An Approach Automatic Change Detection Method for Satellite Images using Deep CNN

Abstract

In geoscience, Change Detection (CD) is a useful method for analysing land surface changes using data from Earth observation and for uncovering links between human activities and environmental phenomena. In this work, a supervised Deep Learning (DL)-based change detection technique was developed to generate an accurate change map, as improving the quality of the binary CD map is a crucial issue in remote sensing images. Due to its strong performance and promise in the realms of pattern recognition and nonlinear problem modelling, DL is gaining traction as a means of overcoming the CD issue using multi temporal remote sensing imageries. Using DL methods, specifically Convolutional Neural Networks (CNN), may help divide environmental changes into two classes: with and without human intervention. To identify the same shift between two SAR pictures, we suggested a Deep CNN in this study. To classify the changes and properties of RSI images, a CNN is employed for supervised classification. A differential picture is generated using the information supplied by the CNN's convolution layers without prior training on objective difference images. Research has made use of the NASA satellite picture collection. Therefore, the change detection approach may be implemented in a supervised manner, making it suitable for usage with any classification algorithm or CNN that has already been pre-trained.