HYRET-CHANGE: A HYBRID RETENTIVE NETWORK FOR REMOTE SENSING CHANGE DETECTION

Mustansar Fiaz, Mubashir Noman, Hiyam Debary, Kamran Ali, Hisham Cholakkal

---

🏆 Highlights

• HyRet-Change: We propose a Siamese-based framework, which can seamlessly integrate the merits of convolution and retention mechanisms at multi-scale features to preserve critical information and enhance adaptability in complex scenes change detection (CD). Specifically, we propose a hybrid plug-and-play feature difference module (FDM) to explore rich feature information utilizing both self-attention and convolution operations in a parallel way. This unique integration, at multi-scale features, leverages the advantages of both local features and long-range contextual information. We introduce a retention mechanism in our novel FDM to mitigate the limitations of standard self-attention.
• Local-Global (LG)-Interaction Module: We introduce an adaptive interaction between local and global representations to exploit the intricate relationship contextually to strengthen the model’s ability to perceive meaningful changes while reducing the effect of pseudo-changes.
• Experiments: Our extensive experimental study over three challenging CD datasets demonstrates the merits of our approach while achieving state-of-the-art performance.

Citation

@inproceedings{fiaz2025hyret,
  title={HyRet-Change: A hybrid retentive network for remote sensing change detection},
  author={Fiaz, Mustansar and Noman, Mubashir and Debary, Hiyam and Ali, Kamran and Cholakkal, Hisham},
  booktitle={IGARSS 2025-2025 IEEE International Geoscience and Remote Sensing Symposium},
  year={2025},
  publisher={IEEE}
}
@inproceedings{noman2024changebind,
  title={Changebind: A hybrid change encoder for remote sensing change detection},
  author={Noman, Mubahsir and Fiaz, Mustansar and Cholakkal, Hisham},
  booktitle={IGARSS 2024-2024 IEEE International Geoscience and Remote Sensing Symposium},
  pages={8417--8422},
  year={2024},
  organization={IEEE}
}
@article{noman2024elgc,
  title={ELGC-Net: Efficient local--global context aggregation for remote sensing change detection},
  author={Noman, Mubashir and Fiaz, Mustansar and Cholakkal, Hisham and Khan, Salman and Khan, Fahad Shahbaz},
  journal={IEEE Transactions on Geoscience and Remote Sensing},
  volume={62},
  pages={1--11},
  year={2024},
  publisher={IEEE}
}
@article{noman2024remote,
  title={Remote sensing change detection with transformers trained from scratch},
  author={Noman, Mubashir and Fiaz, Mustansar and Cholakkal, Hisham and Narayan, Sanath and Anwer, Rao Muhammad and Khan, Salman and Khan, Fahad Shahbaz},
  journal={IEEE Transactions on Geoscience and Remote Sensing},
  volume={62},
  pages={1--14},
  year={2024},
  publisher={IEEE}
}

Contact

If you have any question, please feel free to contact the authors. Mustansar Fiaz: mustansar.fiaz@ibm.com or Mubashir Noman: mubashir.noman@mbzuai.ac.ae.

References

Our code is based on Changebind repository. We thank them for releasing their baseline code.