Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

AttMetNet: Attention-Enhanced Deep Neural Network for Methane Plume Detection in Sentinel-2 Satellite Imagery

Rakib Ahsan, MD Sadik Hossain Shanto, Md Sultanul Arifin, Tanzima Hashem

TL;DR

This work tackles methane plume detection in Sentinel-2 imagery by introducing AttMetNet, a methane-aware framework that fuses the Normalized Difference Methane Index (NDMI) with an attention-enhanced U-Net. By appending NDMI as a 13th input channel and incorporating attention gates in skip connections, the model learns to prioritize plume-related features while suppressing background noise. A focal loss addresses severe class imbalance, and training on a real plume dataset demonstrates improved scene-level and pixel-level performance (higher F1 and mIoU, lower FPR) compared to strong baselines. The approach offers a robust, data-efficient solution for operational methane monitoring with potential for scalable deployment in remote sensing pipelines.

Abstract

Methane is a powerful greenhouse gas that contributes significantly to global warming. Accurate detection of methane emissions is the key to taking timely action and minimizing their impact on climate change. We present AttMetNet, a novel attention-enhanced deep learning framework for methane plume detection with Sentinel-2 satellite imagery. The major challenge in developing a methane detection model is to accurately identify methane plumes from Sentinel-2's B11 and B12 bands while suppressing false positives caused by background variability and diverse land cover types. Traditional detection methods typically depend on the differences or ratios between these bands when comparing the scenes with and without plumes. However, these methods often require verification by a domain expert because they generate numerous false positives. Recent deep learning methods make some improvements using CNN-based architectures, but lack mechanisms to prioritize methane-specific features. AttMetNet introduces a methane-aware architecture that fuses the Normalized Difference Methane Index (NDMI) with an attention-enhanced U-Net. By jointly exploiting NDMI's plume-sensitive cues and attention-driven feature selection, AttMetNet selectively amplifies methane absorption features while suppressing background noise. This integration establishes a first-of-its-kind architecture tailored for robust methane plume detection in real satellite imagery. Additionally, we employ focal loss to address the severe class imbalance arising from both limited positive plume samples and sparse plume pixels within imagery. Furthermore, AttMetNet is trained on the real methane plume dataset, making it more robust to practical scenarios. Extensive experiments show that AttMetNet surpasses recent methods in methane plume detection with a lower false positive rate, better precision recall balance, and higher IoU.

AttMetNet: Attention-Enhanced Deep Neural Network for Methane Plume Detection in Sentinel-2 Satellite Imagery

TL;DR

This work tackles methane plume detection in Sentinel-2 imagery by introducing AttMetNet, a methane-aware framework that fuses the Normalized Difference Methane Index (NDMI) with an attention-enhanced U-Net. By appending NDMI as a 13th input channel and incorporating attention gates in skip connections, the model learns to prioritize plume-related features while suppressing background noise. A focal loss addresses severe class imbalance, and training on a real plume dataset demonstrates improved scene-level and pixel-level performance (higher F1 and mIoU, lower FPR) compared to strong baselines. The approach offers a robust, data-efficient solution for operational methane monitoring with potential for scalable deployment in remote sensing pipelines.

Abstract

Methane is a powerful greenhouse gas that contributes significantly to global warming. Accurate detection of methane emissions is the key to taking timely action and minimizing their impact on climate change. We present AttMetNet, a novel attention-enhanced deep learning framework for methane plume detection with Sentinel-2 satellite imagery. The major challenge in developing a methane detection model is to accurately identify methane plumes from Sentinel-2's B11 and B12 bands while suppressing false positives caused by background variability and diverse land cover types. Traditional detection methods typically depend on the differences or ratios between these bands when comparing the scenes with and without plumes. However, these methods often require verification by a domain expert because they generate numerous false positives. Recent deep learning methods make some improvements using CNN-based architectures, but lack mechanisms to prioritize methane-specific features. AttMetNet introduces a methane-aware architecture that fuses the Normalized Difference Methane Index (NDMI) with an attention-enhanced U-Net. By jointly exploiting NDMI's plume-sensitive cues and attention-driven feature selection, AttMetNet selectively amplifies methane absorption features while suppressing background noise. This integration establishes a first-of-its-kind architecture tailored for robust methane plume detection in real satellite imagery. Additionally, we employ focal loss to address the severe class imbalance arising from both limited positive plume samples and sparse plume pixels within imagery. Furthermore, AttMetNet is trained on the real methane plume dataset, making it more robust to practical scenarios. Extensive experiments show that AttMetNet surpasses recent methods in methane plume detection with a lower false positive rate, better precision recall balance, and higher IoU.

Paper Structure

This paper contains 20 sections, 2 equations, 4 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Our contributions are summarized as follows:
  3. Related Works
  4. Methodology
  5. Normalized Difference Methane Index
  6. Model Architecture
  7. Loss Function
  8. Experiments and Results
  9. Dataset
  10. Experiment Setup
  11. Baselines
  12. Baselines
  13. Evaluation Metric
  14. Training Configuration
  15. Comparison with Baselines
  16. ...and 5 more sections

Figures (4)

  • Figure 1: Architecture of AttMetNet. NDMI is first computed from channel 11 (B11 band) and channel 12 (B12 band) of raw 12-channel Sentinel-2 data. Then it is concatenated with the original 12 spectral channels to form a 13-channel input, which is fed into the Attention U-Net model.
  • Figure 2: Structure of attention gate. Here, $g$ is the gating signal and $x$ is the encoder feature through the skip connection.
  • Figure 3: Comparison of Grad-CAM heatmaps illustrating AttMetNet activation with and without NDMI. Adding NDMI as a 13th channel results in more focused and accurate localization of target regions, as indicated by the closer correspondence between the heatmaps and ground truth.
  • Figure 4: Comparison of different model predictions for methane plumes in different geographical regions. (a) Turkmenistan (38.5602°, 54.2129°) on 7 July 2024 (b) Algeria (28.6373°, 7.6165°) on 3 January 2024 (c) Algeria (31.7779°, 5.9951°) on 27 July 2023 (d) USA (32.1068°, -103.7154°) on 19 February 2024 (e) USA (32.3635°, -101.3277°) on 4 September 2023 (f) Yemen (15.5641°, 45.7987°) on 2 January 2023 (g) Turkmenistan (39.4614°, 53.7766°) on 1 December 2024.