11n的车辆目标检测算法OA

Currently,vehicle detection algorithms fail to well balance between accuracy and the number of parameters.To address the issue,this paper proposes an improved object detection algorithm YOLO-CCLW based on YOLOv11n.First,ConvFormer was integrated with convolutional gated linear units to design the C3k2_ConvFormer_CGLU module,significantly enhancing the network's global feature extraction capabilities.Then,the original detection head was replaced by a shared convolutional detection head,further reducing computational complexity and parameter count.Finally,the CIoU loss function was replaced with Wise-MPDIoU to enhance the model's localization and detection capabilities.Experiments were conducted on the KITTI dataset to verify the performance of the YOLO-CCLW algorithm.Results show YOLO-CCLW improves accuracy by 1.2％,recall rate by 4.6％,and mAP@0.5 by 3.0％compared to that of the traditional YOLOv11n.Meanwhile,the model's parameter count is down to 2.0M and the computation to 5.3G.While achieving superior detection accuracy,the improved algorithm requires fewer parameters.