| 摘 要: 针对现有目标检测模型在小目标检测中存在精度不足、误检率高等问题,提出了一种基于 YOLOv11的改进型道路目标检测算法—CZD-YOLO。设计CPMSFA模块,以提升多尺度特征的提取能力;引入ZSA融合模块,使模型聚焦关键目标区域;设计 DDCNV 检测头,增强模型对小目标的捕捉能力。CZD-YOLO 在公开数据集KITTI和nuScenes分别进行实验。结果表明:CZD-YOLO的召回率分别提升了6.1%和4.5%;检测精度分别提升了2.3%和6.4%;平均精度均值分别提升了4.4%和5.4%。通过可视化结果分析进一步验证了模型的有效性。 |
| 关键词: 目标检测 YOLOv11 多尺度特征 小目标检测 |
|
中图分类号:
文献标识码: A
|
|
| Road Target Detection Algorithm Improved Based on YOLOv11 |
|
CHENG Dashuai, TANG Pei, PAN Jiahao, QIU Zetao
|
(School of Automotive Engineering, Yancheng Institute of Technology, Yancheng, 224000, China)
cs2595667133@163.com; tangpei@ycit.edu.cn; panjh093733@163.com; qzt1520691084@163.com
|
| Abstract: To address the challenges faced by existing object detection models—such as insufficient accuracy and high false positive rates in small object detection—this paper proposes an improved road object detection algorithm based on YOLOv11, named CZD-YOLO. First, a CPMSFA module is designed to enhance multi-scale feature extraction. Second, a ZSA fusion module is introduced to help the model focus on key target regions. Finally, a DDCNV detection head is developed to improve the model’s ability to detect small objects. Experiments are conducted on public datasets KITTI and nuScenes, showing that CZD-YOLO improves recall by 6.1% and 4.5% , precision by 2.3% and 6.4% , and mean Average Precision by 4.4% and 5.4% , respectively. Visual analysis of the detection results further confirms the effectiveness of the proposed model. |
| Keywords: object detection YOLOv11 mult-i scale features small object detection |
