基于CF-YOLO的雾霾交通标志识别
Foggy traffic sign recognition based on CF-YOLO
吴攀超 1郑卓纹 1王婷婷 1孙琦1
作者信息
- 1. 东北石油大学电气信息工程学院,黑龙江大庆 163000
- 折叠
摘要
针对现有交通标志检测模型在雾霾环境下出现漏检、错检以及参数较大等问题,设计一种基于YOLOv5s改进的CF-YOLO检测模型.为加强在雾霾环境中对交通标志的检测能力,提出一种基于颜色衰减先验的自适应伽马变换图像预处理算法;为增强对目标的定位能力及检测精度,将坐标注意力机制融合到网络中;为实现模型轻量化,引入FasterNet-Block构建网络.实验结果表明,改进算法在雾霾环境下交通标志检测相比原YOLOv5模型权重减少了 2.3 MB,精度提高了 8.5个百分点.
Abstract
Aiming at the problems of leakage,wrong detection and large parameters of the existing traffic sign detection model in haze environment,a CF-YOLO detection model based on the improvement of YOLOv5s was designed.An adaptive gamma trans-form image preprocessing algorithm based on the color decay prior was proposed to enhance the ability to detect traffic signs in hazy environments.To enhance the localization ability and detection accuracy of the target,the coordinate attention mechanism was integrated into the network.To realize model light weighting,FasterNetBlock was introduced to construct the network.Experimental results show that the improved algorithm reduces the weight of traffic sign detection in haze environment by 2.3 MB compared to that of the original YOLOv5 model,and improves the accuracy by 8.5 percentage points.
关键词
交通标志识别/目标检测/卷积神经网络/坐标注意力机制/颜色衰减先验/伽马变换/深度学习Key words
traffic sign recognition/target detection/CNN/coordinate attention/color decay prior/Gamma transform/deep learning引用本文复制引用
基金项目
国家自然科学基金项目(52074088)
黑龙江省博士后科研启动基金项目(LBH-Q21086)
黑龙江省省属高校基本科研业务费基金项目(2022TSTD-04)
出版年
2024
NETL
NSTL
万方数据