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.
The recognition of car make and model is a kind of problem of fine-grained classification area. It＇s hard to classify them due to the subtle difference among classes compared to other common image recognition problems. While powerful feature-found ability of convolutional neural network（CNN） on static image has made remarkable achievements in the image classification problem. Therefore, a complete model based on CNN is designed by combining the large open source data sets, region segmentation is applied to raise the accurate while the interaction is designed according to the characteristics of mobile internet. Experiment shows that this method can effectively solve the problem.
作者
黎哲明
蔡鸿明
姜丽红
LI Zheming CAI Hongming JIANG Lihong(School of Software, Shanghai Jiao Tong University, Shanghai 200240, Chin)
出处
《东华大学学报（自然科学版）》
CSCD
北大核心
2017年第4期472-477,共6页
Journal of Donghua University(Natural Science)
基金
国家自然科学基金资助项目(61373030
71171132)
关键词
车型识别
细粒度分类
卷积神经网络
区域分割
图像处理
car make and model recognition
fine-grained classification
convolutional neural network
region segmentation
image processing
