Transforming Plant Breeding with AI-Enabled Phenotyping at CIMMYT
As agricultural challenges intensify due to climate change, population growth, and resource scarcity, institutions like the International Maize and Wheat Improvement Center (CIMMYT) are stepping up to pioneer innovative solutions. At the heart of CIMMYT’s strategy is the integration of Artificial Intelligence (AI) into plant breeding through AI-enabled phenotyping. This approach harnesses cutting-edge technology to transform how plant traits are measured, making the entire process more sustainable, scalable, and objective.
The Role of AI in Phenotyping
Recent breakthroughs in AI are offering powerful new tools for agriculture, especially in the realm of phenotyping—the measurement of observable physical traits in plants. Utilizing computer vision, machine learning, and mobile device technologies, CIMMYT has developed methods that significantly improve the accuracy, consistency, and speed of trait assessment.
For instance, the ImageSafari initiative, funded by the Gates Foundation, explores how vision and language-based AI models can elevate plant trait analysis beyond traditional human methods. By doing so, this project aims to minimize errors, save time, and reduce costs, fundamentally changing the landscape of plant breeding.
What truly sets this initiative apart is its accessibility. By leveraging smartphones, users can soon generate standardized metrics for morphological traits and disease severity. This democratization of data collection empowers farmers and breeders alike, extending opportunities for participation beyond conventional research stations.
The Five-Step AI Phenotyping Pipeline
CIMMYT’s approach to AI-powered phenotyping includes a systematic pipeline designed to enhance how traits are measured in the field. It begins with capturing geo-referenced images using smartphones or tablets, equipped with tools developed by Quantitative Engineering Design (QED.ai). The first phase involves curating and annotating these images to build high-quality datasets.
1. Image Acquisition
Images are systematically captured in the field, leveraging the ubiquity of smartphones. This initial step sets the stage for comprehensive data collection.
2. Trait Identification
Advanced AI models are trained on these datasets to accurately identify key traits—such as plant counts, pod numbers, and symptoms of diseases. Validation across diverse environments ensures that these models are accurate and reliable.
3. Deployment of User-Friendly Apps
The best-performing models are made accessible through user-friendly apps or cloud-based APIs, providing breeders with instant, in-field trait predictions. This transformation not only speeds up the phenotyping process but also enhances its objectivity, empowering breeders with real-time data.
4. Data Quality Monitoring
To ensure high-quality data, protocols are established to monitor and validate the accuracy of the collected images continuously.
5. Feedback Loops
Through practical support systems like dashboards and communication tools (such as WhatsApp), quick feedback mechanisms are implemented to troubleshoot and improve data quality, further enhancing overall efficiency.
The Impact of ImageSafari
CIMMYT plays a pivotal role in the ImageSafari project, which aims to develop advanced digital phenotyping tools using AI. By collaborating with QED.ai and several CGIAR and NARS centers, this initiative focuses on compiling high-quality image datasets essential for developing robust computer vision models.
With mobile tools integrated into a high-performance infrastructure, breeders and technicians across different countries systematically capture field images. These images, linked with extensive metadata from CIMMYT’s Enterprise Breeding System (EBS), create a foundation for scalable AI-driven phenotyping tools that can support real-time trait prediction.
As Victor Musau from CIMMYT proudly states, the project has already amassed over 1 million images from various crops such as finger millet and maize. This impressive dataset not only reflects CIMMYT’s commitment but also sets the stage for expanding to additional crops, such as wheat.
Scaling Efforts Across Africa with ADCIN
The Africa Dryland Crops Improvement Network (ADCIN) is facilitating the mainstreaming of AI-enabled phenotyping within dryland crop improvement. During a pan-African workshop, scientists received training on mobile imaging applications and barcode workflows to enhance their skills.
The workshop focused on three crucial areas:
-
Standardized Imaging Protocols: Establishing harmonized Standard Operating Procedures (SOPs) for capturing multi-angle, multi-stage images of crops like sorghum and millet.
-
National Campaign Mobilization: Aligning image collection efforts with regional trial calendars to ensure robust, timely datasets.
- Real-time Feedback Loops: Implementing systems to monitor data quality and offer immediate troubleshooting support.
This collaborative framework aims to create scalable AI-powered breeding solutions across Africa, marking a significant step toward agricultural innovation on the continent.
Building Foundations for Future Innovations
CIMMYT’s investment in AI-based phenotyping, led by the ImageSafari initiative and strategic collaborations within ADCIN, is paving the way for data-driven advancements in plant breeding. With coordinated image collection efforts set to commence in August 2025, ongoing capacity-building workshops, and open-access training materials will ensure extensive adoption, local ownership, and the sustainability of AI-enhanced breeding pipelines.
As William Wu from QED.ai aptly notes, “AI is not magic. It requires data!” This sentiment underscores the importance of foundational datasets that will drive future innovations in agricultural practices and food security.
Through such comprehensive initiatives, CIMMYT is not just reshaping plant breeding but also contributing to a more sustainable and resilient agricultural future.