Integrated Digital Mural Quality Assessment Using Deep Visual Features and Fuzzy Analytic Hierarchy Process

Digital murals face degradation issues like fading and cracking, requiring objective and interpretable evaluation methods. To overcome the subjectivity and low efficiency of manual assessment, the study proposes an improved deep neural network that maps extracted features into four criteria: color quality, texture details, structural integrity, and degradation degree. Using the fuzzy analytic hierarchy process (FAHP) with triangular fuzzy numbers, the study computes global weights and obtains a comprehensive score via weighted linear aggregation. Experimental results show the network reduces root mean square error to 0.102 and mean absolute error to 0.089. Inter-class distance increases by 33.8%, and t-SNE visualization shows four quality clusters in four-corner separation. The comprehensive score monotonically decreases from 0.924 to 0.376 as degradation worsens, with texture details showing the most significant drop. The model effectively captures mural quality deterioration, aligning with real-world conservation cognition. This work provides an objective, quantitative, and automated tool to assist experts in quality screening and restoration prioritization for large-scale murals.