Using Artificial Intelligence to Accelerate Crop Improvement

The Challenge

Developing improved crop varieties relies on understanding how physical plant characteristics, such as the size and shape of seeds and seed pods, are influenced by genetics. For plant breeders, these characteristics can be linked directly to valuable traits including yield, resilience and productivity.

Traditionally, collecting this information has required researchers to manually measure thousands of individual fruits and seed pods. This process is labour-intensive, time-consuming and prone to human error, creating a significant bottleneck in crop improvement programmes.

As breeding programmes seek to analyse increasingly large populations of plants, there is a growing need for faster, more consistent and scalable approaches to phenotyping, the process of measuring observable plant traits.

The Solution

Researchers from the Institute of Biological, Environmental and Rural Sciences (IBERS) and the Department of Computer Science at Aberystwyth University have developed a suite of artificial intelligence tools that automate the measurement of plant fruits and seed pods.

Combining computer vision, deep learning and plant science, the project enables researchers to automatically identify seed pods within images and measure important characteristics including:

• Length
• Width
• Area
• Predicted volume

The resulting measurements provide consistent, objective and highly accurate datasets that can be generated at a scale that would be impractical using manual approaches.

Delivering High-Throughput Phenotyping

One of the key outcomes of the project is MorphPod, an AI-powered image analysis tool capable of measuring large numbers of seed pods automatically. Using a dataset of more than 330,000 siliques (seed pods) from 362 plant genotypes, the team demonstrated how artificial intelligence can be used to perform high-throughput phenotyping, generating detailed measurements rapidly and consistently across very large plant populations.

In a single study, MorphPod collected detailed measurements from more than 300,000 individual fruits, many while they remained attached to the plant. This demonstrated the potential of deep learning to transform large-scale plant analysis and significantly reduce the time required for data collection.

Alongside MorphPod, researchers have also developed DeepCanola, a related tool designed specifically for brassica crops such as oilseed rape and cabbage. Together, these tools provide a flexible framework that can be applied across multiple crop species.

Connecting Plant Traits to Genetics

The value of the project extends far beyond automated measurements.

MorphPod generates detailed phenotypic data that can be linked directly to a plant's genetic makeup. Researchers used the measurements generated by the system to perform quantitative trait locus (QTL) analysis, identifying regions of DNA associated with pod size, shape and other important morphological traits.

The approach successfully reproduced previously known genetic associations, including links to the well-characterised ERECTA gene, while also revealing new candidate genomic regions associated with fruit morphology.

Importantly, the study is among the first demonstrations of deep learning-generated phenotypic data being used successfully for genetic discovery in plant biology. This provides researchers with a powerful new approach for uncovering the genetic factors that influence crop performance.

Impact

The project demonstrates how artificial intelligence can help accelerate plant breeding by:

• Automating previously labour-intensive measurements
• Producing consistent and unbiased phenotypic data
• Reducing the time required for large-scale analysis
• Enabling researchers to study much larger plant populations
• Supporting more accurate genetic analysis
• Identifying genes associated with valuable crop traits
• Accelerating the development of improved crop varieties

The technology has already shown promise across a range of agricultural species, including oilseed rape, cabbage, oats, barley and wheat, demonstrating its potential beyond the model plant systems in which it was first developed.

Open Science and Global Reach

In line with open science principles, the research team has made both MorphPod and its associated image annotation tools freely available as open-source resources.

This allows researchers and breeders around the world to replicate, adapt and apply the technology to their own crops and breeding programmes, helping to accelerate innovation across the wider plant science community.

Why It Matters

Global agriculture faces increasing pressure to produce more food while improving sustainability and resilience to environmental challenges.

By combining plant science, artificial intelligence and advanced imaging technologies, the Aberystwyth University team is helping to remove one of the major barriers to crop improvement. The project enables researchers and breeders to gather better-quality data, make more informed decisions and accelerate the development of future crop varieties.

As Kieran Atkins, PhD researcher and project lead, explains:

"AI tools like the one we have developed have the potential to revolutionise how we can develop new varieties of crops. It really is a game changer."

Looking Ahead

MorphPod and DeepCanola showcase the growing role of artificial intelligence in agricultural research. By combining advanced imaging, deep learning and genetic analysis, the project is helping to future-proof agriculture through faster, more data-driven crop improvement.

With the tools now freely available and adaptable to a wide range of species, the research provides a foundation for the next generation of AI-enabled plant breeding, supporting the development of crops that are higher yielding, more resilient and better equipped to meet future challenges.