基于改进YOLOv7x的车型识别算法

Vehicle Type Recognition Algorithm Based on Improved YOLOv7x

为提高车型识别精度并改善误检问题,本文提出了一种改进YOLOv7x的车型识别算法.首先,对BIT-Vehicle数据集进行提取并人为划分,使数据集符合模型要求;其次,将CBAM(Convolutional block attention module)注意力机制加入backbone网络中,在模型参数增幅较小的情况下,提升主干网络的特征提取能力;最后,引入SIoU损失函数,利用predict box和groundtruth box之间的向量角度,重新定义了损失函数,提高了模型检测的mAP0.5(mean Average Precision,IoU=0.5)与准确率.实验结果表明,改进YOLOv7x算法整体识别结果优于YOLOv7x算法,改进YOLOv7x算法的mAP0.5和准确率分别为98.4%和97.4%,分别提高了0.5%和1.1%,具有更优的精度和更低的误检率,较好地满足了车型识别的需求.

In order to improve the accuracy of vehicle type recognition and improve the problem of false detection,this paper proposes an improved YOLOv7x vehicle type recognition algorithm.Firstly,this paper extracts the BIT-Vehicle dataset and divides it manually to make the dataset meet the requirements of the model;Secondly,it adds CBAM attention mechanism to backbone network to improve the feature extraction ability of backbone network with small increase of model parameters;Finally,it introduces the SIoU loss function,redefines the loss function and improves the mAP0.5 and precision of model detection by utilizing the vector angle between the predict box and the groundtruth box.The experimental results show that improved YOLOv7x algorithm is better than YOLOv7x algorithm.For the improved YOLOv7x algorithm,mAP0.5 and precision are 98.4%and 97.4%,increase by 0.5%and 1.1%respectively,improved YOLOv7x vehicle type recognition algorithm has better accuracy and lower false detection rate,which better meets the needs of vehicle recognition.