Automatic Defect Detection of Pavement Diseases

Pavement disease detection is an important task for ensuring road safety. Manual visual requires a significant amount of time and effort. Therefore, automated identification technique required to guarantee that city tasks are performed. However, due the irregular shape large-scale differences in diseases, as well imbalance between foreground background, challenging. Because this, we created deep convolution neural network—DASNet, which can be used identify diseases automatically. The network employs deformable instead regular feature pyramid’s input, adds same supervision signal multi-scale features before fusion, decreases semantic difference, extracts context information by residual enhancement, reduces loss top-level map. Considering unique problems background common, therefore, introduce sample weighted function. In order prove superiority effectiveness this method, it compared latest method. A large number experiments show method superior accuracy other methods, specifically, under COCO evaluation metric, with Faster RCNN baseline, proposed obtains 41.1 mAP 3.4 AP improvement.