Hyperspectral Image Classification using Transfer Learning

Abstract

Deep learning is rapidly being used in the field of hyperspectral image (HSI) classification due to recent improvements in computing power and evolution in deep learning technology. Usually, deep learning models contain myriad parameters, and to achieve an optimal performance they require many labeled samples. HSI datasets in general have very few samples available as the collection of training samples is expensive and time-consuming. Therefore, many researchers focus on building deep learning models that provide near-optimal solutions with very few labeled samples. In this paper, we focus on this topic and provide an organized review of pertinent literature. Furthermore, we have performed several experiments with various transfer learning models on different HSI datasets and the results are condensed to show that even though HSI datasets contain few samples, the transfer learning models perform better on these datasets compared to many other state-of-the-art models.