Water scarcity is a critical global issue, with leaks and pipe failures accounting for approximately 30% of urban water loss. These inefficiencies contribute significantly to the ongoing water crisis. Early detection of leaks offers numerous benefits, including improved water distribution efficiency, reduced repair costs, and minimized environmental impact.
To address these challenges, we propose a novel approach for water leak detection using acoustic sensor signals. Our method introduces two key innovations:
- The use of a complex-valued Convolutional Neural Network (CNN) to enhance signal representation and analysis.
- Sensor fusion leveraging multiple modalities, such as hydrophones and accelerometers, to improve detection reliability.
In addition to leak detection, we also tackle the less-explored problem of leak classification, further enhancing the practical utility of our approach. We validate our method through experiments on two distinct datasets, achieving a balanced accuracy of 99% for both detection and classification tasks. Our results demonstrate that the proposed solution outperforms recent state-of-the-art methods, highlighting its potential to significantly improve water resource management.
- Image and Sound Processing Lab (ISPL)
- Department of Electronics, Information, and Bioengineering (DEIB)
- Politecnico di Milano
- Daniele Ugo Leonzio – LinkedIn
- Sara Mandelli – LinkedIn
- Paolo Bestagini – LinkedIn
- Marco Marcon – LinkedIn
- Stefano Tubaro – LinkedIn
The code will be released upon acceptance.