Dense Multiscale Feature Learning Transformer Embedding Cross-Shaped Attention for Road Damage Detection

Road damage detection is essential to the maintenance and management of roads. The morphological road contains a large number multi-scale features, which means that existing algorithms are unable effectively distinguish fuse multiple features. In this paper, we propose dense multiscale feature learning Transformer embedding cross-shaped attention for (DMTC) network, can segment information in images improve effectiveness detection. Our DMTC makes three contributions. Firstly, adopt mechanism expand perceptual field extraction, its global improves description network. Secondly, use module integrate local at different scales, so able overcome difficulty detecting targets. Finally, utilize multi-layer convolutional segmentation head generalize previous get final result. Experimental results show our network could pavement pothole patterns more accurately than other methods, achieving an F1 score 79.39% as well OA 99.83% on cracks-and-potholes-in-road-images-dataset (CPRID).