Evaluation of Machine Learning and Deeplearning Algorithms Applied to Earth Observation Data for Change Detection in Polarimetric Radar Images

The aim here is to detect changes in polarimetric radar images (VV and VH) from the Sentinel 1A and 1B satellites. These changes are of a general nature, and can be linked to natural disasters. They include earthquakes, flooding, sea water pollution, deforestation, crop evolution, land surveying and climate change. In our work, we are interested in changes in the city of Douala. We used a stack of two images, one taken before the change and the other taken after the change. Our contribution is to set up a method based on the fusion of machine learning, deeplearning and algebraic methods in order to obtain more efficient results. This model involves firstly applying machine learning methods such as random forest and algebraic methods based on Minkowski and Kolmogorov algorithms in parallel to our pre-processed images. The second step is to generate a deep learning model based on convolutional neural networks, which will take the images from the algebraic and machine learning methods as input in order to generate two output classes (changed and not-changed).