The Proposal aims to develop deep learning algorithms to perform semantic segmentation of high-resolution multispectral (HRMS) in an unsupervised manner. The segmentation results will be invariant to the domain-shift problem for the scenes belonging to different imaging conditions and satellite sensors.