Performance governing factors of biogeography based land cover feature extraction: An analytical study

In recent years, nature inspired remote sensing image classification has become a global research area for acquiring the geo-spatial information from satellite data. The findings of recent studies are showing strong evidence to the fact that various classifiers perform differently when applied to images having different natural terrain features. This paper is an analytical study and a performance based characterization of the most recent nature inspired image classification technique i.e. Biogeography based Optimization (BBO) that has been used for focused land cover feature extraction [6]. The paper explores the behavior of BBO over different terrain features of a multi-spectral satellite image and establishes the fact that the classification efficiency of BBO for a given land cover feature is proportional to the degree of disorder of the Digital number (DN) values of the pixels comprising that land cover feature when viewed in any of the bands of the multi-spectral satellite image. More precisely, the classification efficiency of BBO on a terrain feature is inversely proportional to the entropy for that feature when viewed in any of the bands of the multi-spectral satellite image. For verification, we calculated the entropies for each of the land cover feature in two bands and found the same results in both the bands, which proves our proposed concept. The dataset on which the proposed concept is demonstrated is the 7-band cartoset satellite image of size 472 × 576 pixels of the Alwar region in Rajasthan. The results indicate that BBO is able to classify the homogeneous regions i.e. the regions with the lower entropy, more efficiently than the regions which show a greater degree of heterogeneity, i.e. higher entropy.