融合整体与局部特征的车辆型号识别方法

Vehicle type recognition method fusing global and local features

针对现有的深度学习模型忽略从特定图像区域提取局部特征的问题,结合智能交通系统领域的具体应用,提出融合整体与局部特征的车辆型号识别方法。首先,根据车头图像的纹理特点与车牌位置,将车头的全局图像分为多个局部图像;然后,使用优化的卷积神经网络模块分别提取图像的全局特征与局部特征,将全局与局部特征输入到多层全连接神经网络进行特征融合;最后,使用softmax进行车辆型号识别。实验表明,提出的方法能够提取有效的全局与局部特征,具有训练收敛快、识别准确率高的优点。

Since the deep learning model ignores the local feature extracted in a particular image region, a vehicle type re- cognition method fusing the global and local features is proposed by combining the specific application in the intelligent transpor- tation system field. According to the texture feature of the image front of the vehicle and licence plate location, the global image front of the vehicle is divided into multiple local images. The optimal convolutional neural network module is used to extract the global feature and local feature of the image. And then the global feature and local feature are input into the multilayer fully- joint neural network for feature fusion. The softmax is used to recognize the vehicle types. The experimental results demonstrate that the method can extract the effective global and local features, and has the advantages of fast convergence rate and high recognition accuracy.