A Comparative Study of Unsupervised Machine Learning Algorithms for Recognition of Change in Vegetation, Water and Built up Land Areas

Change detection is a trendy and versatile research topic in the area of application of remote sensing that includes disaster assessment, forest monitoring, urban sprawl and many more. In a growing era, it has become necessary to introduce more and more machines to the industry for better efficiency and accuracy. Each and every task of image processing is tedious and needs a lot of concentration for better accuracy. In remote sensing while finding the Land cover change detection, generally manual process is applied which consumes too much time and unworthy effort. Therefore, here arises the need of automatic change detection which originates different algorithms for image analysis. In this study we have tried to analyze the effectiveness of two type of unsupervised machine learning algorithm and to detect the change in major classified classes in satellite imagery. The image has been classified in three major classes of vegetation, Built up and water bodies. A comparative study of unsupervised machine learning algorithms has been carried out. Landsat satellite imageries of Allahabad have been used to fulfill the required objectives. Indices majorly NDWI, MSAVI2 and NDBI have been calculated for dominant classes. Two different algorithm K-means and FCM based on machine learning concept of partition and fuzzy have been used. Different types of Land cover (vegetation, Built up and water) have been identified while implementing in MATLAB. The percentage change has been observed and compared with finding decreased percentage trend of vegetation and water while increased percentage of Built up class.