DiAGRI deploys artificial intelligence at the edge — processing imaging and sensor data directly within the growth chamber facility for real-time insights and automated control. Our AI research spans multi-modal data fusion, predictive analytics, and embedded model deployment.
Edge AI Architecture
Traditional cloud-based AI workflows introduce latency that is incompatible with real-time plant monitoring and control. DiAGRI's edge AI architecture addresses this by:
- Deploying GPU-accelerated edge servers within the growth chamber facility for on-site inference
- Running AI models that process imaging data (3D scans, hyperspectral, thermal, fluorescence) in near real-time
- Sending summarised results and metadata to the cloud, while keeping raw data processing local
- Enabling closed-loop automation where AI-driven decisions (e.g., adjust light spectrum, increase humidity) are executed within seconds of data acquisition
Multimodal Data Fusion
Plants generate rich, complex data across multiple sensing modalities. DiAGRI develops AI methods that fuse these diverse data streams:
- 3D geometry from laser scanners → plant architecture, leaf area, biomass
- Hyperspectral reflectance → biochemical composition, nutrient status
- Thermal signatures → transpiration, water stress
- Chlorophyll fluorescence → photosynthetic efficiency
- Environmental time series → growth context and conditions
Our approach uses CLIP-inspired multi-modal alignment techniques to learn shared representations across modalities, enabling more robust and generalisable plant phenotyping models.
Predictive Analytics Platform
DiAGRI's AI platform provides:
- Automated phenotyping — extract quantitative plant traits (height, leaf count, biomass, health indicators) from imaging data without manual measurement
- Growth prediction — forecast plant development, yield, and harvest timing based on current state and environmental conditions
- Anomaly detection — identify early signs of stress, disease, or equipment malfunction
- Optimisation recommendations — suggest environmental parameter adjustments to achieve target outcomes