Landscape Change Detection Using Auto-optimized K-means Algorithm

This paper presents landscape change detection for sustainable environment monitoring and planning of urban areas. An integration of principle component analysis (PCA) and K-mean algorithm is used in the study. The PCA offers the advantages of data redundancy, but results are difficult to interpret. K-mean provides an efficient interpretation of PCA results, but the selection of cluster K affects the detection results. Therefore, an automatic selection of the optimized value of K is proposed in the paper. A Calinski-Harabasz criterion in contrast to the Silhouette criterion is proposed to find the optimized value of K. The results are obtained for both Panchromatic and multispectral satellite images of Pune and Ahmedabad city of India respectively. The experiment analysis shows that the Calinski-Harabasz criterion gives an accurate value of K with better change detection in the images.