Skip to main content

Change Detection

Visualize change

Change detection is a procedure to analyse changes in satellite images taken at two different points in time. The changes in these multi-temporal satellite images can be calculated in various ways. We will present the two most important computer-aided techniques, namely image algebra and post classification change detection.

 

Image Algebra Change Detection

This method needs a lot of calculation. The spectral characteristics of the satellite images are manipulated in order to get a new image showing change values only. The simplest example is the subtraction of two images. This makes sense if both images are similar regarding their spectral characteristics - if one image has different spectral values for a tree because of shadowing, the resulting change image will not depict the actual change. But if the spectral characteristics are similar, this method highlights changes without needing prior knowledge.

 

Image Algebra

Image Algebra Change Detection: The pixel values are substracted, the resulting images shows the differences between the images.

Post Classification Change Detection

This awkward expression stands for the analysis of changes using two previously classified images. Similar to the method mentioned above, the images are subtracted from each other and, thus, the changes become apparent. Other than in image algebra, it can be seen which class has changed and how much - and to which class the pixels are assigned afterwards. The new image can be interpreted easily and is ready for direct use, on condition of well-classified original images. The original information is not contained in the classified image and, therefore, cannot be considered during the change detection process. Thus, the post-classification change detection depends on the accuracy of the image classification.

 

Post Classification Change Detection


Post-Classification Change Detection: Example showing the class "settlement". Green areas represent increase, yellow stagnation and red areas decrease.



Conclusion:

Classification and Change detection are the most important analysis techniques of remote sensing. Change detection examines the changes in spectral characteristics of a region in the course of time in order to determine the processes leading to changes in land use or land cover. To analyse the changes, either original images (image algebra) or classified images (post-classification) are processed.