Wavelet Entropy based Band Selection for Hyperspectral Images

Abstract

Hyperspectral imaging (HSI) provide rich spectral and spatial information providing a detailed information about the location but it also has disadvantages like the large data effects the classification leading to the curse of dimensionality. Band Selection (BS) is the best approach to lower the dimensions of hyperspectral images. This paper mainly focuses on band selection by Ranking bands based on wavelet entropy and forming a hyperspectral cube through bands. Each band represents a unique spatial information, which is initially filtered and combined through entropy-based wavelet selection procedure. In contrary to other approaches, the experiment shows promising results, and suggests that clustering of the bands is informative while reducing the dimensions since adjacent bands have strong correlations. By utilizing wavelet entropy, an image can be represented more efficiently with less storage and transmission space required. This procedure reduces the burden on classification and increasing its performance.