Scene Discovery by Means of K-means Clustering Process and BPNN with Multispectral Satellite Images

Abstract

By substantial growths of remote recognizing have permitted the accession of evidence used on land scene discovery at divergent spatial scales. Substantial collections of remote sensing imagery have furnished a concrete foundation for multispectral analysis of the environment and the influence of human activities. The goal of the remote sensing investigation is to measure the areas of the categories in activity information or offer data that can be utilized to guide sampling strategies. Changes on the scene can be identified as changes in the ‘spectral space’ involved by an image pixel. In this paper, the proposed technique concentrated on the Object-based scene discovery method, which integrates a system for the arrangement of multispectral satellite images into various pre-established scene cover periods. This work incorporates Non-Modified Histogram Equalization (NMHE), K-Means Clustering with Backpropagation neural network (BPNN) to identify the changed areas using remote sensing images. Supervised classification is the technique used for acquiring the results which aims to progress the performance of Back Propagation Neural Network concerning change detection. Experimental results on various multispectral satellite images displays the correctness of the technique and images obtained from different time periods have shown that this approach is comparatively outperforming the conventional change detection method. As a final point, a vigorous and high-comparison scene discovery result can be achieved.