The present invention discloses a convolutional neural network-based asphalt pavement crack classification and recognition method. Road cracks are classified according to different repair strategies of cracks of different widths and shapes; sample pictures are marked correspondingly and preprocessed so as to train a constructed convolutional neural network; and the trained convolutional neural network is adopted to classify crack information in pictures, and the severity levels of cracks are divided according to the widths and shapes of the cracks. Since the crack information in the pictures is automatically classified according to a pre-classification mode, and the severity levels of the cracks are divided, and therefore, efficiency of crack identification can be improved, road maintenance and repair work can be greatly facilitated; the convolutional neural network algorithm is used as a classifier to classify the road cracks; the convolutional neural network is a layered neural network composed of convolutional layers and sampling layers which are alternately distributed; and the convolutional neural network can learn features from training data implicitly and has greater advantages in the classification of cracks with no regular and significant features.