ICICLE | September 23, 2026
https://icicle.osu.edu/news/2026/09/icicle-demonstrates-ai-cyberinfrastructure-precision-agriculture-farm-science-review-2026
LONDON, Ohio — At Farm Science Review 2026, the ICICLE AI Institute and its collaborators will demonstrate how artificial intelligence, high-performance computing, edge devices, and interoperable geospatial data can work together to support practical, field-level agricultural decision-making.
The demonstrations highlight a central ICICLE goal: making advanced AI capabilities accessible beyond centralized cloud environments by connecting data collection, model inference, computing infrastructure, and user-facing tools into usable end-to-end workflows. Led by The Ohio State University, ICICLE is the National Science Foundation-funded AI Institute for Intelligent Cyberinfrastructure with Computational Learning in the Environment.
From drone imagery to precision spot spraying
Visitors will see an end-to-end precision spot-spraying workflow that begins with drone-based field scouting and ends with targeted treatment. Rather than spraying an entire field uniformly, the workflow identifies areas where weeds are present and produces a treatment plan that distinguishes spray zones from no-spray zones. This targeted approach has the potential to reduce unnecessary pesticide application, lower treatment costs, and support more sustainable crop-management practices.
The workflow is coordinated by an orchestrator that connects each stage:
• A laptop-based flight planner creates a drone scouting mission.
• Captured imagery is transferred automatically to a high-performance computing cluster.
• Zero-shot annotation models identify plants and weeds in the imagery.
• The resulting detections are incorporated into a GeoPackage containing the field boundary, detected objects, and designated spray and no-spray areas.
• The system generates a waypoint-based drone flight plan for targeted spraying.
• The spray drone can then execute the treatment mission using the generated waypoint file.
The demonstration illustrates how cyberinfrastructure can turn raw aerial imagery into an actionable, location-specific management plan with limited manual data handling.
Accelerating annotation with Smart Labeler
ICICLE will also demonstrate Smart Labeler, a tool designed to reduce one of the most persistent barriers to applying AI in agriculture and ecology: creating high-quality labeled datasets.
Users can provide simple coordinate-based or text prompts to identify objects of interest in new image collections. They can then inspect, alter, filter, and analyze the generated labels. The system supports both:
• Zero-shot labeling, in which models identify objects without task-specific examples supplied by the user.
• Few-shot labeling, in which previously annotated images serve as examples or ground-truth references for labeling new imagery.
The demonstration integrates multiple foundation-model components across proposal and classification stages, including SAM3, OWLv2, and DINOv3. Together, these models support the detection and classification of objects relevant to agricultural and ecological datasets.
By shortening the path from raw data to usable annotations, Smart Labeler can help researchers and practitioners prepare training and evaluation data more efficiently while retaining the ability to review and refine model outputs.
Bringing AI to the field with IEMS
A third demonstration focuses on operating AI systems at the agricultural edge—where cameras and sensors collect data in the field, rather than only in remote computing environments.
Using ICICLE’s Intelligent Edge Management Service (IEMS), the team will demonstrate deployment and management of a YOLO weed-detection model on a field-hardened NVIDIA Jetson AGX Xavier device connected to Hikrobot GigE cameras. IEMS packages and deploys the application container and model weights to the device, streams live camera data to a portal, and enables an operator to start, stop, and monitor inference from a laptop.
The workflow is designed to avoid manual configuration of individual devices. A model can be prepared in the lab, hardware can be enrolled, and the detector can be deployed to an outdoor sensing site through a coordinated edge-management process.
This capability is especially relevant for agricultural applications in which connectivity, latency, privacy, data volumes, or operational needs make it valuable to run AI near the point of data collection.
Asking questions of a field
The team will also present Ask the Farm, a conversational AI tool for exploring field-level agricultural and environmental information. Built on harmonized public geospatial data layers ingested through ICICLE AI’s Earth Data Hub, Ask the Farm enables users to select a field at the Molly Caren Agricultural Center—or upload their own boundary—and ask questions in plain language.
The system draws on data layers that include:
• Sentinel-2 satellite imagery
• USDA crop maps
• NRCS soil-survey information
• USGS watersheds, streams, wetlands, elevation, and slope data
• Rainfall information
Ask the Farm is designed to answer questions about crop history, soils, rainfall, drainage, and landscape characteristics. Responses are generated using expert co-designed, open-source foundational large language models operating on national academic high-performance computing resources. The tool grounds its responses in the available source data, reports relevant source, spatial-resolution, and date information, and explicitly identifies questions that cannot be answered from the available records.
The demonstration shows how cyberinfrastructure can make complex, multi-source geospatial data more discoverable and usable through a conversational interface—while maintaining attention to data provenance and limitations.
A connected vision for agricultural AI
Together, these demonstrations show several parts of an integrated agricultural AI ecosystem: automated data collection, scalable computing, AI-assisted labeling, edge deployment, geospatial data harmonization, and natural-language access to field intelligence.
At Farm Science Review, ICICLE’s team will invite visitors to examine these workflows, observe live demonstrations, and discuss how similar capabilities could support weed management, in-field sensing, ecological monitoring, and data-driven agricultural research.
Farm Science Review 2026 runs September 22–24 at The Ohio State University’s Molly Caren Agricultural Center near London, Ohio.