A Rapid Intelligent Screening of a Three-Band Index for Estimating Soil Copper Content

Research has widely validated three-band spectral index as a simple, valid, and highly accurate method of estimating the copper content of soil. However, selecting the best band combination from hundreds of thousands, even millions of candidate combinations in hyperspectral data, is a very complicated problem. To address this issue, this study collected a total of 170 soil samples from the Aktas copper-gold mining area in Fuyun County, Xinjiang, China. Then, two algorithms including Competitive Weighted Resampling (CARS) and Stepwise Regression Analysis (STE) were applied to pick the bands from the original and first-order derivative spectra, respectively. A three-band index model was developed using the selected feature bands to estimate soil copper content. Results showed the first-order derivative spectrum transforms the spectral curve into a sharper one, with more peaks and valleys, which is beneficial for increasing the correlation between bands and copper content compared with the original spectrum. Moreover, integrating first-order derivative spectroscopy with CARS makes it possible to precisely identify key spectral bands and outperforms the dimensionality-reduction capabilities compared with the integration of STE. This strategy drastically reduces the time spent screening and is proven to have similar model accuracy, as compared to the individual group lifting method. Specifically, it reduces the duration of an 8 h task down to a mere 2 s. An intelligent screening of three-band indices is proposed in this study as a method of rapidly estimating copper content in soil.