基于轻量WACNN的交通标志识别  

Traffic Sign Recognition Based on Light WACNN

作　　者：黄知超[1] 李栋[1] Huang Zhichao;Li Dong(Guilin University of Electronic Technology,Guilin,Guangxi 541004,China)

机构地区：[1]桂林电子科技大学,广西桂林541004

出　　处：《应用激光》2019年第1期119-123,共5页Applied Laser

基　　金：广西科技开发攻关资助项目(项目编号:14122007-1);广西教育厅资助项目(项目编号:2015JGA202)

摘　　要：现有交通标志识别技术,存在高识别率高功耗或者低识别率低功耗的问题,构建了新轻量级WACNN的识别算法。首先,利用TensorFlow构建6层卷积神经网络,其中前三层为卷积池化层,四层为1×1卷积层,五层为全连接层,六层为输出层,前四层再加入批量归一化方法。其次,使用直方图均衡对交通图像预处理。最后,模型在GTSRB上进行实验,实验结果表明,所提模型不仅极大缩短了训练时间,且识别准确率也能达到了97%。The existing traffic sign recognition technology has the problems of high recognition rate,high power consumption or low recognition rate and low power consumption.Aiming at this problem,a new light WACNN of recognition algorithm is constructed.Firstly,six layers convolutional neural network are constructed by using TensorFlow,in which the first three layers are convolutional pooling layers,the fourth layer is 1×1 convolutional layer,the fifth layer is fully connected layer,the sixth layer is output layer,and the first four layers are then added batch normalization method.Secondly,histogram equalization is used to preprocess traffic sign images.Finally,the model is tested on GTSR.The experimental results show that the proposed model not only greatly shortens the training time,but also the recognition accuracy can reach 97%.

关 键 词：交通标志识别 卷积神经网络 批归一化 图像预处理 

分 类 号：TP319[自动化与计算机技术—计算机软件与理论]