Costas Panagiotakis Homepage

Flood Detection

Figure 1:Graphical abstract of our proposed unsupervised approach [1].

Figure 2:Graphical abstract of our proposed unsupervised Deep Learning approach [2].

Figure 3:Graphical abstract of our proposed unsupervised Deep Learning approach [3].

Presentation of paper [2] G. Simantiris and C. Panagiotakis, Unsupervised Deep Learning for Flood Segmentation in UAV imagery, 13th IAPR Workshop on Pattern Recognition in Remote Sensing (PRRS), presented by G. Simantiris.

In [1] we propose a novel unsupervised semantic segmentation method for fast and accurate flood area detection utilizing color images acquired from Unmanned Aerial Vehicles (UAVs). To our knowledge so far, this is the first fully unsupervised method for flood area segmentation in color images captured by UAVs, without the need of pre-desaster images.

In [2], we present a novel unsupervised Deep Learning method for flood segmentation in Unmanned Aerial Vehicle imagery. This method utilizes automatically generated labels as masks for the training process, eliminating the need for actual ground truth data.

In [3], we present a novel methodology for generating and filtering synthetic Unmanned Aerial Vehicle (UAV) flood imagery to enhance the generalization capabilities of segmentation models.

Datasets - Code - Experiments

code [1]: https://www.mathworks.com/matlabcentral/fileexchange/167961-flood-segmentation
Flood Area Segmentation Dataset (290 images) [1-2]: https://www.kaggle.com/datasets/faizalkarim/flood-area-segmentation
Flood Semantic Segmentation Dataset (663 images) [1-2]: https://www.kaggle.com/datasets/lihuayang111265/flood-semantic-segmentation-dataset
Real, Synthetic (SD_s) and semi-synthetic (SD_ip) Datasets [3]: https://drive.google.com/file/d/1XLU5tpOONZ1zjdEgWUdpuVVFN2DRZz7c/view?usp=sharing

Figure 3. Original images (a) to (d), respective ground truth (e) to (h), and segmentation results from UFS-HT-REM (i) to (l), U-Net (m) to (p), and FCN_ResNet-50 (q) to (t). The flood is overlaid in blue color and each segmentation’s F1-score is also reported. We used the descending sorted differences between U-Net and UFS-HTREM segmentations to showcase the top 0th, 25th, 50th and 75th percentile results for comparison [2].

Related Publications

[1] Georgios Simantiris and Costas Panagiotakis, Unsupervised Color Based Flood Segmentation in UAV imagery, vol. 16, no 12, Remote Sensing, 2024.

[2] G. Simantiris and C. Panagiotakis, Unsupervised Deep Learning for Flood Segmentation in UAV imagery, 13th IAPR Workshop on Pattern Recognition in Remote Sensing, 2024.

[3] G. Simantiris, K. Bacharidis and C. Panagiotakis, Closing the Domain Gap: Can Pseudo-Labels from Synthetic UAV Data Enable Real-World Flood Segmentation?, 25(12), Sensors, 2025.