Overview:

CIMMYT is integrating artificial intelligence, computer vision, and mobile devices to make phenotyping—measuring plant traits—faster, more scalable, and more objective for dryland staple crops across Africa. This approach reduces labor, boosts consistency, and accelerates breeding timelines from field data capture to decision-making.

What’s new:

• Smartphone-first phenotyping: Breeders capture geo-referenced field images on phones/tablets; AI models convert images into trait measurements in near real time.
• Vision-and-language AI (ImageSafari): A pipeline that uses annotated images to train, validate, and deploy models for traits, minimizing human error and subjectivity.
• Low-barrier deployment: Cloud APIs and mobile apps put advanced phenotyping into resource-constrained programs without specialized hardware.
• Partnership engine: Built with QED.ai, the Alliance of Bioversity International & CIAT (ABC), the Gates Foundation, and the Africa Dryland Crops Improvement Network (ADCIN).

Five-step pipeline:

1. Capture: Collect geo-tagged, standardized plot images in the field.
2. Annotate & Train: Label traits; train machine-learning models on diverse germplasm and environments.
3. Validate: Test for accuracy, robustness, and transferability across sites/seasons.
4. Deploy: Integrate into mobile apps or APIs for real-time trait scoring.
5. Decide: Feed reliable trait data into selection indices to advance lines faster.

Why it matters:

• Speed: Cuts phenotyping bottlenecks and shortens breeding cycles.
• Precision: Consistent, objective trait scoring across locations and teams.
• Access: Democratizes high-quality phenotyping for national programs and partners.
• Scale: From pilot plots to multi-site networks using the same digital workflow.

At a glance: