基于轻量化改进型YOLOv5的车辆检测方法

Vehicle Detection Method Based on Lightweight Improved YOLOv5

摘要: 本文建立了一种基于计算机视觉的检测前方车辆模型，实现对前方目标车辆的实时检测。本文提出一种基于YOLOv5算法的轻量化车辆检测模型。首先，搭建了YOLOv5检测网络，针对误检漏检问题，使用ConvNeXt重建YOLOv5的主干网络来进行特征提取，以提升网络的细粒度特征融合能力，提高检测精度；然后，为有效解决图像尺度特征变换较大问题其次，在主干网络中引入坐标注意力机制引入提高了主干特征提取效率，进一步提升了算法的特征提取能力；其次，对模型进行剪枝操作，使模型更加轻量化。实验结果表明，改进YOLOv5算法平均精度均值达到97.41%，较原始算法提升5.3%，剪枝后模型检测速率达到175 f/s，较原速率提升了69.4%，证明了该算法可以满足对车辆的实时检测要求。

Abstract: In this paper, a vehicle detection model based on computer vision is established to realize real-time detection of the target vehicle in front. This paper proposes a lightweight vehicle detection model based on YOLOv5 algorithm. First, the YOLOv5 detection network is built. For the small target mod-el in the image, the backbone network of YOLOv5 is reconstructed by ConvNeXt to extract features, so as to improve the fine-grained feature fusion ability of the network and improve the detection accuracy; Secondly, in order to effectively solve the problem of image scale feature transformation, the introduction of coordinate attention mechanism in the backbone network improves the effi-ciency of backbone feature extraction and further improves the feature extraction ability of the al-gorithm; Thirdly, prune the model to make it more lightweight. The experimental results show that the average accuracy of the improved YOLOv5 algorithm reaches 97.41%, which is 5.3% higher than the original algorithm. The model detection rate after pruning reaches 175 f/s, which is 69.4% higher than the original rate. It is proved that the algorithm can meet the requirements of real-time vehicle detection.