Intelligent Identification, Characteristics Extraction and Driving Factors of Scaling off on Earthen Sites: A Comparison of K-Means, Fuzzy C-Means, and Experience-Based Methods

Scaling off is a prevalent deterioration on earthen sites, evolving through crust nucleation, maturation, and detachment. Accurate identifying and evaluating scaling off is crucial for conservation of earthen sites. However, traditional methods are subjective, labor-intensive, and intrusive. In this study, an intelligent assessment system was developed to diagnose scaling off development degrees, guiding method selection in terms of accuracy, cost, and preservation needs. Using an experience-based method as reference, K-means and Fuzzy C-means were compared on medium-scale wall point cloud datasets and validated with multi-temporal data. Results have shown that the framework is applicable to the degree assessment of scaling off, achieving an accuracy above 80%, and maintaining an average absolute error below 5%; K-means demonstrated broader applicability, while Fuzzy C-means exhibited limited robustness in early-stage degradation. The transition rates of three stages are not uniform. Mature crust units had the largest area, longest perimeter, and lowest crack density; soil moisture migration is the key driver influencing the severity of scaling off. These findings support comprehensive assessment, precise measurement, and targeted conservation, while being adaptable to other deterioration forms, promoting sustainable preservation of cultural heritage.