Research On Traffic Sign Recognition Method Based On Convolution Neural Network

With the development and application of unmanned in intelligent transportation system,traffic sign recognition become more and more attention,the traffic sign recognition accuracy will directly affect the traffic and pedestrian safety,many experts and scholars are committed to finding new ways to solve this problem,and constantly improve the accuracy of recognition.In recent years,the emergence of deep learning has also made a new way of thinking for the identification of traffic signs.In this paper,we use convolution neural network to explore the problem of traffic sign recognition.Based on the classic convolution neural network LeNet-5 model,we improve the network structure,and implement a high-performance traffic sign recognition algorithm based on this model.The main work of this article is as follows:(1)Construct the TSR-LeNet convolution neural network model for traffic sign recognition.In the LeNet-5 model based on the model,by adding a truncated normal distribution function,the random initialization strategy,discarding,activation learning rate optimization algorithm,in the connection layer increased after softmax layer,and through the experiment to determine the parameters of the network to construct a network model for the classical LeNet-5 model in traffic sign recognition is prone to overfitting insufficient.For further improvement of the TSR-LeNet model,to solve the softmax inner class in the classification process from the big problem,introducing center loss function on the softmax loss function is optimized,the training of higher cohesive network,to further improve the recognition accuracy of network.(2)Based on the TensorFlow framework,we carried out experiments on the augmented GSRTB German traffic sign dataset.Compared with the existing traditional recognition methods,the TSR-LeNet convolution neural network model implemented in this paper has higher classification accuracy.