Search

Diagnostic absorption features can indicate the existence of specific materials, which is the foundation of mineral analysis with optical remote sensing data. In hyperspectral data processing, the most commonly used method to extract absorption feature, is Continuum Removal (CR). As for multispectral data, Principle Component Analysis and other indirect methods were used to extract absorption information, and little research has been done on full-band absorption feature extraction. Classification of similar minerals is one of the major difficulties in mineral spectral analysis, while there is no valid index for spectral difference between similar mineral groups. Absorption feature extraction may improve the classification accuracy, but there is no research to investigate the impact of absorption feature extraction on spectral difference between similar minerals. This paper summarized the principle of mineral spectral difference, and proposed the concept of Class Separability Ratio (CSR), which was verified to be a valid index for spectral difference between similar mineral categories. Through comparison experiments on alunite and kaolinite spectra, including USGS spectral library spectra and resampled spectra in accordance with the band settings of HYPERION, ASTER and OLI, the impact of absorption feature extraction on spectral difference between similar minerals were investigated. Experimental results show that valid absorption feature extraction can greatly enhance the spectral difference between similar minerals, and the spectral difference is positively correlated with spectral resolution. Besides, the results of CR can be severely affected by spectral resolution and band center positions, and the absorption feature spectra extraction results for multispectral datasets need to be improved. This research laid the foundation of precise identification between similar mineral categories, and provided valuable reference for the band settings of future geology remote sensing sensors.

The precise analysis of mineral abundance is a key difficulty in hyperspectral remote sensing research. In the present paper, based on linear spectral mixture model, the derivative of ratio spectroscopy (DRS) was introduced for spectral unmixing of visible to short-wave infrared (Vis-SWIR; 0.4 - 2.5 microm) reflectance data. The mixtures of different proportions of plaster and allochite were analyzed to estimate the accuracy of the spectral unmixing model based on DRS. For the best 5 strong linear bands, the Pearson correlation coefficient (PCC) of the abundances and the actual abundances were higher than 99.9%, while the root mean square error (RMSE) is less than 2.2%. The result shows that the new spectral unmixing model based on DRS is simple, of rigorous mathematical proof, and highly precise. It has a great potential in high-precision quantitative analysis of spectral mixture with fixed endmembers.