Deep learning-based potato disease detection for variable rate spraying

Overview

This solution applies deep learning to automatically detect diseased potato plants in field conditions. By processing images of healthy and infected plants through a trained model, the system enables real-time disease identification during field operations. The technology is designed to integrate with a variable rate sprayer platform, allowing site-specific application of agrochemicals only where disease or weed pressure is detected. This targeted approach helps farmers reduce input costs, limit over-application of petrochemicals, and minimize environmental degradation.

The project builds on prior work in the laboratory, where a masters student successfully tested a variable rate sprayer model for weed detection. The current effort extends that foundation to broader disease detection across potato crops, with the goal of developing a prototype suitable for patent registration and eventual commercialization.

Technical specifications

• Deep learning-based image classification trained on labeled images of healthy and diseased potato plants
• Real-time detection capability intended for integration with existing variable rate sprayer field equipment already developed in the laboratory
• Site-specific application logic that triggers agrochemical spraying only when disease or weed presence is confirmed, reducing blanket application
• Scalable image dataset approach allowing expansion to additional potato diseases common to Northern North America
• Variable rate sprayer compatibility with the existing field buggy platform for field validation

Technology readiness level

The underlying variable rate sprayer platform has been previously tested in the laboratory for weed detection. The current stage focuses on expanding the deep learning model to detect potato diseases beyond weeds. Future validation will involve capturing and labeling additional field images of a common Northern North American potato disease, training the model, and deploying it on the existing field buggy for real-time detection and site-specific spraying trials. Following successful field validation, the team plans to develop a commercial prototype and pursue patent registration.