In order to improve the efficiency and accuracy of pavement crack detection, the enhanced convolution neural network is introduced into the recognition of pavement crack image. First, the original image is preprocessed by linear gray transformation to reduce the influence of noise on recognition. After several steps, such as structure design, algorithm training and experimental sample testing, the pavement crack identification model is established. Finally, through the experiments in MATLAB, the recognition model can effectively identify the pavement cracks, and the recognition rate can reach 92.8%. Experimental results show that compared with other algorithms, the algorithm proposed in this paper has the advantages of high efficiency, accurate results, and can meet the engineering needs.