AggCrack: An Aggregated Attention Model for Robotic Crack Detection in Challenging Airport Runway Environment

Haifeng Li, Jianping Zong, Rui Huang, Zhongcheng Gui, Dezhen Song

Abstract: Crack detection is essential for guaranteeing airport runway structural reliability. An efficient solution we take is to employ a robot equipped with a camera to perform inspection task. However, automatic crack detection for airport runway is challenging, as the runway surface is seriously polluted by fuel stain and aircraft wheel mark, and the cracks need to be detected luare usually extremely thin. Thus, we propose a CNN model, AggCrack, to perform the crack detection task. AggCrack adopts an innovative semantic-level attention mechanism on the edges of the targets to focus the model on crucial features, and combines edge information and semantic segmentation for more accurate crack detection. We have implemented the algorithm and have it extensively tested on an airport runway dataset collected by our inspection robot from four different airport runways. Compared with four existing deep learning methods, experimental results show that our algorithm outperforms all counterparts. Specifically, it achieves the Precision, Recall and F1-measure at 84.24%, 70.36% and 76.68%, respectively.