IMAGE CLASSIFICATION USING ARTIFICIAL BEE COLONY ALGORITHM

Abstract

Remote sensing is the most important provider of the various data sources that are used in GIS. It has been globally used for knowledge elicitation of earth’s surface and atmosphere. Land cover mapping, one of the widely used applications of remote sensing is a method for acquiring geo-spatial information from satellite data. We have attempted here to solve the land cover problem by image classification using one of the newest and most promising Swarm techniques of Artificial Bee Colony optimization. Artificial Bee Colony algorithm (ABC) is a recent development and has emerged as an efficient global optimization technique. It is inspired from the intelligent foraging behaviour of honey bees. In this paper we propose an implementation of ABC for satellite image classification. Classification plays a very important role in image processing and with the increase in images being acquired and archived, optimal classification tool for different application domains is needed. The objective here is to utilize the bee communication and food search method of information exchange and hence achieve maximum classification accuracy. We also present a new dimension to analyze the efficiency of classification techniques. It is the feature extraction from heterogeneous regions. It is very important for any method to correctly identify all features in such regions. We have used the Heterogeneity factor of an image to analyse the actual efficiency of our method. Various techniques overlook this for overall classification accuracy. The results produced by artificial bee colony algorithm are compared with the results obtained by other traditional and soft computing techniques to show the effectiveness of our proposed implementation. We have analysed the feature extraction of Heterogeneous regions in an image for different soft computing techniques like Page | iv fuzzy, cAntMiner, hybrid ACO-BBO and shown that the accurate classification of such regions is independent of kappa coefficient as the accuracy assessment parameter. The results produced by artificial bee colony (ABC) algorithm are compared with the results obtained by other techniques like Minimum Distance Classifier (MDC), Maximum Likelihood Classifier (MLC), Biogeography Based Optimization (BBO), and Membrane Computing (MC), hybrid Flower pollination/ Bacterial Foraging (FPAB/BFO) and Fuzzy classifier to show the effectiveness of our proposed implementation. We go beyond the multispectral regime to the hyperspectral images and their dimensionality reduction problem. The scope of ABC has not been explored much yet but we can utilize the bee optimization in reducing the dimension of hyperspectral images. We have discussed the dimensionality reduction problem with analysis of various existing reduction techniques and propose a theory of Artificial Bee Colony dimension reduction for the same