Pipelines play a crucial role in chemical industrial production. However, due to long operating cycles, seal failures, and internal corrosion, hazardous chemical media are prone to leak, potentially leading to serious accidents such as explosions. To address the limitations of existing pipeline leak detection methods—specifically their insufficient recognition accuracy and poor robustness in noisy environments—this paper proposes an Acoustic Emission (AE)-driven leakage state recognition method based on wavelet time-frequency maps and the Inception-V3 deep network. First, a pipeline leak experimental platform was constructed, and AE signals were collected. The signals were denoised through wavelet decomposition reconstruction. Then, the continuous wavelet transform (CWT) was applied to perform time–frequency analysis of the AE signals, generating wavelet time–frequency maps as the dataset. Finally, a deep learning classification model based on Inception-V3 was developed to identify different pipeline leak states. Experimental results show that the proposed method achieves a recognition accuracy of 99.6%. Compared with other network models and feature-based support vector machine (SVM) models, this method exhibits superior robustness in high noise and high recognition accuracy under small leakage conditions, confirming its effectiveness and advantages in pipeline leak detection.
Zheng et al. (2026) studied this question.