To improve the accuracy and real-time performance of the intelligent assisted driving system in estimating the road adhesion coefficient, a deep learning algorithm for road recognition based on visual information was devised to achieve the pre-estimation of the road adhesion coefficient.A compression convolution mechanism is designed to reduce the network operation parameters.Further,the feature map global average is used to replace the fully connection layer to improve the fitting performance of the network.Also, a pavement recognition depth convolution neural network DW-VGG is constructed.The self-built pavement image dataset is used to train the network. The test results show that the DW-VGG network based on the proposed multi-layer knowledge distillation algorithm has a high recognition accuracy, the score of classification performance evaluation index F1 score is 96.57%, reduces the time and space costs of the network effectively, it only takes 32.06ms to identify a single image, and the prediction model is only 5.63M.