Hybrid convolution and transformer network for coupler fracture failure pattern segmentation recognition in heavy-haul trains

•A failure pattern automated recognition method named HCT for heavy-haul train couplers fractures is proposed.•A dataset (DJTU-Seg) for coupler fracture recognition was created, which includes 5026 images.•A multi-scale loss function is proposed, it improves the network performance without increasing the forward reasoning time.•Experiments demonstrate that the proposed HCT yields competitive performance on both the DJTU-Seg and DWTT-Seg datasets.