Region Expansion of a Hyperspectral-Based Mineral Map Using Random Forest Classification with Multispectral Data

Observation images from hyperspectral (HS) sensors on satellites and aircraft can be used to map minerals in greater detail than those multispectral (MS) sensors. However, the coverage of HS is much less that MS images, so there are often cases where cover entire area interest while only a part it. In this study, we propose new method more reasonably expand mineral an image with such cases. The uses various indices sensor’s band values as input image-based classes output. Random forest (RF) two-class classification then applied iteratively determine distribution each turn, starting most consistent map. also involves correction misalignment between selection variables by RF multiclass classification. was evaluated comparison other methods Cuprite area, Nevada, using Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) Hyperspectral Imager Suite (HISUI) Advanced Spaceborne Thermal Emission Reflection Radiometer (ASTER) As result, all region-expansion HS–MS pair, including proposed method, showed better performance image. had highest performance, inter-mineral averages F1-scores for overlap non-overlap areas were 85.98% 46.46% AVIRIS–ASTER pair 82.78% 42.60% HISUI–ASTER respectively. Although region lower region, high precision accuracy almost minerals, few pixels. Misalignment factor degrades requires precise alignment, but could suppress effect misalignment. Validation studies different regions will carried out future.