Rapid detection of pavement cracks is important for road maintenance and rehabilitation, but the traditional crack detection method is time-consuming, labor-intensive and low accuracy. Therefore, an improved U-net neural network model is proposed in this study. By adjusting the model structure and fine-tuning parameters, the U-net model can accurately and automatically identify pavement cracks. In this paper, a new semi-automatic marking software is developed to label pavement cracks based on Canny edge detection and Otsu segmentation algorithms, and the labeled 2D laser images are used as the training dataset. In addition, data enhancement methods are used to augment the training database. In the experimental stage, the cross-entropy loss function is used to compute error differences between the predicted value and the true value based on Adam optimization algorithm. Findings show that the improved U-net model is better than the original U-net model and the fully connected neural network model in terms of detection accuracy and algorithm robustness. This study provides a solution for the rapid detection of pavement diseases, which will be beneficial to road maintenance management department which can rapidly take corrective measures to ensure road traffic safety.