Pavement Crack Identification Based on Deep Learning and Denoising Model

Based on the current classification of asphalt concrete, cracks can only be classified as cracked and no cracking. And most of the crack extraction methods can only be applied on cement concrete at present. To improve this situation, an improved crack classification method based on deep learning is proposed, and an improved crack extraction method based on multilevel image denoising is raised. The improved crack classificatory method can divide the cracks of concrete pavement into no cracking. Transverse crack, longitudinal crack, and reticular cracks. The improved crack extraction method developed a multilevel image denoising frame by using many image denoising methods. The experiment result shows that the accurate rate of crack classification is 97.3%, 98.6%, 97.4%, 97.1%, respectively. And the average accuracy rate is 88.49%. The improved crack extraction method can extract cracks on the asphalt concrete, even if it has a lot of distractions.