Satellite-based Infrastructure Change Detection using Deep Learning

Infrastructure development and land-use changes are critical indicators of urban growth and environmental impact. Monitoring these changes using satellite imagery is essential for urban planning, disaster management, and environmental conservation. However, manual analysis of satellite data will be complex and often lacks precision. This Research addresses this challenge by implementing an automated change detection system on land area using deep learning models.This approach used the Siamese U-Net and U-Net architectures to analyze pre-change and post-change satellite images, identifying and masking regions where infrastructure or land-use changes have occurred. These model trained with LEVIR-CD dataset, which contains high- resolution satellite image pairs and corresponding change masks.The Siamese U-Net achieved an accuracy of 96.81% while the U-Net achieved a accuracy of 95.54% in detecting changes, with the results visualized as binary masks and overlaid on the input images for better interpretation.