Triplet-Loss Driven Optimization for Improved Methane Leak Detection

Methane leaks continue to pose significant threats to the environment, public health, and safety, necessitating more effective detection methods. Building upon our previous work, which utilized one-shot learning for methane leak detection using hyperspectral data, this paper introduces an approach employing triplet loss to further enhance detection accuracy. By leveraging the discriminative power of triplet loss, we establish a robust framework to compare unidentified methane signatures with known methane references. Through extensive evaluation on AVIRIS data from the historic Aliso Canyon methane leak incident, we demonstrate the superiority of our proposed method over traditional one-shot learning techniques. Notably, our approach achieves remarkable gains in both accuracy and speed. By refining the detection process, our study contributes to the advancement of methane leak detection, providing crucial insights for mitigating environmental and public health impacts.