Work

Hyperspectral Methane Plume Segmentation Through Foundation Computer Vision Models

Year: 2025

Type: article

Abstract: Efficient large-scale monitoring of methane emissions is crucial for sustainable environmental management, necessitating advanced remote sensing techniques that process hyperspectral imagery to provide detailed insights into the Earth’s surface and atmosphere. This technology is vital for identifying methane plumes from point sources within the fossil fuel industry, essential for mitigating climate change and minimizing environmental degradation. Despite advancements, current remote sensing methods often generate significant false alarms that require manual verification by experts, limiting scalability and practicality. To overcome these challenges, our study introduces a novel hybrid model combining a matched filter with the Segment Anything Model (SAM), a foundation model renowned for its zero-shot generalization capabilities. This model efficiently segments methane plumes from hyperspectral images without additional training on customized datasets, streamlining the detection process and enhancing both efficiency and scalability. Additionally, we have developed an iterative scheme to refine methane signatures and improve anomaly detection accuracy. This approach addresses the challenge of acquiring data for cold-start segmentation and reduces the computational burden of training on specialized datasets. The combined capabilities of our methodology not only make it cost-effective but also robust for sensor-agnostic implementation. By leveraging remote sensing and cutting-edge machine learning, our method improves the precision and reliability of environmental monitoring, crucial for advancing global environmental management and sustainability efforts.

Source: IEEE Access

DOI: 10.1109/access.2025.3597991

Authors: Yuan Zi, Chenyu Zhou, Jiefu Chen, Zhu Han, Yueqin Huang, Xuqing Wu

Institution: University of Houston

Language: English

Topic: Remote-Sensing Image Classification

Subfield: Media Technology

Field: Engineering

Domain: Physical Sciences

SDG: Climate action

Open Access status: gold

Estimated APC paid: $2,075