Geo-Spatial Object Data Analytics Based Change Detection in Aerial Imagery

Change detection in remote sensing is indispensable for monitoring areas of interest. While radiometric changes depict seasonal and man-made changes appearing in each pixel, the comprehension of such changes are subjective and often limited to climate, geological or ecological studies. In a first, this work utilizes an object detection algorithm to create an object data layer per image, and compares this geospatial object data layer from imagery of two different times using data analytics. This results in a focused approach, where the seasonal variations, vegetation etc., do not distract the change detection process. The object classes, such as buildings, vehicle lots, etc, also provide a semantic understanding of the changes. The algorithm is tested on available aerial imagery datasets from two different dates, having different resolutions. A popular detection algorithm trained on the xView 60 class dataset is employed in this study.