融合整体与局部特征的车辆型号识别方法

发布时间：2019-04-13 11:53

【摘要】：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。
[Abstract]:In view of the problem that the existing depth learning models ignore the problem of extracting local features from specific image regions, combined with the specific application of intelligent transportation system, a vehicle model recognition method combining global and local features is proposed. Firstly, according to the texture characteristics of the front image and the location of the license plate, the global image of the head is divided into several local images. Then, the optimized convolution neural network module is used to extract the global and local features of the image, and the global and local features are inputted to the multi-layer fully connected neural network for feature fusion. Finally, the vehicle model recognition is carried out by using softmax. Experiments show that the proposed method can extract effective global and local features, and has the advantages of fast training convergence and high recognition accuracy.
【作者单位】：桂林电子科技大学信息与通信学院;
【基金】：国家科技支撑计划课题资助项目(2014BAK11B02) 广西科学研究与技术开发计划资助项目(桂科攻14122007-5) 桂林电子科技大学研究生创新项目(YJCXS201521)
【分类号】：U495;TP391.41

【相似文献】