Method and system for domain aware semi-supervised learning

Abstract

Classification of images is inherently a semi-supervised classification problem. Often, the labeled pixels and the unlabeled pixels in the image may have different distribution. Hence classification accuracy of such images is affected. The present disclosure proposes an umbrella framework for semi-supervised learning that considers the domains shifts in labeled and unlabeled pixels. The method proposed a two way optimization solution using deep learning models based on spectral features, spatial features, and fused spectral-spatial features. The model is trained in such a way that it is not only trained on the correct class of pixel but also on the source category of the pixel, for example, labeled pixel or unlabeled pixel. The error in the pixel class is minimized, whereas the error in the source category is encouraged simultaneously.