Incremental Learning Approaches for Flood Detection in Dynamic River Environments

• Otávio Ferracioli Coletti USP
• Saulo Matos Neves USP
• Caetano Mazzoni Ranieri UNESP
• Gowri Sankar Ramachandran Queensland University of Technology
• Raja Jurdak Queensland University of Technology
• Jó Ueyama USP

DOI: https://doi.org/10.5753/eniac.2025.12389

Abstract

Identifying urban floods promptly can mitigate risks to public safety. Traditional models for this task rely on static datasets and are ill-suited to handle continuous environmental variation. This work presents an image-based flood monitoring system that integrates an incremental learning pipeline capable of adapting in real-time to distributional shifts. The model is periodically updated through cloud-based retraining informed by expert feedback. We evaluate different replay buffer strategies and demonstrate that the buffer replacement policy has a greater influence on performance than the buffer size. Our results show that incremental learning with selective replay significantly improves model robustness in dynamic flood monitoring scenarios.

References

Bayram, F., Ahmed, B. S., and Kassler, A. (2022). From concept drift to model degradation: An overview on performance-aware drift detectors. Knowledge-Based Systems, 245:108632.

Dou, G., Chen, R., Han, C., Liu, Z., and Liu, J. (2022). Research on water-level recognition method based on image processing and convolutional neural networks. Water, 14(12).

Howard, A. G., Zhu, M., Chen, B., Kalenichenko, D., Wang, W., Weyand, T., Andreetto, M., and Adam, H. (2017). Mobilenets: Efficient convolutional neural networks for mobile vision applications. CoRR, abs/1704.04861.

Lin, T.-Y., Goyal, P., Girshick, R., He, K., and Dollár, P. (2017). Focal loss for dense object detection. In Proceedings of the IEEE International Conference on Computer Vision (ICCV), pages 2980–2988.

Machado, G. P. (2024). Floods in south brazil: more than an environmental crisis. The Lancet, 404(10447):24–25.

Morelli, A. B. and Cunha, A. L. (2021). Measuring urban road network vulnerability to extreme events: An application for urban floods. Transportation research part D: transport and environment, 93:102770.

Mulimani, M. and Mesaros, A. (2024). Online domain-incremental learning approach to classify acoustic scenes in all locations. In 2024 32nd European Signal Processing Conference (EUSIPCO), pages 96–100.

Negrão, P. A. S., Ranieri, C. M., Matos, S. N., and Ueyama, J. (2024). Image segmentation with zero-shot generalization for flood prediction in urban environments. In Proceedings of the 22nd IFAC Symposium on Information Control Problems in Manufacturing (INCOM), Vienna, Austria. Forthcoming.

Tao, R., Liang, L., Chen, H., Pan, W., Li, X., Liu, W., and Zhang, H. (2024). A review of cutting-edge sensor technologies for improved flood monitoring and damage assessment. Sensors, 24(13):7090.

Tian, S., Li, L., Li, W., Ran, H., Ning, X., and Tiwari, P. (2024). A survey on few-shot class-incremental learning. Neural Networks, 169:307–324.

van de Ven, G. M., Tuytelaars, T., and Tolias, A. S. (2022). Three types of incremental learning. Nature Machine Intelligence, 4(12):1185–1197.

Wang, L., Zhang, X., Su, H., and Zhu, J. (2024). A comprehensive survey of continual learning: Theory, method and application.

Yilmaz, A. E. and Demirhan, H. (2023). Weighted kappa measures for ordinal multi-class classification performance. Applied Soft Computing, 134:110020.

Zeng, Y.-F., Chang, M.-J., and Lin, G.-F. (2024). A novel ai-based model for real-time flooding image recognition using super-resolution generative adversarial network. Journal of Hydrology, 638:131475.

Zhang, Z., Zhou, Y., Liu, H., Zhang, L., and Wang, H. (2019). Visual measurement of water level under complex illumination conditions. Sensors, 19(19).