| 摘 要: 在雾天环境下,针对难以捕捉车辆的特征导致车辆检测精度低的问题,提出一种基于YOLOv8改进的雾天环境下车辆检测算法———CSM-YOLOv8(Cross Scale Multi-YOLOv8)。首先,将C2f模块的第一个卷积替换为SA-Net模块,使其能够处理不同尺度的特征图。其次,引入SEAttention(Squeeze and Excitation Networks)注意力机制,通过自适应地学习特征通道之间的相关性,捕捉雾天环境下的车辆特征。最后,加入MHSA(Multi-Head Self-Attention)模块通过将注意力机制分解成多个头部并行计算。实验结果表明,CSM-YOLOv8的检测平均精度均值(mAP)提高了1.9百分点。由此可见,改进后的模型有效克服了雾天环境中车辆特征模糊导致检测精度低的问题,实现了对车辆的准确检测。 |
| 关键词: YOLOv8;车辆检测;雾天环境;注意力机制 |
|
中图分类号: TP391
文献标识码: A
|
|
| Improved Vehicle Detection in Foggy Conditions Based on YOLOv8 |
|
GUO Jun, SHAO Mengzhen, CHEN Xinyu,YANG Yue
|
(School of Automotive Engineering, Yancheng Institute of Technology, Yancheng 224051, China)
gj_njau@163.com; shaomzz@163.com; 2093944703@qq.com; 227094923@qq.com
|
| Abstract: In foggy conditions, capturing vehicle features is challenging, leading to low accuracy in vehicle detection. In view of this problem, this paper proposes an improved vehicle detection algorithm for foggy conditions based on YOLOv8, named CSM-YOLOv8 (Cross Scale Multi-YOLOv8). Firstly, the first convolution of the C2f module is replaced with an SA-Net module, enabling it to handle feature maps of different scales. Secondly, the SEAttention ( Squeeze and Excitation Networks) mechanism is introduced, which adaptively learns the correlations between feature channels to capture vehicle features in foggy conditions. Lastly, the MHSA (Multi-Head Self-Attention) module is added, allowing the attention mechanism to be decomposed into multiple heads for parallel computation. Experimental results show that the detection mean Average Precision (mAP) of CSM-YOLOv8 improves by 1. 9 percentage points. This indicates that the improved model effectively addresses the issue of low detection accuracy due to the blurring of vehicle features in foggy conditions, achieving accurate vehicle detection. |
| Keywords: YOLOv8; vehicle detection; foggy conditions; attention mechanism |
