Hierarchical existential prior based on expanded pseudo-label for crack detection

Road crack detection approaches based on the image processing technique have attracted much attention during the past decade due to their convenience and efficiency, but most of them cannot achieve the expected performances due to the complex background interference and severe category imbalance of road images. This paper presents a hierarchical existential prior based on an expanded pseudo-label for crack detection. In particular, the framework contains three variants of U-Net, and each sub-network is trained by pseudo-labels generated by transforming semantic categories of non-crack pixels distributed in the neighborhoods of crack ones. Notably, the expansion degrees of labels for three sub-networks are set in hierarchical descending order. In other words, the crack samples of pseudo-labels for the latter sub-network are a subset of pseudo-labels for the former one, and we define it as an existential prior, which can optimize the network in a coarse-to-fine fashion and refine the detection result gradually. In addition, we utilize a hybrid loss consisting of IoU, SSIM, and focal loss to optimize the network in different aspects, including image-aspect, patch-aspect, and pixel aspect in the training phase, which can improve the structural representation capability of the model. In addition, we present a dynamic hyper-parameter adjustment strategy to balance the weight coefficients of different loss terms, which can enhance the robustness of the model for various practical scenes. Finally, the proposed method achieves 11.36%, 29.76%, and 26.73% in terms of Fβ on CrackTree200, Crack Forest, and ALE datasets, respectively, which sufficiently demonstrate its effectiveness and superiority.