Hyperspectral Prediction Model for Mixed Salts on the Surfaces of Temple Murals in Northern China Based on Simulated Samples

Abstract. Murals are invaluable treasures of human civilization, yet they face irreversible damage from salt migration and crystallization caused by environmental fluctuations. This study develops a hyperspectral approach to monitor common salts (anhydrous sodium sulfate (Na₂SO₄) and anhydrous calcium chloride (CaCl₂)) on mural surfaces. Using an ASD-FieldSpec4 spectrometer, we analyze simulated mural samples with varying salt concentrations. For data correction, spectral preprocessing was conducted using Savitzky-Golay smoothing (SG), Standard Normal Variate (SNV), and Multiplicative Scatter Correction (MSC). Additionally, the natural logarithm (NL) transformation was introduced and combined with spectral preprocessing to enhance spectral features. The Competitive Adaptive Reweighted Sampling (CARS) algorithm selected characteristic wavelengths, and the Savitzky-Golay-Natural Logarithm-Multiple Linear Stepwise Regression model demonstrated the best predictive capability. For Na₂SO₄ prediction, the model achieved optimal performance with an R² of 0.934 and an RMSE of 0.0678% using characteristic bands at 550 nm, 560 nm, 817 nm, 1318 nm, and 1911–2349 nm. For CaCl₂ detection, it showed excellent accuracy with an R² of 0.987 and an RMSE of 0.0162% at key bands of 404–448 nm, 812 nm, 1137 nm, 1314–1333 nm, and 1922–2465 nm. We predicted multiple salt concentrations on the simulated temple murals using a method based on the NL and the CARS algorithm. We demonstrate a technical approach for hyperspectral detection of mixed salts on the surfaces of typical temple murals in northern China.