Hyperspectral image classification using semi-supervised learning with label propagation

Hyperspectral Image generally contains hundreds of spectral bands and thus provides a huge amount of information for a particular scene. Despite this, the classification task for hyperspectral image is considered difficult due to less number of labeled samples available. In recent years, deep learning algorithms have grown as the most significant and highly effective for classification tasks. But these algorithms require a huge amount of labeled data which is not suitable for hyperspectral images as getting labeled data is costly. To mitigate this problem, we can employ semi-supervised learning techniques that can address the issue of less labeled samples for training. In this paper, we have used label propagation technique to improve the performance of the CNN model using semi-supervised learning. By considering this semi-supervised learning strategy, we can obtain comparative performance on hyperspectral data using very less number of labeled samples.