Algorithm for Small Defect Detection from Challenging Airport Runway Images Based on Bidirectional Feature Pyramid

Airport runway defect detection is indispensable to guarantee the safety of aircraft landing and taking-off, automating the inspection process has been the focus of airport management department. Defects such as crack, seam-broken, and corner fracture, which are common to airport runway, are characterized by extremely small sizes, similar shapes, and low contrast with the background, making the automatic detection task challenging. Thus, we propose an algorithm based on bidirectional feature pyramid to perform the automatic detection. The proposed network applies VGG-16 as backbone to extract features, reducing the information loss during down sampling to retain more detailed features of small defects. A bidirectional feature pyramid module is designed to enhance feature fusion, which can balance the feature information of small defects in different scales. The module expands the receptive field on the basis of Feature Pyramid Network (FPN), which can obtain the global semantic information of the target, and achieve better classification and localization performance. Compared with five existing deep learning methods, experimental results show that our proposed algorithm outperforms all counterparts, making the automatic detection and monitoring of airport runway more efficient.