Land Cover / Land Use map using machine learning

Object - based image classification of Sentinel 2 imagery using machine learning:

Land classification for generating land use / land cover maps, are in great use for the planning and management purpose of a region/nation. It has been concluded that, pixel-based image classification of satellite images is the most standard approach followed by these communities. However, as said by Blaschke, the LULC map generated using this approach has a grainy effect. On the other hand, high performing tools like eCognition where object-based classification can be undergone easily, is a proprietary platform creating barrier of access.

Therefore, my project aims to test the applicability of a standalone open source software, "Orfeo Toolbox", for the object-based image classification of high resolution sentinel 2 imagery. The project involves two machine learning algorithms, libSVM and Random Forests. Further, the change detection map for the region of interest using the same open source toolbox will be produced. The results of this project confirms the standard, easy and realiability of the LULC maps generated using the object-based approach, which does not only have overall higher accuracy but also are visually appealing and easy to interpret land cover changes.

(Presented at Canadian Symposium of Remote Sensing 2023)