Researchers at the University of Illinois have unveiled a groundbreaking approach to crop breeding that integrates artificial intelligence (AI) with sensor technologies. Environmental Scientist Leakey and Computer Scientist Varela invented Efficiently Supervised Generative and Adversarial Network (ESGAN). AGRA seeks to push the use of AI to hasten agricultural field research by reducing the reliance on human annotated training data. This latest approach leans on the might of a generative adversarial network (GAN). It showcases two competing AI models that ultimately work in tandem to improve the accuracy of crop assessments, including those for Miscanthus grasses.
The development comes at a time when the agricultural sector faces increasing pressure to adapt to changing climate conditions and optimize crop productivity. To find out, ESGAN has been rigorously tested against today’s AI training protocols through a series of experiments. The findings demonstrate its potential to create high fidelity images and reliably score flowering traits across multiple varieties of Miscanthus grasses.
How ESGAN Works
ESGAN operates using two primary components: a generator (G) and a discriminator (D). The generator creates realistic appearances of plant characteristics. In parallel, the discriminator judges these images, functionally sorting out real versus fake data. This competitive dynamic ensures that the models are always pushing each other to improve their performance.
Specifically, we used ESGAN to classify individual Miscanthus grasses as either flowering or non-flowering. UAV imagery from the 2020 growing season was used for this analysis. Scientists screened thousands of hybrid Miscanthus varieties, each with distinct flowering traits and schedules.
“Over time, the models improve one another,” – Varela
This iterative pipeline of learning from data to data not only speeds up the process of data generation, but decreases reliance on human annotation. According to Leakey, the new approach has “reduced the requirement for human-annotated data by one-to-two orders of magnitude.” This reduction is immensely important given how laborious and time-consuming visual inspections have typically been a part of crop breeding.
Impact on Crop Breeding
The impact of ESGAN’s capabilities goes well beyond making data collection more efficient. Flowering time is a key trait that influences productivity and adaptability in crops such as Miscanthus. From the deep South to the PNW, AGREE strongly influences agricultural prosperity on any growing land.
“Flowering time is a key trait influencing productivity and the adaptation of many crops, including Miscanthus, to different growing regions.” – Leakey
The ESGAN model intends to enable more widespread applications in crop improvement, covering multiple traits and species. By easing AI tool adoption in crop breeding, the researchers hope to bolster the bioeconomy and enhance sustainable agricultural practices.
Leakey was hopeful for the future uses of their work. “We hope our new approach can be used by others to ease the adoption of AI tools for crop improvement involving a wider variety of traits and species,” he stated.
Future Collaborations and Research Directions
Leakey and Varela are teaming up with Miscanthus breeder Eric Sacks. Together, they will employ ESGAN to gather high-quality data from a multistate Miscanthus breeding trial. This partnership will help to more widely validate ESGAN’s effectiveness and its adaptability across various geographical contexts.
As they expand their research into other areas, the team is hopeful that ESGAN can transform agricultural practices. Pairing this cutting-edge AI technology with time-tested breeding practices will help us produce crops that are more resilient to climate stressors and more productive.
“And the approach paves the way to overcome similar challenges in other areas of biology and digital agriculture,” Leakey noted.
The adoption of ESGAN into crop breeding is a model for how innovation can solve old problems within agriculture. With the world facing challenges of food security, innovations such as these are ideal examples of how new technology can pave the way for sustainable agricultural innovation.