Change Detection in Satellite Images using Deep Learning

For the purpose of tracking environmental changes, urbanization, and natural disasters, change detection in satellite imagery is essential. New developments in deep learning, especially by creating “after” photographs from “before” photos, Generative Adversarial Networks (GANs) provide promising solutions for modeling possible changes over time. This paper investigates a GAN-based technique for identifying and labeling significant alterations in satellite imagery. A discriminator refines these by distinguishing between the actual and the artificial, while a generator produces believable modifications as part of the suggested architecture. A binary mask that graphically displays the identified changes is then produced as a result of thresholding the images. By demonstrating the model’s accuracy, precision, recall, and F1 measure against ground-truth data, GANs’ potential to consistently Automate the process of identifying changes in satellite photos.