Cross-Domain Meta-Learning under Dual Adjustment Mode for Few-Shot Hyperspectral Image Classification, TGRS, 2023.

This is a code demo for the paper: Cross-Domain Meta-Learning under Dual Adjustment Mode for Few-Shot Hyperspectral Image Classification.

Fig. 1: Overall framework of the DMCM. The dual-adjustment mode-based cross-domain meta-learning framework comprises the following four parts: (a) feature extraction, (b) cross-domain FSL, (c) dual adjustment mode, and (d) target dataset classification.

References

If you find this code helpful, please kindly cite:

@article{hu2023cross,
  title={Cross-Domain Meta-Learning under Dual Adjustment Mode for Few-Shot Hyperspectral Image Classification},
  author={Hu, Lei and He, Wei and Zhang, Liangpei and Zhang, Hongyan},
  journal={IEEE Transactions on Geoscience and Remote Sensing},
  year={2023},
  publisher={IEEE}
}

Some of our reference projects are listed below, and we are very grateful for their research:

• Graph Information Aggregation Cross-Domain Few-Shot Learning for Hyperspectral Image Classification, TNNLS, 2022.
• Deep Cross-domain Few-shot Learning for Hyperspectral Image Classification, TGRS, 2022.
• Few-shot Learning with Class-Covariance Metric for Hyperspectral Image Classification, TIP, 2022.

Requirements

CUDA Version = 11.7

Python = 3.9.7

Pytorch = 1.12.0

Sklearn = 0.24.2

Numpy = 1.20.0

Matplotlib = 3.4.3

Spectral = 0.22.4

Dataset

1. Target domain data set: Indian Pines (IP)/ Pavia University (UP)/Pavia Center (PC)

You can download the hyperspectral datasets in mat format at: http://www.ehu.eus/ccwintco/index.php/Hyperspectral_Remote_Sensing_Scenes, and move the files to ./Datasets folder.

2. Source domain data set: Chikusei

The source domain hyperspectral datasets (Chikusei) in mat format is available in: https://github.com/Li-ZK/DCFSL-2021.

You can also download our preprocessed source domain data set (Chikusei_imdb_128.pickle) directly in pickle format, please move to the ./Datasets folder. However, the data is too large to upload, if necessary, please contact my email: [email protected].

An example dataset folder has the following structure:

Datasets
├── Chikusei_imdb_128.pickle
├── IP
│   ├── indian_pines.mat
│   ├── indian_pines_gt.mat
├── paviaU
│   ├── paviaU.mat
│   └── paviaU_gt.mat
└── paviaU
    ├── PaviaCenter.mat
    └── PaviaCenter_gt.mat

Usage:

An example of DMCM method:

1. Download the required data set and move to folder./Datasets.
2. If you down the source domain data set (Chikusei) in mat format, you need to run the script ./Datasets/Generate_source_dataset.py to generate preprocessed source domain data.
3. Taking 5 labeled samples per class as an example, run DMCM_IP/UP/PC.py.

• --test_lsample_num_per_class denotes the number of labeled samples per class for the target domain data set.
• --tar_input_dim denotes the number of bands for the target domain data set.