Location: Sustainable Agricultural Systems Laboratory
Project Number: 8042-30400-001-073-S
Project Type: Non-Assistance Cooperative Agreement
Start Date: Sep 1, 2025
End Date: Aug 31, 2029
Objective:
Expand the Digital Ag Science Hub (DASH) capacity to enterprise phenotyping solutions for plant breeders, plant scientists, and farmer. This funding
Specifically we will:
1. Expand SCINet compute resources - Enable integration between SCINet (the USDA high-performance cluster), relational databases, and web applications.
2. Build BenchBot imaging systems - Deploy at ARS research stations to automate image collection for training computer vision models.
3. Construct automated data pipelines - Transfer images from BenchBot directly to SCINet.
4. Curate a National Plant Agricultural Image Repository - Host on SCINet and in the cloud for broad accessibility.
5. Develop machine learning operations (MLOps) - Automate generation of neural networks for machine vision tasks.
6. Calibrate and validate deployed models - Ensure accuracy and reliability in real-world settings.
7. Develop a modular camera system - Support deployment of machine vision models across multiple farm platforms (hand-held, tractor-mounted, robotic rovers, UTVs).
8. Provide education and training - Train users on computer vision, AI tools, and phenotyping solutions using satellites, low-cost cameras, and drones.
Approach:
The Digital Agricultural Systems Hub (DASH) will deliver enterprise-level artificial intelligence solutions by tightly integrating infrastructure, imaging systems, machine learning operations, and training pipelines into a unified framework. The approach ensures that farmers and ARS researchers and breeders can access deployable, scalable, and user-friendly tools for phenotyping and precision agriculture.
1. Target Identification and Scoping
Establish a standardized intake process that identifies phenotyping targets in collaboration with researchers. This process will assess technical feasibility, platform requirements, and barriers to success, ensuring that DASH resources are directed toward the most impactful applications.
2. Imaging Systems and Data Pipelines.
Deploy BenchBot imaging systems at ARS research stations to automate image collection and construct automated pipelines that transfer images seamlessly into SCINet. Synthetic image generation approaches will augment these pipelines to ensure robust training datasets across crops and environments.
3. Compute Infrastructure and Repositories
Expand SCINet compute resources to support integration with relational databases and web applications. A National Plant Agricultural Image Repository will be curated on SCINet and in the cloud, providing a centralized, accessible platform for high-quality training data.
4. Annotation and Model Development
Provide open-source annotation and labeling tools optimized for agricultural phenotyping. Develop machine learning operations (MLOps) pipelines that automate the generation, training, and version control of neural networks, ensuring reproducibility and scalability across multiple targets.
5. Deployment and Modular Platforms
Create deployment tools that enable trained models to run seamlessly on SCINet, the cloud, and edge devices. Develop a modular camera system adaptable to diverse farm platforms—including hand-held devices, tractor-mounted systems, robotic rovers, and UTVs—ensuring broad applicability across research and production settings.
6. Calibration, Validation, and Feedback Loops
Deploy models into real-world agricultural environments for calibration and validation. Define continuous feedback mechanisms to capture performance data, driving iterative improvement of algorithms and ensuring reliability in practice.
7. Education and Training
Provide accessible education and training through online tutorials, workshops, and direct support. Training will cover computer vision and AI applications, phenotyping solutions using satellites, low-cost cameras, drones, and SCINet resources, building community-wide capacity in digital agriculture.