SGPL

Semantic Guided Prototype Learning for Cross-Domain Few-Shot Hyperspectral Image Classification

Requirements

To run this project, you will need to add the following environment variables to your .env file

python = 3.8.8

torch == 1.12.1+cu113

torchvision == 0.13.1+cu113

Datasets

• source domain dataset

  • Chikusei
  • Botswana
  • Houston
  • KSC

• target domain datasets

  • Salinas
  • PaviaU
  • Indian Pines

An example datasets folder has the following structure:

data
├── SA
│   ├── salinas_corrected.mat
│   └── salinas_gt.mat
├── UP
│   ├── PaviaU.mat
│   └── PaviaU_gt.mat
├── IP
│   ├── indian_pines_corrected.mat
│   ├── indian_pines_gt.mat
└── source_data
    └── the upcoming source domain dataset

The four datasets of the source domain should be stored at locations of your choosing based on the addresses specified in the particular code.

Usage

1. Download the required source and target domain datasets and store them in their specific locations.

2. Preparation of pre_training dataset.

• generate_source_CH_data_process.py, generate_source_HS_data_process.py,

• generate_source_HS_data_process.py, generate_source_KSC_data_process.py,

• generate_source_process.py. Then you will obtain the datasets required for pre-training.

3. Pre_training.

• train_SGPL_source.py. Then you will obtain the weight of pre-training model.

4. Fine-tuning.

• train_SGPL_SA.py,train_SGPL_UP.py,train_SGPL_IP.py. Then you will obtain the weight of fine-tuning model.

5. Testing.

• test_SGPL_SA.py,test_SGPL_UP.py,test_SGPL_IP.py. Then you will obtain the result of different datasets.

Supplement

To facilitate a faster code execution, we have provided the weights of the pre-trained model in the file. You can directly proceed to the fine-tuning phase, significantly reducing your time.