An efficient intelligent detection method for water pipeline leakages utilizing homologous Multi-Modal signal fusion

• Propose a novel method for water pipeline leakage detection by fusing homologous multi-modal signals • Develop a lightweight dual-branch network for efficient leakage feature extraction • Achieve a detection accuracy of 99.58% outperforming existing methods In industrial and civil fields, the leakage of water pipelines not only results in resource wastage but may precipitate significant environmental and safety issues. Consequently, to achieve timely and accurate leak detection within water pipeline systems, this paper proposes a novel lightweight leak detection method based on the fusion of homologous multi-modal signals. First of all, audio signals collected in the field are segmented and sent to two network branches respectively for feature depth extraction. Among them, a multi-channel convolutional neural network (MCCNN), which consists of multiple parallel one-dimensional convolutional neural networks (1D-CNN) of different scales, is designed to extract local timing features. Meanwhile, a two-dimensional feature extraction module is developed, in which the above audio sequences are converted into MEL spectrograms, spectrograms, and chromagrams to explore the spatiotemporal correlation and component independence. Then, a lightweight branch structure is constructed for extracting and enhancing global spatiotemporal features by combining the Squeeze-and-Excitation (SE) module with an optimized Reparameterized Vision Transformer (RepVit). Finally, the results of the fusion of one-dimensional and two-dimensional branches will be input into the Long Short-Term Memory (LSTM) layer to realize the intelligent identification and detection of water pipeline leakage . The experimental results show that the accuracy of the proposed method for water supply pipe leakage classification is as high as 99.58%, and the comparison experiments also verify that the method in this paper has more accurate classification performance and robustness in the face of noise interference.