Enhancing Pavement Distress Detection Using a Morphological Constraints-Based Data Augmentation Method

Pavement distress data in a single section usually presents long-tailed distribution, with potholes, sealed cracks, and other distresses normally located at the tail. This distribution will seriously affect performance robustness of big data-driven deep learning detection models. Conventional augmentation algorithms only expand amount by image transformation fail to enlarge diversity. Due such drawback, this paper proposes novel two-stage pavement pattern, which mask is generated randomly according geometric features first stage; second stage, distress-free fused transformed into image. Furthermore, two convolutional networks, M-DCGAN MDTMN, are designed complete generation task stages separately. In comparison algorithms, quality diversity results proposed better than algorithms. addition, tests conducted indicate that expanded dataset can raise IoU from 48.83% 83.65% maximum, augmented algorithm contributes more performance.