Suresh et al. (2025) U-Net++ and ConvLSTM based Semantic Event Stream Processing for Real-Time Flood Monitoring

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: Springer Link (Chiba Institute of Technology)
• Year: 2025
• Date: 2025-10-08
• Authors: A. Suresh, Smitha Kurian, Subhra Chakraborty, P. Rajeswari, Leeladhar Gudala
• DOI: 10.1051/itmconf/20257901041/pdf

Research Groups

Not explicitly mentioned in the provided text.

Short Summary

This paper introduces UConvFloodNet, a deep learning framework that integrates image enhancement, U-Net++, and ConvLSTM to achieve highly accurate real-time flood monitoring by segmenting flood zones and tracking their temporal evolution.

Objective

• To develop a fast and accurate method for real-time flood monitoring by enhancing satellite multispectral images, precisely segmenting flood zones, and dynamically tracking flood evolution over time.

Study Configuration

• Spatial Scale: Local to regional (focused on detecting submerged regions and flood zones).
• Temporal Scale: Real-time monitoring, tracking changes frame by frame, capturing flood evolution over time.

Methodology and Data

• Models used: UConvFloodNet, which incorporates Wiener filtering for noise removal, pixel intensity conversion for spatial contrast enhancement, U-Net++ for segmentation (flood zones, water bodies, land covers), and ConvLSTM for temporal tracking.
• Data sources: Satellite Multispectral Images; evaluated using the Sen1Floods11 and S1GFloods datasets.

Main Results

• The UConvFloodNet achieved an accuracy of 99.31% on the Sen1Floods11 dataset.
• It also achieved an accuracy of 99.06% on the S1GFloods dataset.
• The proposed method demonstrated superior performance compared to existing approaches, such as the Compact Convolutional Tokenizer integrating U-Net and Vision Transformer (CCT-U-ViT).

Contributions

• Introduction of UConvFloodNet, a novel, integrated framework for real-time flood monitoring that combines image enhancement, precise spatial segmentation, and dynamic temporal tracking.
• Significant improvement in flood detection accuracy (99.31% and 99.06%) over existing state-of-the-art methods.
• Addresses the limitations of satellite multispectral images (limited spectral bands, low spatial resolution) through advanced image processing techniques.

Funding

Not explicitly mentioned in the provided text.

Citation

@article{Suresh2025UNet,
  author = {Suresh, A. and Kurian, Smitha and Chakraborty, Subhra and Rajeswari, P. and Gudala, Leeladhar},
  title = {U-Net++ and ConvLSTM based Semantic Event Stream Processing for Real-Time Flood Monitoring},
  journal = {Springer Link (Chiba Institute of Technology)},
  year = {2025},
  doi = {10.1051/itmconf/20257901041/pdf},
  url = {https://doi.org/10.1051/itmconf/20257901041/pdf}
}