Spudfinder 6500: Creating a radar-based system for pre-harvest potato yield mapping, year two

Overview

Yang Dr. Peter Marchetto University of Minnesota Engineering and Technology (Devices) Potatoes require heavy watering and intensive fertilization, which negatively impacts the environment. A system designed to mitigate these environmental impacts and improve farming efficiency would be beneficial for farmers. Precision agriculture techniques accomplish this by documenting spatial variability across and within fields. One key precision agriculture technique is yield mapping. Mapping yields throughout a field allows farmers to determine how various factors like water and fertilizer usage influence farming efficiency, leading to informed decisions about how much water and fertilizer are necessary to apply to the field. Current potato yield mapping methods occur during harvest. This means that farmers can’t use their data during the growing season, instead having to wait until future years to benefit. To mitigate current yield mapping problems, last year we built a robot that detects individual underground potatoes, known as a phenocart. This phenocart uses ground-penetrating radar and machine learning for data collection and processing, allowing for non-invasive, pre-harvest detection of underground potatoes. This year, we improved our cart’s drive system by designing a chain drive using a self-designed sprocket, improved our radar’s casing by mitigating interference, tested our radar system with clusters of potatoes rather than single potatoes, and used new machine learning models to analyze the radar data and estimate the mass of buried potatoes. This work suggests methods for using radar and machine learning to estimate the mass of underground potatoes and has the potential to help farmers save money and reduce the environmental impacts of potato farming.