Algorithm to select Optimal Spectral Bands for Hyperspectral Index of Feature Extraction

Abstract

Background/Objectives: Water body management and food-grains supply will be challenging tasks. Selecting spectral bands for accurate area estimation of water body and selected vegetation (crop) is the objective of this study. Methods/ Statistical Analysis: Indices such as Normalized Difference Water Index (NDWI), Normalized Difference Vegetation Index (NDVI) are used for extracting water and vegetation respectively from other features. Generally bands providing high index values for the target are utilized for extraction. In this study the bands giving high index value for selected target and low index for surrounding are selected. Bands selected in this method provide better extraction and accurate area estimation. Findings: The NDWI and NDVI are based on multispectral data and have less number of combinations for band sets selection. Though the proposed method is derived from NDWI, it utilizes the Hyperspectral data that has narrow and hundreds of bands. Another advantage of this method is it utilizes the index value of target and surrounding features. It selects suitable band set by iteration method and provides accurate extraction of the targets and area estimation. The performance of the bands selected in this method was tested with coastal and inland water bodies. Area estimated with this method matches with NDVI and MNDWI values. Applications/Improvements: This method selects suitable bands to estimate area of water body and vegetation. This estimation will be useful for water body management and food production forecasting.