RESEARCH ON DETECTION METHOD OF SALINITY IN SALINE-ALKALI SOILS BASED ON FUSION DATA

Abstract

Efficient monitoring of soil salinity is of significant importance for the utilization and ecological restoration of saline–alkali soils. This study investigates typical saline–alkali soils by employing hyperspectral imaging and machine vision technologies to develop a quantitative prediction model for soil salinity. First, spectral and image data of soil samples were acquired. Subsequently, spectral preprocessing was performed using Standard Normal Variate (SNV), Multiplicative Scatter Correction (MSC), and Moving Average smoothing (MA). Characteristic spectral bands were extracted using the Competitive Adaptive Reweighted Sampling (CARS), Variable Combination Population Analysis (VCPA), Iteratively Retaining Informative Variables (IRIV), and a combined VCPA–IRIV approach. Finally, the selected spectral features were fused with image features to establish a Support Vector Regression (SVR) model. The results demonstrated that, compared with single-source data, the SVR model based on feature-level data fusion significantly improved the prediction accuracy of soil salinity. Among the tested models, the SNV + VCPA–IRIV + SVR combination achieved the best performance (Rc² = 0.9889, RMSEC = 0.4790, Rp² = 0.9569, RMSEP = 1.0484, RPD = 3.4423). The results indicate that feature-level data fusion combining hyperspectral and machine vision data can effectively predict soil salinity in saline–alkali soils, while improving model stability compared to single-data-source approaches.