A rapid increase in the quantity as well as the quality of remote sensing data asks for new methods to come to an automated approach for change detection in satellite imagery. This research uses an interdisciplinary approach on the question how machine learning with the input of times-series data can be used to come to an accurate, flexible, and automated change detection system for remote sensing.

A novel approach of using machine learning combined with graph based image segmentation on a running average in a time series of images is proposed. On validation using imagery of the newly constructed Marker Wadden island, the automated system proved effective in detecting significant changes.

With the system a satellite image analyst can monitor large areas and focus on assessing areas flagged by the system. It improves change detection analysis in terms of speed and quality and severely reduces the human workload.