摘要
文章针对车辆型号分类中存在车辆不同视角影响的问题,提出了一个视角相关的卷积神经网络(viewing angle relative convolutional neural network,VAR-NET)模型。该模型包含视角预测和分类2个子网络,其中视角预测子网络用于提取车辆的拍摄视角信息,分类子网络用于提取车辆特征并实现其分类。在公开数据集CompCars和Standford Cars上的实验结果表明,VAR-NET模型在多视角车辆图像上取得了很好的识别效果,其识别率高于一些其他经典的网络模型。
Aiming at the influence of different viewing angles of vehicles in the classification of vehicle models, this paper proposes a viewing angle relative convolutional neural network(VAR-NET). The model includes two sub-networks, one is the viewing angle prediction sub-network, which is used to extract viewing angle information of the vehicle;the other is the classification sub-network, which is used to extract vehicle features and classify them. The experimental results on public datasets CompCars and Standford Cars show that VAR-NET can achieve good recognition results on multi-view vehicle images, and its recognition accuracy is higher than that of other classical network models.
作者
朱文佳
付源梓
金强
余烨
ZHU Wenjia;FU Yuanzi;JIN Qiang;YU Ye(Anhui Baichenghuitong Technology Co.,Ltd.,Hefei 230088,China;School of Computer and Information,Hefei University of Technology,Hefei 230601,China)
出处
《合肥工业大学学报（自然科学版）》
CAS
北大核心
2020年第2期205-210,279,共7页
Journal of Hefei University of Technology：Natural Science
基金
国家自然科学基金青年科学基金资助项目(61906061)
安徽省重点研究和开发计划资助项目(201904d07020010)
关键词
车型识别
卷积神经网络(CNN)
精细分类
视角预测
vehicle model recognition
convolutional neural network(CNN)
fine-grained classification
viewing angle prediction
