Non-destructive classification of root vegetable dimensions and varieties

Abstract

Monitoring the size and biomass of root vegetables during growth is essential for yield estimation, harvest timing, and crop health assessment, yet existing methods are typically destructive and unsuitable for repeated measurements. We demonstrate how 8 electrode electrical impedance tomography measurements can be used to monitor the properties of growing potatoes, carrots, and ginger by only making contact with the surrounding soil. Following theoretical validations and preliminary reconstructions, we introduce an in-situ device for Non-Destructive Root Vegetable Impedance Monitoring (NDRIM-8). Its machine learning models are trained on multi-channel data to classify vegetable type, size, and biomass, with XGBoost achieving 99–100 percent accuracy across five size classes of known location crops. When positional uncertainties are introduced, the system maintains an 81.3 percept size classification accuracy and R2 of 0.91. Our results demonstrate how impedance measurements can be used to identify and characterize root vegetables under variable soil conditions, offering strong potential for continuous, non-invasive biomass monitoring in agricultural environments.

Publication
In Computers and Electronics in Agriculture