Leakage in water supply pipelines results in wasted water resources, increased costs, and serious issues such as ground subsidence and environmental pollution, significantly impacting residents’ lives and urban operations. Therefore, effective leak detection is essential for conserving water resources, protecting the environment, and ensuring infrastructure safety. Traditional detection methods suffer from low efficiency, high costs, and limited accuracy. In recent years, deep learning-based approaches have advanced water leakage detection, but challenges remain, including difficulty in capturing long-term dependencies and local features, as well as weak generalization ability. To address these challenges, this paper proposes a Dual Attention Contrastive Representation Learning-based Water Pipeline Leakage Detection (DACRL-WPLD) method. This approach captures long-term dependencies and local features through a multi-scale design and dual attention mechanism while improving generalization ability via contrastive learning. Experimental results demonstrate that, compared to the MTCNN model, DACRL-WPLD improves accuracy, recall, and F1-score by 3.64, 1.66, and 1.55 percentage points, respectively. Compared to the OCSVM model, it improves these metrics by 2.00, 4.66, and 2.39 percentage points, respectively. Furthermore, experiments conducted on publicly available datasets such as SMD, MSL, SMAP, SWAT, and PSM further validate DACRL-WPLD’s strong generalization ability across different application scenarios. These results confirm that the proposed method significantly enhances the accuracy and efficiency of water leakage detection, making it a promising solution for real-world applications.