Weiyue et al. (2026) WAVES: Fine-grained Water Type Mapping via Vision-Language Alignment under Supervision without Pixel-level Labels

Identification

• Journal: IEEE Transactions on Geoscience and Remote Sensing
• Year: 2026
• Date: 2026-01-01
• Authors: Shi Weiyue, Sui Haigang
• DOI: 10.1109/tgrs.2026.3674221

Research Groups

This paper presents WAVES, a novel method for fine-grained water type mapping that utilizes vision-language alignment, enabling effective supervision without the need for pixel-level labels.

Objective

• To develop and evaluate WAVES, a method for fine-grained water type mapping, by leveraging vision-language alignment to overcome the dependency on pixel-level annotated data for supervision.

Study Configuration

• Spatial Scale: Fine-grained mapping.
• Temporal Scale: Not specified.

Methodology and Data

• Models used: A vision-language alignment model (WAVES).
• Data sources: Implied to be image data (for vision) and textual descriptions (for language).

Main Results

• The proposed WAVES method demonstrates the feasibility of achieving fine-grained water type mapping through vision-language alignment, effectively operating under supervision without requiring pixel-level labels.

Contributions

• Proposes WAVES, a novel approach for fine-grained water type mapping.
• Introduces a method for supervision that does not require pixel-level labels, significantly reducing annotation burden.
• Leverages vision-language alignment for enhanced water type classification.

Funding

Citation

@article{Weiyue2026WAVES,
  author = {Weiyue, Shi and Haigang, Sui},
  title = {WAVES: Fine-grained Water Type Mapping via Vision-Language Alignment under Supervision without Pixel-level Labels},
  journal = {IEEE Transactions on Geoscience and Remote Sensing},
  year = {2026},
  doi = {10.1109/tgrs.2026.3674221},
  url = {https://doi.org/10.1109/tgrs.2026.3674221}
}