Mineral Identification and Mapping by Synthesis of Hyperspectral VNIR/SWIR and Multispectral TIR Remotely Sensed Data With Different Classifiers

Hyperspectral data, which have fine continuous spectrum, have been recognized to be more suitable for the detailed identification and classification of land surface, especially for minerals. The combination of the hyperspectral visible/near-infrared (VNIR) and shortwave infrared (SWIR) data with the hyperspectral thermal infrared (TIR) data is proven to be an effective way. However, how those effects are and what are the effects of introduction of multispectral TIR data on the minerals identification and classification are not well studied. To fully evaluate those effects, this article tries to use both simulated data and real data to testify the practicability of introduction of multispectral TIR data for the accuracies of mineral identification and classification. Four classifiers, i.e., spectral angle mapping, spectral feature fitting, orthogonal subspace projection, and adaptive coherence/cosine estimator, are selected in the experiment. Compared with the results using hyperspectral data alone, the introducing of multispectral TIR data in identification and classification has improved accuracies for both the simulated and real data. The overall accuracies are improved about 4%-13% for the simulated data and about 1%-5% for the real data by using different classifiers. Those improvements prove that the spectral diagnosed characteristics in TIR region even for multispectral data help identify and classify minerals. Although the improvements for real data are not well obvious due to the low spatial resolution, the multispectral TIR data are still effective supplements for hyperspectral VNIR and SWIR data in mineral identification and classification.