Deep Learning based Pavement Crack Detection System

Abstract The pavement crack causes the highway service life to shorten, safety hidden danger increase. low efficiency and high cost of manual inspection makes it difficult detect cracks. This paper proposes a fast efficient deep learning detection system. CRACK2000, an image segmentation dataset with complex interference background multiple types, is constructed based on perspective transformation cropping. scheme corrects images by transformation. extraction depth features completed applying U-Net network. Finally, condition index PCI (pavement index) calculated quantifying different types information results. experimental results show that Precision, Recall, F1-score AUC network are 76.67%, 72.32%, 74.43% 99.46% respectively. values reflect method more capable filtering out from cracked images. automatic system designed in this can accurately locate classify location category cracks, perform quantitative evaluation obtain deterioration road section corresponding repair recommendations, enhancing practicality detection.