基于改进卷积神经网络的车型识别  

Vehicle identification based on improved convolutional neural network

作　　者：陈立潮[1] 卜楠 潘理虎[1,3] 曹建芳[2] 张英俊[1] CHEN Li-chao;BU Nan;PAN Li-hu;CAO Jian-fang;ZHANG Ying-jun(School of Computer Science and Technology,Taiyuan University of Science and Technology,Taiyuan 030000,China;Department of Computer Science and Technology,Xinzhou Teachers University,Xinzhou 034000,China;Institute of Geographic Science and Natural Resource Research,Chinese Academy of Sciences,Beijing 100101,China)

机构地区：[1]太原科技大学计算机科学与技术学院,山西太原030000 [2]忻州师范学院计算机科学与技术系,山西忻州034000 [3]中国科学院地理科学与资源研究所,北京100101

出　　处：《计算机工程与设计》2019年第11期3331-3336,3348,共7页Computer Engineering and Design

基　　金：山西省中科院科技合作基金项目(20141101001);"十二五"山西省科技重大专项基金项目(20121101001);山西省科技攻关基金项目(20141039);山西省重点研发计划基金项目(201603D121031)

摘　　要：为解决传统车型识别方法提取特征信息单一、识别精度不高、效率低的问题,将卷积神经网络引入目标识别问题中,利用其清晰、高效的泛化能力完成车型的特征学习,围绕模型的框架结构设计和内部参数优化两个方面进行研究,提出一种基于改进的Alex Net网络模型。将循环神经网络与卷积神经网络融合嵌入二级框架,设计自定义池化方式并对参数更新过程方法进行合理组合,通过提取浅层和高层的组合特征保证训练过程输入信息的多样性,使特征表达更加精确,网络性能更加高效。将该模型应用于视频监控图像车型识别任务中,通过在BIT-vehicle数据集上的一系列对比实验验证了所提模型的有效性。To solve the problem that the traditional vehicle recognition method extracts single characteristic information with low recognition accuracy and efficiency,the convolutional neural network was introduced into the target recognition problem and its clear and efficient generalization ability was used to complete the feature learning of the vehicle.The framework structure design and internal parameter optimization of the convolutional neural network model were studied,an improved Alex Net network model was proposed.Recurrent neural network and convolutional neural network were integrated into a two-level framework,customized pooling method was designed and reasonable combination of parameter updating process method was conducted.The combination features of shallow and high-level layers were extracted to ensure the diversity of input information in the training process,making feature expression more accurate and network performance more efficient.The proposed model was applied to the video surveillance image vehicle identification task.The effectiveness of the proposed model was verified by a series of comparative experiments on the BIT-vehicle data set.

关 键 词：车型识别 Alex Net卷积神经网络 循环神经网络 特征融合 池化 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]