DCDRNet: Detail–Context Decoupled Representation Learning Network for Efficient Crack Segmentation | Synapse