Abstract: In the process of corrosion acoustic emission monitoring of in-service pipelines, the flow noise due to the flow of the medium inside the pipe will also be picked up by the sensors, which results in the corrosion signals being overwritten and thus making it difficult to determine whether corrosion has occurred or not. To address this problem, a corrosion damage acoustic emission identification model based on variational mode decomposition (VMD), Hilbert transform (HT), and deep bi-directional GRU neural network (BiGRU) was proposed. The model was able to adaptively transform the original signal into multiple eigenmode components, and extracted the instantaneous frequency and spectral entropy of each component to construct a multidimensional time-series feature matrix, and then establish the mapping relationship between the original signal and the multidimensional features. In order to verify the effectiveness of the method, corrosion acoustic emission monitoring experiments of in-service pipelines were carried out, and the experimental results showed that the proposed model had good robustness under the flow noise environment, and the recognition accuracy of the monitoring data reached 96.3%, and it was able to provide a new solution for the corrosion acoustic emission monitoring of in-service pipelines.