| 摘 要: 针对车辆在雾霾天气下驾驶对道路目标的检测能力不足问题,提出一种基于改进 YOLOv11的目标检测模型 YOLO-EVC。采用去雾网络 DehazeNet对雾天图像进行去雾处理。在特征提取部分引入SPPFCSPC-G 新型池化结构。为增强对小目标的特征提取能力,在特征融合部分增加 EVC注意力机制。提出 C3_Res2模块降本增效。此外,引入自适应空间特征融合检测头 ASFF_Detect,有效地滤除空间域中的信息冲突。实验结果表明:YOLO-EVC在内部采集数据集上准确率达到90.6%,相比 YOLOv11检测平均精度均值(mAP)提高了1.9个百分点,召回率提高4.56个百分点,增强了对密集的小型车辆和行人的检测能力,对于未来汽车雾天自动驾驶有应用潜力。 |
| 关键词: 雾天目标检测 深度学习 YOLOv11 小目标检测 |
|
中图分类号: TP391
文献标识码: A
|
| 基金项目: 自然科学基金-甘肃省科技计划资助(24JRRA656);2022年横向课题:农业产品物资销售模式的数据统计分析(loonG20220201) |
|
| Traffic Target Detectionin Hazy Weather Based on YOLO-EVC |
|
GAO Zhaoyu, ZHAO Xia
|
(College of Information Science and Technology, Gansu Agricultural University, Lanzhou 730070, China)
1073324120842@stu.gsau.edu.cn; 58892778@qq.com
|
| Abstract: Aiming at the problem of insufficient detection ability of road targets when vehicles driving in hazy weather, a target detection model YOLO-EVC based on improved YOLOv11 is proposed. Firstly, the dehazing network DehazeNet is used to perform dehazing processing on images taken in hazy weather. In the feature extraction part, a new pooling structure named SPPFCSPC-G is introduced. To enhance the feature extraction ability for small targets, the EVC attention mechanism is added in the feature fusion part. The C3_Res2 module is proposed to reduce costs and improve efficiency. In addition, the adaptive spatial feature fusion detection head ASFF_Detect is introduced to effectively mitigate information conflicts in the spatial domain. Experimental results show that the Mean Average Precision(mAP)reaches 90. 6% on the sel-f collected dataset, which is 1. 9 percentage points higher than that of YOLOv11 in detection of YOLO-EVC and 4.56 percentage points higher in recall rate. The detection ability for dense small vehicles and pedestrians is enhanced, which has application potential for future autonomous driving of cars in hazy weather. |
| Keywords: target detection in hazy weather deep learning YOLOv11 small target detection |
