Research on the optimization of multi-class land cover classification using deep learning with multispectral imagery

Yichuan Li, Junchuan Yu*, Ming Wang, Minying Xie, Laidian Xi, Yunxuan Pang, and Changhong Hou

*corresponding author

Updates

• [2024.3.24] Paper submission.
• [2024.4.23] Paper revision
• [2024.4.24] Code release

Dataset

• The Worldview3 data used in this study was released during the Dstl Satellite Imagery Feature Detection competition, provided by the Defense Science and Technology Laboratory of the United Kingdom. the original data can be download through this link: kaggle dtsl dataset
• The training data used for the optimal model can be downloaded via this link: training dataset
• The panoramic data used for testing can be downloaded via this link: testing dataset

Code

Structure of the proposed Model

Click the links below to download the checkpoint for the corresponding model type.

• [Baseline model]: Google drive

  • Save the file in your download directory: /checkpoint/baseline.hdf5

• [OUNet model]: Google drive

  • Save the file in your download directory: /checkpoint/OUNet.hdf5

• The supporting library information of the code is shown below:

Package	Version
GDAL	3.4.3
h5py	3.1.0
matplotlib	3.7.2
numpy	1.22.1
tensorflow	2.10.0
pandas	1.5.0
sklearn	1.1.2

Acknowledgement

• HeyWhale Community Provided the computational platform for this work.

• kaggle Provided the dataset for this work.

If you find this study helpful, please star this repo OUNet_for_multi-class_land_cover_classification, and cite using this BibTeX:

@article{
  title={TResearch on the optimization of multi-class land cover classification using deep learning with multispectral imagery},
  author={Lichuan Li, Junchuan Yu*, Ming Wang, Ming Ying, Laidian Xi, Yunxuan Pang, and Changhong Hou}
  year={2024}
}