智能交通中的车型识别技术研究  

Research on vehicle identification technology based on deep learning

作　　者：马秀 谭福奎[2] 李震 李良荣 Ma Xiu;Tan Fukui;Li Zhen;Li Liangrong(School of Big Data and Information Engineering,Guizhou University,Guiyang 550025,China;Department of Physics and Electronic Science,Xingyi Normal University for Nationalities,Xingyi 562400,China)

机构地区：[1]贵州大学大数据与信息工程学院,贵州贵阳550025 [2]兴义民族师范学院物理与电子科学系,贵州兴义562400

出　　处：《电子技术应用》2020年第5期68-73,共6页Application of Electronic Technique

基　　金：国家自然科学基金项目(61361012);贵州省科技计划项目(黔科合平台人才[2017]5788号);2017年度兴义民族师范学院教授基金项目(17XYJS26)。

摘　　要：车型识别技术在智能交通系统中发挥着重要的作用,研究了一种基于快速RCNN的车型识别算法。首先通过目标检测网络进行图像检测,并利用卷积特性进行特征提取,经过RPN算法得到建议区域;然后基于框架区域的快速卷积神经网络对车辆进行精细化检测,并分割目标区域;最后基于Softmax-SVM集成模型实现车型的识别。实验表明,该算法可以降低计算的多重性,并提高准确度、灵敏度、特异度和精密度。Vehicle recognition technology plays an important role in intelligent transportation system.This paper studies a vehicle recognition algorithm based on fast RCNN.Firstly,image detection is carried out through the target detection network,and feature extraction is carried out by using the convolution feature.Then,based on the fast convolutional neural network of frame region,the vehicle is precisely detected and the target region is segmented.Finally,based on Softmax-SVM integration model,the identification of vehicle models is realized.Experiments show that the proposed algorithm can reduce the multiplicity of calculation and improve the accuracy,sensitivity,specificity and precision.

关 键 词：快速RCNN 车型识别 Softmax-SVM 识别率 RPN 

分 类 号：TN919.8[电子电信—通信与信息系统]