Hyperdimensional CD using untrained models

Change detection in hyperspectral images and any other hyperdimensional images

The proposed method uses an untrained lightweight deep model, initialized with some weight initialization strategy for feature extraction from bi-temporal hyperdimensional images.

To run the code please download the datasets and put them in datasets folder.
Then cd to deepImagePriorNonlinear directory and run the code for Santa Barbara dataset by using following command:
python deepImagePriorSantaBarbara.py --manualSeed 40
The manual seed can be varied.

Citation

If you find this code useful, please consider citing:

@article{saha2021change,
  title={Change Detection in Hyperdimensional Images using Untrained Models},
  author={Saha, Sudipan and Kondmann, Lukas and Song, Qian and Zhu, Xiao Xiang},
  journal={IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing}
  year={2021},
  publisher={IEEE}
}