With Farmgazer, students are bridging the gap between machine learning and modern agriculture—helping farms detect problems early through automation and user-informed design.
By Justin Horne
A team of graduate students from the Global Innovation Exchange (GIX), the University of Washington’s engineering-and-business institute for emerging technology leaders, is tackling a major issue in large-scale farming: monitoring crops for problems before they can cause significant damage. The students’ project, sponsored by Microsoft and Nelson Farms, led to an AI-powered yield monitoring system that aims to help farmers detect problems such as pests, weeds, and crop diseases.
The Problem: Large Farms, Slow Detection
According to the team’s research, 14% of yield losses from pests and diseases could be prevented through early intervention. The team’s goal was to improve monitoring efficiency using AI-powered automation, helping farmers act before problems escalate.
Modern farms are vast, and often span 1,000 acres or more, making manual monitoring slow and inefficient. “It takes a week for one or two people to manually walk around the farm and monitor if there’s any problem happening,” explained Haoran Zeng, a student in the MS in Technology Innovation (MSTI) program, the flagship degree at GIX. That delay can mean issues like pests, weeds, water pooling, or disease spread before they’re noticed, leading to significant crop loss.
The Solution: AI-Driven Monitoring with Agriculture-Specific Models
To build their system, the team has been working on Farmgazer, a capstone project that was initiated by another student team at GIX last year, which uses cameras and computer vision to monitor crops. “They had good hardware as a monitoring device, but there is still room for improvement in developing a more intelligent and seamless system to ease farm owners' high cognitive load from diverse information and problems,” said MSTI student Wanling Yu.
“Our user research told us that farmers wanted help categorizing the problems,” added Yu. “So now, after the analysis, it’ll tag the image, and users can quickly choose which category they want to focus on.” This approach saves farmers significant time by letting them immediately pinpoint potential threats rather than sifting through all the photos manually.
The Role of Hardware: Sensors and Deployment at Scale
While the primary focus is software, the team is making numerous hardware adjustments. “We are mainly focusing on the software part, but we’re also trying to make the hardware more compact,” said Zeng. “Right now, their enclosures are just boards bought off the internet. We want to design enclosures that fit the farming needs better,” he added.
“And right now, the system only uses cameras to detect issues,” added Yu. “We’re adding sensors to give them even more contextual information.
Even with this advanced hardware and AI, monitoring large farms requires many devices. “For a 1,000-acre farm, they’ll probably need 50 to 60 devices,” estimated Zeng.
Learning from Farmers: Trust and Usability Challenges
One of the biggest concerns farmers raised was trust; specifically, farmers were concerned about the risk of putting so much faith in AI recommendations. “False negatives are very important,” explained Yu. “If there’s a problem and the AI doesn’t detect it, that can contribute to crop loss. This insight from farmers was crucial in shaping our design”
“Our system gives an initial categorization to each potential problem and gives the farmers the ability to manually correct any mistakes. It gives them more confidence in the system,” said Yu. This human-in-the-loop approach makes the system more credible and reliable while also improving the AI’s accuracy for future detections.
Next Steps: Scaling and Field Deployment
Right now, the GIX team is still developing the software, but their goal is to deploy a functioning prototype on a real farm. “We plan to scale both the software and hardware,” said team member Joel Zhu. “We also want to deploy it on our sponsor’s farm and have his employees try the app on their phones.”
The hardware design also needs further refinement, particularly making the enclosure weatherproof and durable for field conditions. The team also hopes to integrate AI-powered camera panning and tilting to improve coverage. “We’ve gotten great feedback from Dr. John Raiti [GIX Technical Advisor and Associate Teaching Professor], and have some ideas,” said Yu. “We want to build in a low-cost distributed automatic panning and tilting camera system to get a broader vision of the farm.”
A critical part of making the system viable for real-world use came from the work of team member Zia Sun, who was responsible for building the project’s backend infrastructure and deploying it to the cloud. Sun designed and implemented the FastAPI-based interface that serves as the central communication hub between the AI models, hardware devices, and mobile application. “Without a fast and reliable backend, the system can’t deliver timely insights to farmers,” said Sun.
Beyond API development, Sun also led the deployment of the application to a cloud platform, enabling real-time access, centralized data management, and system scalability. This cloud-based architecture is essential for supporting large-scale farm operations, where dozens of devices generate continuous data across hundreds or even thousands of acres. Sun’s contributions not only made the current prototype functional – they also laid the groundwork for future expansion and commercialization of the system.
The Bigger Picture: AI Meets Agriculture
For the students, this project isn’t just about building technology – it’s an introduction to new types of problems. “It’s quite exciting and different from the tech we’re used to,” said Zhu. “So much of it depends on weather or nature, and you just have to accept that and learn how to do better.”
Visiting a working farm also reinforced the project’s impact. “The farm visit was so exciting,” said Yu. “Seeing the real crops really helped us realize that our efforts can make a real impact on the farm.”
With the agriculture industry doesn’t appear inherently tech-driven or connected, the team hopes their work will help make farms smarter, more efficient, and more resilient in the face of growing challenges.
Ready to work on projects like these? Learn more about the MS in Technology Innovation or apply to be a student.
Interested in sponsoring a project? GIX capstone projects offer corporate sponsors up to six months of access to international and interdisciplinary teams of graduate students in the University of Washington’s project-based Master of Science in Technology Innovation. Learn more here.
Back to News
