Hyperspectral Image Classification Using Deep Matrix Capsules

Deep Matrix Capsules is based on the concept of matrix capsules with Expectation-Maximization (EM) routing algorithm, specifically designed to accommodate the nuances in the HSI data to efficiently exploit spectral-spatial relationships with reduced computational complexity.

Deep Matrix Capsules Architecture for HSI Classification

• PyTorch v1.6.0
• Python v3.9
• Scikit Learn v1.1
• Numpy v1.20

Fig: The Indian Pines dataset classification result (Overall Accuracy 99.93%) of Deep Matrix Capsules using 50% samples for training. (a) RGB Composition. (b) Ground-truth classification Map. (c) Classification map corresponding to Deep Matrix Capsules. (d) Class legend.

Fig: The Salinas Scene dataset classification result (Overall Accuracy 100.00%) of Deep Matrix Capsules using 50% samples for training. (a) RGB Composition. (b) Ground-truth classification Map. (c) Classification map corresponding to Deep Matrix Capsules. (d) Class legend.

Fig: The University of Pavia dataset classification result (Overall Accuracy 99.99%) of Deep Matrix Capsules using 50% samples for training. (a) RGB Composition. (b) Ground-truth classification Map. (c) Classification map corresponding to Deep Matrix Capsules. (d) Class legend.

@INPROCEEDINGS{10028853,
  author={Ravikumar, Anirudh and Rohit, P N and Nair, Mydhili K and Bhatia, Vimal},
  booktitle={2022 International Conference on Data Science, Agents & Artificial Intelligence (ICDSAAI)}, 
  title={Hyperspectral Image Classification Using Deep Matrix Capsules}, 
  year={2022},
  volume={01},
  number={},
  pages={1-7},
  doi={10.1109/ICDSAAI55433.2022.10028853}}

The following repositories were used for this work

• Matrix Capsules using PyTorch
• HybridSN
• HSI - Traditional to Deep Models
• SpectralNET
• HSI Classification
• Double-Branch Dual-Attention Mechanism Network

Copyright (c) 2023 Rohit P N and Anirudh Ravikumar. Released under the MIT License. See LICENSE for details.