Traffic Sign Recognition Based On Improved VGG Net

In the field of pattern recognition,the detection and classification of targets is a very important research topic.The traffic sign recognition task extended from this has a high research value in real life.In recent years,the role of intelligent transportation systems(ITS)and driverless technology in urban transportation has become increasingly important.Especially for driverless cars,how to accurately identify traffic signs has an important guiding role for vehicle path planning and control.In 2006,Hinton et al.first proposed the concept of Deep Learning.Until recently,deep learning has successfully guided the direction of artificial intelligence.Convolutional Neural Network(CNN),as a very important model in deep learning,has achieved remarkable results in the field of image recognition.VGG Net is a convolutional neural network proposed in 2014,and achieved the second place in the ILSVRC 2014.This paper refers to the structure of VGG Net,and designs a new type of network structure for specific traffic sign recognition tasks.It is proved in experiments that this new type of network structure performs very well in the task of traffic sign recognition.The main work of this paper is as follows:(1)This paper proposes a new type of convolutional neural network,named I VGG(Improved VGG).The IVGG is designed to be 9 layers in view of the small size of the identified traffic sign image.Referring to the convolution and pooling operations used in VGG Net,the operations of Dropout and Batch Normalization are added to IVGG to improve the generalization ability of IVGG and effectively avoid over-fitting.Experiments show that the verification and testing on the data set have achieved the desired results(2)In academia,the mainstream data set used for research is the German Traffic Sign Recognition Benchmark(GTSRB),which contains 43 types of German traffic signs,totaling more than 50,000 images.However,there is a problem with the sample distribution of this data set.The sample size of these 43 types of traffic signs is inconsistent,and the difference is large,and the number of some traffic signs is also less.In order to make the experimental results more scientific and accurate,this topic enhances the data of the entire data set according to the idea of equalization,so that the sample size of each type of traffic sign is expanded to some extent,and the number of samples of each type in the newly generated data set.Consistent and provide a sufficient number of samples for training,further enhancing the recognition of the IVGG network model.