A Review of Deep-Learning Methods for Change Detection in Multispectral Remote Sensing Images

: Remote sensing is a tool of interest for a large variety of applications. It is becoming
increasingly more useful with the growing amount of available remote sensing data. However, the
large amount of data also leads to a need for improved automated analysis. Deep learning is a
natural candidate for solving this need. Change detection in remote sensing is a rapidly evolving
area of interest that is relevant for a number of fields. Recent years have seen a large number of
publications and progress, even though the challenge is far from solved. This review focuses on deep
learning applied to the task of change detection in multispectral remote-sensing images. It provides
an overview of open datasets designed for change detection as well as a discussion of selected models
developed for this task—including supervised, semi-supervised and unsupervised. Furthermore, the
challenges and trends in the field are reviewed, and possible future developments are considered.