Mapping hydrothermally altered minerals with AST_07XT, AST_05 and Hyperion datasets using a voting-based extreme learning machine algorithm

Abstract

Detailed hydrothermally altered mineral mapping is important for mineral exploration. ASTER data are commonly combined with Hyperion data to classify the hydrothermally altered minerals. However, when machine learning algorithms are applied to the shortwave infrared (SWIR) bands of an ASTER reflectance image (AST_07XT), misclassification of Mg-OH group minerals is the major source of errors. In this study, an ASTER emissivity image (AST_05) and the AST_07XT SWIR bands are integrated to map minerals in the Duolong area, Tibetan Plateau. The results show that ASTER thermal infrared (TIR) bands can successfully identify Mg-OH group minerals. To improve the performance of classification, a novel voting-based extreme learning machine (V-ELM) algorithm is introduced to map hydrothermally altered minerals. The classification based on the ASTER SWIR-TIR data gets good identification of Mg-OH group minerals, which is better than those acquired using SWIR and TIR data alone. Moreover, these results also show that the AST_05 TIR bands cannot discriminate Al-OH group minerals. Compared with the classification applied to AST_07XT SWIR data, the classification applied to the ASTER SWIR-TIR data can achieve a higher overall accuracy (99.01%). The ASTER data results are spatially consistent with those of the Hyperion data. In accordance with the image processing results, a new deposit associated with felsic intrusions has been validated by field investigations. These schemes are promising for mineral mapping in the Tibetan Plateau.