Abstract <p>As a critical means of energy transportation, oil and gas pipelines transport essential resources like oil and natural gas. However, as the service life of these pipelines increases, they encounter various adverse factors that gradually compromise their structural integrity, making them prone to defects. These defects may lead to serious safety incidents, posing threats to human lives, property, and the ecological environment. Traditional defect recognition methods often fall short in accuracy and reliability due to the complexity of the pipeline operating environment. This paper proposes a novel damage recognition method for oil and gas pipelines that combines multiscale adaptive convolution (MSDAC) with a long short-term memory network (LSTM). The MSDAC module integrates a multiscale convolution network with a channel attention mechanism, enabling it to extract multiscale defect features from the original acoustic signals and dynamically adjust feature weights at different scales to enhance key feature responses. The output from the MSDAC module forms a new multiscale feature vector, which serves as input for the LSTM to capture contextual features with temporal dependencies, improving defect identification performance. The proposed method is validated using acoustic signals from actual pipeline defects. Experimental results demonstrate that its accuracy, precision, recall, and F1-score exceed 95% for identifying defects such as No defect, Crack, Delamination, Pit, and Perforation. The method outperforms traditional models, showcasing excellent noise resistance and generalization ability across varying conditions, thus confirming its efficacy in oil and gas pipeline defect recognition and classification.</p>