Few-Shot Hyperspectral Image Classification Through Multitask Transfer Learning

Ying Qu, Razieh Kaviani Baghbaderani, Hairong Qi

Abstract: Despite a plethora of works on Hyperspectral image (HSI) classification, all state-of-the-art approaches need to train the classifier in a supervised fashion which requires a large amount of hand-crafted ground truth labels. In addition, the trained classifier may not work on another domain due to different acquisition conditions. Thus, how to preserve the classification accuracy in different domains remains a challenging issue. In this paper, we propose a few-shot HSI classification method to address the above challenges through Dirichlet-net based on multitask transfer learning. The essential contribution of this work is the realization of the transfer learning scheme that can extract shared intrinsic representations from the same type of objects in different domains; so that given a few samples, the classifier trained in the source domain can be directly applied to the target domain without having to collect more ground truth labels from the target domain. Experimental results demonstrate the superiority of the proposed method as compared to the state-of-the-art. The success of this endeavor would largely facilitate the deployment of HSI classification for real-world sensing scenarios.