The advancement of machine learning in oil leak detection systems is transformative, with the potential to revolutionize environmental protection, resource management, and industrial safety. However, transitioning from traditional methods to machine learning-based techniques will be a gradual process over several years, involving the incremental introduction of new features and capabilities already utilized in modern leak detection systems. In this work, we explored machine learning-based leak detection, examining various techniques and their applications in the field. We also reviewed several image analysis methods due to their critical role in safely and efficiently developing and testing leak detection systems. The ultimate goal was to create a fully automated leak detection system. To achieve this, we used a U-Net model fortified with an InceptionResNetV2 backbone, a cutting-edge approach in machine learning. Before implementation, we conducted an in-depth study and explanation of this approach. The model was successfully tested for airborne monitoring, yielding an Intersection over Union (IoU) of 75.30%. The results were presented and discussed in detail.