Weathering is one of the most common causes of building sandstone damage. The evolution of building sandstone in various weathering behaviors is critical for research. An intelligent assessment approach for classifying weathering degree of building sandstone in a humid environment is presented in this study. This synthesis method relates to three parts: microscopic observation of weathering characteristics, hyperspectral acquisition of weathered samples, and machine learning technology for a classification model. At first, weathering process is divided into initial weathered stage, accelerated weathered stage, and stable weathered stage according to the causes and mechanisms of weathering. Secondly, a novel classification method of weathering degree is proposed based on the weathering stage. Then, the mapping relationship between microscopic characteristics and hyperspectral image of shedding samples can be established in the visible and near-infrared spectral ranges (400–1000 nm) according to the change law of spectral absorption feature. Next, the spectral data of building sandstone with different weathering degrees are classified using Random Forest model. Furthermore, the hyperparameters of Random Forest model are optimized by Gray Wolf Optimizer algorithm for better performance. The trained model is finally applied to evaluate the weathering degree of large-scale sandstone walls quantitatively. The whole weathering assessment process is worth recommending for diagnosing and monitoring the building sandstone.
Read full abstract