I’ve done research about change detection techniques and as it seems, deep-learning-based methods have the best outcome among the others. Here is the research about change detection using one of the most effective methods of identifying meaningful changes in very high-resolution (VHR) images.
Abstract. Change detection is now one of the most important duties in remote sensing science. Over time it has been proven that deep-learning-based methods are one of the most effective ways to identify meaningful changes in a pair of co-registered images. In this project, we are using an end-to-end CD method based on an effective encoder-decoder architecture for binary change detection named UNet++, where change maps could be learned from scratch using a VHR dataset called LEVIR-CD+. UNet++ has been an improved version of UNet that uses a series of nested and dense skip connections, rather than only connections between encoder and decoder networks. This research has proven that the more we use nested outputs in the result, the better the results can get.