Abstract: To address the challenge of accurately detecting disease areas in ancient murals with existing methods, a novel ancient mural disease detection model based on a residual dual channel attention U-Net is proposed. Firstly, a residual dual channel module is designed to replace the encoder and decoder in U-Net, constructing a network capable of multi-resolution analysis to detect mural disease areas of varying scales against complex backgrounds. Then, a multi-scale attention gate is incorporated to fuse complementary features from both high and low levels, enabling the network to emphasize salient features of mural disease areas. Finally, a mixed domain attention module is designed to suppress the interference of mural background information, further accurately locating mural disease areas. Additionally, a multi-stage loss addition approach is employed to improve the performance of the network model. Experimental results on the Dunhuang Mogao Grottoes mural dataset and the Yunnan ethnic minority mural dataset demonstrate that the proposed model outperforms other comparative methods in visual perception, achieving F-score metrics of 0.807 7 and 0.728 9, respectively, both higher than those of other comparative methods.