Hyperspectral Emissivity Estimation and Application Based on Machine Learning Method

Land surface emissivity (LSE) can reflect the unique spectral characteristics of land surfaces, which holds significant research value and promising applications. However, due to the limitations of spaceborne hyperspectral thermal infrared (TIR) sensors, there is currently a lack of high spatial resolution LSE imagery. Therefore, this study proposes a methodology based on machine learning and mathematical statistical approaches to extend multispectral emissivity values to hyperspectral LSE spectrums. Furthermore, the Spectral Angle Mapper (SAM) method was used for target identification. The results indicated that the proposed method can estimate hyperspectral LSE curves in high accuracy. Moreover, the target ground class can be identified accurately by matching the spectral angle with the measured LSE curve. This has crucial implications for applications such as land cover classification, mineral identification, geological exploration and soil property research.