Abstract: In recent years, acoustic emission technology has been widely used in pipeline operation safety monitoring and structural integrity maintenance. However, the diversity of weld crack forms in pipelines leads to the complex and variable characteristics of time-varying acoustic emission signals, which limits the improvement of monitoring accuracy. Therefore, a pressure pipeline leakage monitoring system based on Markov transition field image processing technology (MTF-IPT) and a Residual Swin Transformer model was proposed. Firstly, one-dimensional acoustic emission signals were encoded into two-dimensional images based on MTF, which enhanced the correlation and temporal dependence between time-series sampling points while revealing the multi-dimensional phase space trajectories of the signals. Then, a Residual Swin Transformer deep learning network was constructed to extract the key feature information of leakage states from the two-dimensional images, enabling the identification of different leakage states. Experiments with different leakage states were designed to verify the superiority of the proposed method in multiple evaluation indicators, achieving an ideal recognition accuracy of 97.86%. Comparative experiments with other methods further demonstrate the superior performance of the proposed method in online pipeline leakage monitoring, and the system can be deployed for the operational safety maintenance of pipelines in the future.