Hyperspectral-Image-Classification-Based-on-3D-Octave-Convolution-with-Spatial-Spectral-Attention

The official codes for paper "Hyperspectral Image Classification Based on 3D Octave Convolution With Spatial-spectral Attention Network"

Install dependencies

numpy
python==3.6
sklearn
tensorflow==1.5
pycharm

dataset

We conduct the experiments on the University of Pavia data set. To train and test our model, you should 
download the data set and modify image's path according to your needs.

data process

Since the input of our network is patch, it is necessary to preprocess the original hyperspectral image. 
The image preprocessing includes data normalization and patch cutting. 
In our experiments, the patch size is set to 13*13, and the training set and testing set are divided.
If you want to divide the data set, run the program `data_prosess.py'      

train

All the configurations are diaplayed in `model.py', and you can modify them by your needs. Please download the
`model.py' for training and testing data set first.    

train the model

Please run the program 'train.py' and save the parameters. In the `train.py', the path to read the data
should be changed according to your own situation.

test

Please read the saved parameters and run the program 'test.py'. In the `test.py', the path to read the data
should be changed according to your own situation.