Qiu et al. (2026) Visual recognition algorithm for weakly labeled multi-source data fusion of remote sensing images

Identification

• Journal: Scientific Reports
• Year: 2026
• Date: 2026-05-11
• Authors: Zemin Qiu, Jiajun Zou, Shaojiang Liu, Bensheng Yang
• DOI: 10.1038/s41598-026-52254-8

Research Groups

• School of Information and Intelligence Engineering, Guangzhou Xinhua University, China
• School of Artificial Intelligence, Guangzhou Huashang College, China

Short Summary

The study proposes a multi-source data fusion algorithm for remote sensing image recognition that leverages weakly labeled data and multimodal features to improve fine-grained classification and cross-domain generalization.

Objective

• To overcome the heavy reliance on strong annotations and the poor cross-domain generalization typical of existing fine-grained visual recognition methods in remote sensing.

Study Configuration

• Spatial Scale: Image-level analysis of remote sensing data.
• Temporal Scale: Not specified.

Methodology and Data

• Models used:
  • Self-supervised learning framework (utilizing rotation prediction and image restoration tasks).
  • Multimodal cross-attention fusion model (integrating cross-attention and gating mechanisms).
  • Cross-domain contrastive loss (combining contrastive learning with domain adaptation).
• Data sources: Multi-source remote sensing images, textual semantic features, and traditional benchmark datasets.

Main Results

• The proposed algorithm achieved a classification accuracy of 94% on traditional datasets.
• The model outperformed existing baseline models and demonstrated stable performance in real-world application scenarios.

Contributions

• Developed a framework to efficiently utilize weakly annotated data through self-supervised learning and attention-based filtering.
• Bridged the gap between local image features and textual semantic features using a shared embedding space and cross-attention.
• Enhanced cross-domain generalization by identifying domain-invariant features via a novel cross-domain contrastive loss.

Funding

• 2024 University-level Research Project of Guangzhou Xinhua University (No. 2024KYZDZK02)
• Guangdong Province Key Construction Discipline Research Capacity Enhancement Project (Nos. 2021ZDJS144, 2024ZDJS130)
• Characteristic Innovation Category Project of Guangdong Ordinary Colleges and Universities (Nos. 2024KTSCX127, 2025KTSCX220)
• School-level Scientific Research Project of Guangzhou Xinhua University (No. 2024KYCXTD02)
• Young Innovative Talents Category Project of Guangdong Ordinary Colleges and Universities (Nos. 2023KQNCX124, 2024KQNCX076)
• Key Research Platforms and Projects of Regular Higher Education Institutions in Guangdong Province (No. 2025ZDZX3059)

Citation

@article{Qiu2026Visual,
  author = {Qiu, Zemin and Zou, Jiajun and Liu, Shaojiang and Yang, Bensheng},
  title = {Visual recognition algorithm for weakly labeled multi-source data fusion of remote sensing images},
  journal = {Scientific Reports},
  year = {2026},
  doi = {10.1038/s41598-026-52254-8},
  url = {https://doi.org/10.1038/s41598-026-52254-8}
}