TRACE: Thermal Recognition Attentive-Framework for CO2 Emissions from Livestock
A new AI framework analyzes thermal video to track methane from free-roaming cattle, achieving near-perfect segmentation.
A research team has introduced TRACE (Thermal Recognition Attentive-Framework for CO2 Emissions from Livestock), a novel AI framework designed to solve a critical agricultural and environmental monitoring problem. The system analyzes mid-wave infrared (MWIR) thermal video to perform two tasks simultaneously: segmenting CO2 plumes exhaled by individual, free-roaming cattle in each video frame, and classifying the overall emission flux over a sequence of frames. This dual-objective approach is a first, moving beyond simple detection to provide quantifiable, per-animal emission data without physical contact or confinement.
TRACE's technical innovation lies in its specialized architecture. Its core is a Thermal Gas-Aware Attention (TGAA) encoder, which uses the intensity of the gas plume itself as a spatial guide to focus the model's self-attention mechanism on the most relevant areas of the thermal image. This is paired with an Attention-based Temporal Fusion (ATF) module that models the dynamics of the breath cycle across frames. The team employed a four-stage progressive training curriculum to effectively train this complex, multi-task model without performance interference.
Benchmarked on the CO2 Farm Thermal Gas Dataset, TRACE delivered a standout performance. It achieved a mean Intersection over Union (mIoU) score of 0.998 for plume segmentation and topped all metrics for both segmentation and flux classification. This performance surpassed fifteen other state-of-the-art models, including specialized gas segmenters that were significantly larger. Ablation studies confirmed that both the gas-conditioned attention and the temporal reasoning components were essential for the system's precision and discriminative power.
- Achieves 0.998 mIoU for CO2 plume segmentation, outperforming 15 larger models.
- Uses a novel Thermal Gas-Aware Attention encoder to focus on emission hotspots in thermal video.
- Enables continuous, per-animal emission monitoring without physical contact, a first for free-roaming livestock.
Why It Matters
Provides a scalable, accurate tool for agricultural carbon accounting and monitoring livestock metabolic health, key for sustainability and farming efficiency.