Joint Imputation of Multiple Highly Sparse Remote Sensing Observations with Heterogeneous Mixture of Experts

Qian Chen¹, Shuyu Wang², Wengen Li¹, Jihong Guan¹, Zhen Zhang^1
1Tongji University, Shanghai, China   ²Xi'an Shiyou University, Xi'an, China

       

This is the official repository for MoE4MRS, a joint imputation framework for multi-source remote sensing data that integrates a Test-Time Training–based intra-variable learning module with a Mixture-of-Experts–based inter-variable fusion module to capture dynamic temporal dependencies and cross-source correlations, enabling robust reconstruction of highly sparse observations.

🎉Usage

🔧Setup

The required dependencies for this project are listed in requirements.txt. You can install them using pip with the following command:

pip install -r requirements.txt

(Optional) If you still encounter the package missing problem, you can refer to the requirements.txt file to download the packages you need.

If you encounter other environment setting problems not mentioned in this README file, please contact us to report your problems or open an issue.

🔥Train

You can use the following instruction in the root path of this repository to run the training process:

bash scripts/MoE4MRS.sh

This script will run the MoE4MRS model on both YRE and EAST datasets with various mask rates (0.2, 0.5, 0.7, 0.9). The script includes all necessary hyperparameter configurations and will automatically handle the training process for both datasets.

If you want to customize the training, you can modify the parameters in scripts/MoE4MRS.sh to adjust batch size, learning rate, mask rates, or other hyperparameters.

📜License

This project is licensed under the MIT License - see the LICENSE file for details.

📧Contact

If you encounter any problems, feel free to contact the author via email at [email protected] (or [email protected] as a personal email).