Object Based Change Detection using Machine Learning and Multi-source Data
Earths surface is subject to constant change due to natural phenomena or human intervention. The process of identifying those changes is called “change detection” (CD) and is a core problem in environmental monitoring, disaster monitoring, city expansion and land cover. CD approaches, when working with raster data, can be divided into pixel- and object-based change detection. In object-based change detection tasks, as the name suggests, changes are not detected at the pixel level but at the object level.
In the field of geoinformatics, working with two-dimensional (2D) data is popular due to their underlying structure. However, raster data comes with significant drawbacks for object based change detection. For example, the object detection that precedes the change detection in 2D data is usually subject to uncertainties due to perspective and atmospheric effects. Another shortcoming of 2D change identification is the lack of possibilities to identify changes in height as 2D data does not provide elevation information directly. At the same time, studies on the use of deep learning for change detection tasks which also take the third spatial dimension into account, are not sufficiently explored.
This PhD project aims to fill the gap between tasks of object based detection with two- and three-dimensional data. Using deep learning algorithms, the respective advantages of both two- and three-dimensional data are to be combined while compensating for the disadvantages.
