Hyperspectral Image Classification via Modified Stable Prototypical Networks

This research work proposes a new approach for Hyperspectral Imaging (HSI) classification. Even with the success of new deep learning frameworks, HSI classification models tend to have challenges because of a lack of labeled samples. Hyperspectral data is not easily accessible. To overcome these objectives, this research work introduces a modified Stable Prototypical Networks (SPN) to improve the overall outcome of the respective model. The proposed model constructs multiple class prototypes in a low dimensional space and uses few shot learning by calculating the distance between the prototypes and samples. The proposed architecture includes a modified mathematical model instead of traditional approach to improve the model performance. By incorporating the new mathematical approach within existing SPN model provides various advantages like effective transferring of cross-domain knowledge, improved generalization and low annotation issues. The proposed model when experimented with specific evaluation strategies shows great performance demonstrating adaptability and resource management in HSI classification.