Abstract: Oil and gas pipelines, as critical energy infrastructure, play a vital role in safeguarding national energy security and public security. Currently, third-party damage and pipeline deformation pose the greatest safety risks. Traditional monitoring methods suffer from low signal accuracy, limited long-distance stability, and poor adaptability across scenarios. Moreover, existing systems lack data fusion, intelligent analysis, and standardized sensing architectures, hindering compliance with the "1 + 1 + 1 + N" intelligent management and control requirements for pipelines. By comparing the sensing principles of distributed vibration sensing(DVS)and fiber Bragg grating(FBG)arrays, the demodulation algorithms for both were optimized. A convolutional neural network(CNN)-based event recognition model was developed using deep learning. Additionally, a multi-source dataset integrating geographic information, weather data, and video footage around pipelines was established, covering scenarios such as manual excavation, mechanical operations, and vehicle movement, to enhance model performance. Comparative tests were conducted in an experimental field under typical scenarios, including engine vibration, mechanical and pedestrian movement along pipelines, and manual or mechanical excitation at varying distances. Key performance indicators and application outcomes were analyzed, confirming the feasibility of large-capacity, long-distance FBG arrays. The optimized system significantly improves vibration source identification and monitoring accuracy, effectively adapting to third-party damage and pipeline deformation scenarios, and providing a reliable technical solution for pipeline safety monitoring.