基于卷积神经网络的车辆品牌和型号识别

发布时间：2018-01-14 06:22

本文关键词：基于卷积神经网络的车辆品牌和型号识别　出处：《东华大学学报(自然科学版)》2017年04期 　论文类型：期刊论文

【摘要】：车辆品牌和型号的识别属于细粒度分类领域的一类问题,与只针对不同物体的图像识别相比,待分类的车辆品牌和型号之间差异较小,分类较困难.卷积神经网络在静态图像上具有强大的特征发现能力,近年来在图像分类问题中成果显著.结合卷积神经网络和开源的大量标注数据集设计出了完整的车型识别模型,引入区域分割从而提高了识别的准确率,同时根据移动互联网的特性设计了交互方式.通过试验验证,该方法可以有效地解决查询图片识别具体车辆品牌及型号的问题.
[Abstract]:The recognition of vehicle brand and model belongs to a class of problems in the field of fine-grained classification. Compared with the image recognition of only different objects, the difference between the brands and models of vehicles to be classified is smaller. It is difficult to classify. Convolutional neural networks have powerful feature discovery ability in static images. In recent years, there are significant achievements in image classification. Combined with convolution neural network and a large number of open source annotated data sets, a complete vehicle recognition model is designed, and region segmentation is introduced to improve the accuracy of recognition. At the same time, the interactive mode is designed according to the characteristics of mobile Internet, and it is proved by experiments that this method can effectively solve the problem of querying pictures to identify specific vehicle brands and models.
【作者单位】：上海交通大学软件学院;
【基金】：国家自然科学基金资助项目(61373030,71171132)
【分类号】：TP183;TP391.41
【正文快照】： 车型识别有着广泛的应用前景,如,在智能交通领域能够利用车型识别对道路交通情况进行统计,在汽车销售领域可帮助潜在购车用户更好地了解感兴趣的车辆信息.车型识别的范围较广,不仅仅局限于数字和字母,而且由于无法对固定的区域进行特征提取,因此面临着较大的挑战.在车辆图片识

【相似文献】