Rapidity non-destructive detection of pavement surface diseases has become the main method of highway detection. In this paper, the Crack Pattern identification with digital image processing technology and Machine Learning were proposed to effectively train, learn, and recognize three different types of pavement crack morphological. Crack Feature attributes of segmented crack images were extracted based on feature engineering as input information of Support Vector Machine(SVM). This method has a satisfactory classification accuracy, and it does not require extensive training sample crack images. This paper has shown that pre-processing has the advantage of enhancing Crack Pattern information significantly and simplify the input information to improve the classifier performance based on a combination of digital image processing and SVM. For normal photos of pavement cracks with obvious background noises, the classification accuracy is only 0.4693, and it is lower than 0.9236 after the extraction of crack morphology. Besides, we also compared the model accuracy of different optimal parameters and the training time. It is found that the parameters which can achieve the highest verification classification accuracy in the cross-validation algorithm have better identification and classification results.

CICTP