Natural gas pipeline leak detection method based on 1D CNN-Mamba integrated acoustic signal classification

Abstract The transportation of natural gas requires the construction of complex pipeline systems. However, as the pipelines age, damage and leakage incidents occur frequently, making pipeline leak detection particularly critical. This paper proposes an acoustic signal classification method integrated with an improved 1D CNN-Mamba model for pipeline leak detection. The method combines the 1D-CNN and Mamba modules, training them separately before jointly fine-tuning, significantly improving detection accuracy. By utilizing depthwise separable convolutions, the model parameters are reduced, and the computational cost is lowered with the Mamba block. Experimental results demonstrate that the proposed method offers a significant performance improvement, with an average accuracy of 93.24%, a 0.74% increase compared to the PNN in 12 classifications.