A new bit-level hyperspectral tensor data compression method that combines a data-driven quantized neural encoder and channel-wise attention-based enhancement super-resolution.

• Ununtu 18.0
• python 3.7
• Pytorch 1.4

Creat HSI folder and put HSI dataset in, and add corresponding path in the .txt file in the testpath and trainpath
Run the train.py for training and testing.py for testing

In the file Classification

Datasets contains the cropped classification datasets Indian Pines (128×128×172) and Salinas (512×128×172), and their corresponding reconstructed data

checkpointIP and checkpointSalinas contain 10 weight files of the model trained on the dataset IP and Salinas, respectively.

logIP and logSalinas contain 5 txt files respectively, recording the results of 10 classification experiments

IP_ori.txt and S_ori.txt record the accuracies of 10 model weights on the classification dataset and corresponding values of random seeds (select the training samples randomly)

Make sure you have set the training or testing mode, then

python Demo_IP.py

or

python Demo_S.py

to implement the training or testing on the classification datasets