BAND SELECTION USING 3D REGION GROWING ALGORITHM FOR HYPERSPECTRAL IMAGE ANALYSIS

Abstract

Advancements in sensor technology have significantly increased the importance of hyperspectral imaging (HSI) in various computer vision applications for remote sensing. Modern HSI sensors provide unmatched spectral resolution by capturing images from satellites and drones, encompassing the electromagnetic spectrum from 400 to 2500 nanometers. This study explores how to use the abundant spectral data effectively, specifically addressing the issue caused by the high dimensionality of HSI data. We propose a new method that improves the selection of spectral bands for better segmentation accuracy through the use of a 3D Region Growing Algorithm (RGA). Unlike other selection methods that primarily identify statistically distinct bands, our approach focuses on heuristically searching for the most informative bands. This approach introduces a flexible stopping rule dependent on seed pixel intensity, providing precise control over segmentation by adjusting to different image contrasts. Our approach combines spatial and spectral information to achieve context-aware segmentation. This method has been proven to be effective in various datasets like Salinas, Indian Pines, Pavia Centre, and a real-world dataset, showing its potential in remote sensing.