Fuzzy Classification of Semi-urban Features from IRS Satellite Imagery

Abstract

In remote sensing images, a pixel might represent a mixture of class covers, within class variability or other complex surface cover patterns that cannot be properly described by one class. So in order to map a scene’s natural fuzziness or imprecision and to provide more complete information through image analysis, a fuzzy logic based classification procedure is necessary. This fuzzy logic is a knowledge based method which makes no assumption about statistical distribution of the data and therefore reduces classification inaccuracies. Also fuzzy logic is interpretable and can combine expert knowledge and training data. Major advantage of fuzzy is that it allows natural description in linguistic terms of problems that should be solved rather than in terms of relationship between precise numerical values. Hence this paper aiming to study fuzzy classifier as an alternative approach to traditional classification techniques for RS data to classify urban features from satellite image. The ERDAS IMAGINE V9.2 remote sensing software is used in this study. The accuracy assessment was conducted based on Overall Classification Accuracy (OCA) and Kappa Statistics. This experiment was conducted using Erdas Imagine V9.2 RS software. Finally, the suitability of Fuzzy classification is verified on Arasikere Semi-urban area and Overall Classification Accuracy 65.83% and 71.85% was obtained for 360 and 720 training sites with 120 validation sites respectively. By increasing validation sites to 240 for 720 training sites, OCA of 73.33% was achieved.