fengjiaqi927 / hsi_transformer

transformer-based method for HSI classification

Spectral Transformer with Dynamic Spatial Sampling and Gaussian Positional Embedding for Hyperspectral Image Classification

Framework

Enveronment

conda create -n HSI python=3.7.11

conda activate HSI

pip install -r requirements.txt

Attention

HSI cube is generated from superpixel region. It is time-consuming when generating in each batch. To speed up the dataload process, we generate the HSI cube before training network and all cube is storaged in SLIC_samples.npy.

Usage

1. generation of superpixel region

   we use matlab to generate superpixel region. The code is in matlab_code dir.

2. normal cube generation version

   train: python train_IP_normal.py

   test: python test_IP_normal.py

3. cube generated from superpixel region

   2.1 generate while training

   train: python train_IP_speed_normal.py

   test: python test_IP_speed_normal.py

   2.2 generate before training

   generate python prepare_data.py

   train: python train_IP_speed_up.py

   test: python test_IP_speed_up.py

TODO

1. Add article link (done)
2. Upload code (done)