关键词: 卷积神经网络, 交通标志识别, VGG16, 卷积层, 池化层, 批量, 特征图, 正则化

Abstract:

In view of the problem of low recognition rate and long recognition time of the existing traffic sign recognition methods, based on convolution neural network method, an improved convolutional neural network model based on VGG16 is proposed. The number of convolution layers, convolution kernels and pooling layers of the VGG16 network model are modified to enhance the feature extraction ability and simplicity of the network model. The experimental datasets are enhanced by random rotation, scaling, offset and contrast adjustment, and the network model recognition rate is improved by activation function, mixed batch training and early termination regularization methods. In the experiment of the German traffic sign recognition benchmark, the recognition rate of the improved VGG16 network model is 98.98%, and the recognition time of single traffic sign is only 0.24 ms. Compared with other models, the proposed model has obvious advantages in recognition rate and recognition time.

Key words: convolutional neural network, traffic sign recognition, VGG16, convolutional layer, pooling layer, batch size, feature map, regularization