This paper introduces an innovative approach for pipeline leak detection, addressing the noise interference challenges faced by conventional acoustic emission (AE) analysis, particularly with continuous wavelet transform (CWT) images. By enhancing CWT scalograms using Laplacian filtering, Non-Local Means noise reduction, and Adaptive Histogram Equalization (AHE), the method significantly improves contrast and reduces background noise, producing Leak-Enhanced Scalograms (LES). A Convolutional Neural Network (CNN) is employed to extract critical features from these enhanced scalograms, ensuring precise identification of leak-related patterns. For classification between normal and leak conditions, Artificial Neural Networks (ANN) are used in combination with t-Distributed Stochastic Neighbor Embedding (t-SNE) for dimensionality reduction and visualization. This approach demonstrates superior performance across various metrics when tested on an industrial-scale pipeline testbed, surpassing existing techniques and representing a substantial advancement in pipeline leak detection technology.