|
In the investigation, protection and restoration of murals, the exact location and size of fragmentation disease can be labeled to facilitate the subsequent protection and cultural heritage of murals. However, manual labeling is time-consuming and laborious, and the results will be various due to the different experience of experts, which is not conducive to the promotion of intelligent cultural relic protection and restoration. The intelligent labeling of mural diseases through artificial intelligence can greatly improve the efficiency of mural restoration and solve these deficiencies. Therefore, an intelligent labeling method for mural fragments based on gradient-trainable Gabor and U-Net is proposed. In this paper, the disease labeling problem is transformed into the image segmentation problem for disease regions. However, due to the rich texture of the mural image and the complex edges of the fragmentation regions, a lot of detail is lost for disease labeling directly using the U-Net network. Different from previous studies, this method uses gradient-trained Gabor in the encoders to extract texture features of fragmentation disease and obtain more texture information of the disease region. In particular, res2convolution is embedded into the skip connections to narrow the semantic gap between encoder and decoder and better inject the texture information of fragmentation disease into the deep network. Finally, we proved that the method proposed in this paper can realize the intelligent labeling of fragmentation diseases accurately and efficiently through the murals of Han Tomb at Xi 'an Jiaotong University.
|
