Multispectral Satellite Image Segmentation Using Improved Bat Algorithm

This paper is mainly intended to use Bat Algorithm (BA) – based clustering approach for classifying multispectral satellite images. This clustering algorithm provides the partitions that are both bound and self-determined. But the drawbacks of traditional K-means algorithm are: i) it converges easily to a local optimum and ii) identifying the number of optimal clusters is a challenging task. The researchers have tried to unravel these issues by initializing the cluster centers from a priori information. In this paper, we have tried to solve the issues of traditional k-means algorithm by applying the bat algorithm on it. The proposed modified bat algorithm is used to segment the satellite images and extract the useful information from it. In the proposed algorithm, clusters with minimum inter-cluster distance and lesser than the given threshold are identified and merged. This process is repeated till all the inter-cluster distances are greater than the given threshold. The enhancement in the performance of the proposed improved bat algorithm is evident in the experimental results.