Abstract
This study aims to develop and validate an automated approach for large-scale pipeline leak detection using fixed-wing unmanned aerial vehicles (UAVs) equipped with thermal imaging sensors and convolutional neural networks (CNNs). Pipeline leakages represent a critical challenge for energy infrastructure operators because of their environmental, economic, and safety implications, particularly in large and geographically distributed pipeline networks where ground-based inspections are costly and time-consuming. The proposed methodology integrates long-range thermal image acquisition from fixed-wing UAV flights with a deep-learning-based detection pipeline designed to identify and localize leakage events under real-world operating conditions. Unlike conventional approaches that predominantly rely on rotary-wing platforms or visible-spectrum imagery, the presented framework leverages the extended coverage capabilities of fixed-wing UAVs and the robustness of thermal imaging to improve monitoring efficiency across large areas. The main contributions of this work include the development of an integrated thermal-based leak detection framework tailored for fixed-wing UAV operations, the application of CNNs for automated leak identification and localization, and comprehensive validation under both simulated and real-world conditions. Quantitative evaluation demonstrates reliable performance, achieving a leak localization root mean square error of 0.8989 m on straight pipeline segments and an overall detection accuracy of 81%. The proposed approach is directly applicable for pipeline operators, infrastructure monitoring agencies, and energy companies by enabling early leak identification, reducing inspection costs, and supporting safer and more efficient maintenance planning. Furthermore, the presented framework provides a foundation for future research focused on larger annotated datasets and integration with predictive maintenance systems.