Abstract: A method based on convolutional neural network (CNN) is proposed to solve the problem of identifying the degree of acoustic signal leakage in water supply pipe networks. Firstly, the acoustic sensor is used to collect data from the water supply pipe network and form a dataset of leakage acoustic signals. The sparrow algorithm is then employed to optimize variational modal decomposition and wavelet transform for noise removal in the leakage signal. In this paper, the leaked signal is input into CNN to extract 31 characteristic parameters. An improved CNN network is constructed based on residual block and attention mechanism using these extracted parameters, and binary classification models and multi-classification models are established for different degrees of leakage. Experimental results demonstrate that the proposed method achieves recognition rates of 94.46%, 94.82%, and 94.89% for one-dimensional acoustic signals with little, medium, and large leakages respectively; with an average recognition rate of 94.72%. Compared to current advanced methods for recognizing acoustic signals, the proposed method can be applied effectively in classifying and recognizing pipe leakage acoustic signal.