Hyperspectral Image Segmentation and Entropy Calculation for Qualitative Analysis of Grassland

Hyperspectral imaging provides high spectral resolution images of a scene in hundreds of narrow spectral bands. This technique proves to be very useful in Earth Observation applications such as land cover mapping, agriculture crop health assessment and other remote sensing tasks. The qualitative analysis of grassland areas is manually performed by computing the Shannon-Weaver biodiversity index. In this paper we propose a semi-automatic estimation of this biodiversity index in remotely-sensed hyperspectral images. Starting from a spectral reflectance curve chosen to be representative for grassland, we perform the image segmentation based on histogram thresholding of spectral angle mapper (SAM) values. We then compute the entropy for the pixels belonging to the segmented grassland areas. In order to apply the classic Shannon definition of entropy, we perform a clustering for data dimensionality reduction. The evaluation of the proposed method is performed considering a manually generated ground truth.