Change Detection of Urban GIS Maps with High Resolution Satellite Data Using Deep Learning Techniaues

Rapid increase in urban settlements has caused a tremendous effect on natural deposits and habitat. A perfect blueprint is needed for organizing the settlements without degrading the living standards. With the developed spatial technology, good quality of remote sensing data is now available. Using satellite images, change can be detected between two temporal images of the same scene. This change detection helps in the proper management of development proceedings. Exiting machine learning algorithms for change detection uses hand crafted features that may result to over-fitting problems. This paper proposes deep learning based methodology to detect the change in urban landscapes especially buildings. The proposed methodology involves a feature extractor complemented by an attention mechanism to capture the multi-scale dependencies between two pixels taken from two temporal images. With this methodology, both temporal and spatial features are taken into consideration in finding change detection.