Automatic segmentation of airport pavement damage by AM‐Mask R‐CNN algorithm

The airport is an important infrastructure for air transport and urban traffic. pavement damage seriously affects the safety of aircraft take-off landing. Therefore, regular detection critical landing safety. However, small target areas, existing methods cannot effectively achieve detection. In addition, carried out under low light conditions at night, may not be able to accurately detect boundary. To address above problems, automatic algorithm Mask R-CNN integrating attention mechanism (AM-Mask R-CNN) proposed. First, AM-Mask developed based on improved modified by feature pyramid (FPN) dual are fused extract subtle features in images. fusion used. macroscopic microscopic organically combined improve segmentation sensitivity areas dark lighting. experimental results show that average F1-score proposed model 0.9489. mean intersection over union 0.9388. speed can reach 11.8 FPS. Moreover, compared with traditional threshold method Fully Convolutional Networks (FCN) DeepCrack method, effectiveness further proved.