Deep Learning-Based Detection and Classification of Moss and Crack Damage in Rock Structures for Geo-Mechanical Preservation

Abstract

Stone monuments, valuable cultural assets that inspire thoughts of expression and legacy, necessitate effective preservation strategies. This research proposes a novel approach utilizing Deep Learning for automated detection, classification, and evaluation of moss and crack damage. The methodology involves fusing Luminance with Gray-Level Co-occurrence Matrices (GLCM) for data analysis. Monument decay is then classified using a Multi-Layer Neural Network (MLNN) and rigorously validated using metrics such as precision, selectivity, sensitivity, and accuracy. Damage assessment employs indices and categories, with spread parameters evaluating crack damage and colour change levels assessing moss damage. The estimated damage severity guides the design of appropriate recovery and preservation plans. Notably, this innovative method achieves a high classification accuracy of 97% for monument decay, demonstrating its potential for significantly improving the longterm preservation of these invaluable cultural treasures. Major Findings: This research presents a novel deep learning-based model for automated detection and classification of moss and crack damage in stone monuments, achieving a high classification accuracy of 97%. The model effectively assesses decay parameters using an MLNN, demonstrating superior performance compared to existing methods like Local Binary Patterns, with high G-mean, Precision, Specificity, Accuracy, F-measure, and Sensitivity values. The model provides valuable insights for monument preservation by generating damage maps, identifying damage categories and indices, and estimating damage severity, guiding the development of effective recovery and preservation plans.