The difficulty in establishing a reliable model for detecting and identifying boiler pipeline leakage with limited samples is addressed in this research. A support vector machine-based acoustic emission detection technology is proposed, which combines nonlinear and linear methods to classify signal parameters as feature vectors. This approach overcomes the limitations of traditional methods that require a large amount of training data for classification. It demonstrates outstanding advantages in small sample scenarios and binary classification. Firstly, the feature parameters of pipeline leak acoustic emission signals are extracted, and then put the extracted feature parameters into support vector machine classification and training as the input feature vector. Realizing the effective recognition of acoustic emission signals such as knocking, sandpaper friction and broken lead, the SVM can use the measured acoustic emission signals to judge the type of leakage signals, and the accuracy reaches 100% after training. The test results show that the method is effective and feasible in the pipeline leakage diagnosis ,verifying the acoustic emission technique is applied to the feasibility of pipeline leak detection.