Water Body Change Detection and Estimation From Landsat Satellite Images Using Deep Learning

Abstract Identification of urban water bodies from satellite images has been extensively explored in the recent past. Different types of mechanisms have been developed to delineate water bodies from distinct satellite images differing in spatial and temporal aspects. The problem of identifying urban water body with abundant sureness materializes in several applications. Amongst them, natural resource mapping like forest and urban water bodies utilizing satellite image has received much significance over the past few years. This is owing to the reason than both forest and water resources are dependent on each other to enormous exploitation and hence monitoring them at regular time intervals is highly essential for their sustainable management. At the same time, different methods have been adopted to select the pertinent features from satellite images using machine learning and each method has its own advantages and disadvantages. In this work, a method called, Nonlinear Recursive Filter used to preprocessing the satellite image and Convolution neural network (Deep learning method) used for urban water body identification from satellite images is proposed. The proposed CNN method is split into three parts. First, with the satellite images of water bodies taken as input, significant amount of noise are removed and also preserving the edges for future reference with the aid of Nonlinear Recursive Filter-based Preprocessing model. Second with the noise minimized preprocessed water body images, relevant features are selected in Convolution neural network Deep learning model. Finally, appropriate classification between the water bodies and non-water bodies are made by utilizing CNN Model. Experiments are conducted using water body images and simulations are performed to analyze PSNR, water body identification justify that accuracy, Spatial and Temporal analysis.