In the actual traffic environment, the quality of the collected traffic signs is often influenced by the factors such as motion blur, background interference, weather conditions and shooting angles and so on, which poses a great challenge to the accuracy, real-time and robustness of traffic sign automatic identification. Owing to this, a classification recognition algorithm model of improved deep convolution neural network AlexNet is proposed. On the basis of the traditional AlexNet model, this model takes the traffic sign image data set GTSRB taken in the real scene as the research object, modifies the convolution kernels of all coiling layers to 3×3, in order to prevent and reduce the occurrence of over fitting, the dropout layer is added after two fully connected layers. In order to improve the accuracy of traffic sign recognition, two convolution layers are added after the fifth layer of the network model. The experimental results show that the improved AlexNet model is advanced and robust in traffic sign recognition.