Land Cover Clustering for Change Detection using Deep Belief Network

Change detection in recent era plays a major role in the remote sensing areas that includes land cover and water bodies. The application of change detection in remote sensing is considered complicated due to continuous interaction of humans with nature. Further, the change detection in land cover and water bodies plays a major role in global change. Manual change detection process via experts consumes more time to survey the land, whereas detection of changes in land cover and water bodies via an artificial intelligence (AI) model enables faster change detection. However, multi-sensor and multi-temporal satellite data calibration poses a serious problem in remote-sensing application. It makes the entire clustering process a difficult one and hence the accuracy of change detection degrades. In this paper, we apply the task of unsupervised learning to a deep learning model namely deep belief network. Here, the land cover images are clustered using the unlabeled data, where the algorithms are trained in unsupervised manner. The intelligent decisions are made available via deep belief network that clusters the uncategorized the land covers and water bodies images from the satellite image database. The dataset includes Landsat sensors’ datasets in dry semi-arid river basin in Western India. The performance of the unsupervised deep belief network is evaluated in terms of various performance metrics that includes accuracy, precision, recall, f-measure, geometric mean, peak signal noise ratio and mean average percentage error.