AI-Driven Methane Detection for Livestock Monitoring
A comprehensive three-phase research program developing artificial intelligence techniques combined with optical gas imaging technology to detect, segment, and quantify methane emissions from livestock operations.
Project Overview
Livestock methane emissions represent a critical challenge in climate science, making automated monitoring systems essential for effective mitigation strategies.
Combining artificial intelligence with optical gas imaging technology for real-time methane detection and quantification.
Systematic progression from proof-of-concept to practical real-world applications with demonstrated farm deployment capability.
Livestock methane emissions represent a critical challenge in climate science, accounting for 32% of human-caused methane production globally. As the world's population is projected to reach 9.7 billion by 2050, the demand for livestock products will increase by 70%, making automated monitoring systems essential for developing effective climate mitigation strategies.
Our comprehensive research program addresses this challenge through the development of artificial intelligence techniques combined with optical gas imaging technology to detect, segment, and quantify methane emissions from livestock operations. This three-phase research initiative demonstrates a systematic progression from fundamental proof-of-concept work to practical real-world applications.
Research Phases
A systematic three-phase progression from fundamental algorithmic development to practical real-world deployment.
View Complete Technical DetailsResearch Goal
Develop the first transformer-based architecture for low-flow rate methane emission segmentation in livestock applications.
Key Innovation: Gasformer Architecture
- • Encoder: Mix Vision Transformer (MiT-B0) for multi-scale feature extraction
- • Decoder: Light-Ham decoder with Hamburger Matrix Decomposition
- • Achievement: Effective detection at concentrations as low as 10 SCCM
Experimental Setup
- • Equipment: FLIR GF77 OGI Camera (uncooled LWIR, 320×240 resolution)
- • Detection Range: 7-8.5 μm spectral range, optimized for methane's 7.7±0.1 μm absorption band
- • Environment: Ice background for consistent temperature contrast
Datasets Created
Controlled Methane Release (MR) Dataset
- • 9,237 images across 10-100 SCCM flow rates
- • Precision gas flow controller for accurate control
- • Multiple color modes (white hot, black hot, rainbow)
Dairy Cow Rumen Gas (CR) Dataset
- • 340 images from real dairy cow rumen gas samples
- • 24-hour ANKOM batch culture system
- • Direct livestock emission capture
Results & Impact
Validation: Successful transfer learning from controlled to real livestock scenarios
Research Evolution & Future Directions
The progression of our research program demonstrates a carefully orchestrated evolution from fundamental algorithmic development to practical real-world application.
- • First transformer architecture for livestock methane detection
- • Largest annotated dataset for agricultural gas emission analysis
- • Novel dual-task framework combining detection and classification
- • Comprehensive validation against analytical chemistry standards
- • Real-world deployment capability for practical farm applications