New machine learning based functions for TubAR measure skin finish, pressure bruise, and chip and fry color
Date & Time
Tuesday, July 19, 2022, 1:45 PM

TubAR (Tuber Analysis in R) is an image analysis package in the R programming language which provides quantitative measurements of several tuber traits. The first version of the package measured shape and skin traits including tuber length, width, and roundness, skinning percentage, redness, and lightness. In order to increase our capacity to evaluate complex skin traits, we employed machine learning image classification methods to measure traits including skin finish and pressure bruise. Images of tubers from multiple breeding programs exhibiting a variety of skin traits were used to develop image classifier models, and cross validation within this sample of images was utilized to determine the accuracy of classifiers. These models will be used to provide quantitative measurements of tuber skin traits. Upon the release of TubAR in 2021, there was great interest in the packages possible application to measuring color traits in chips and French fries. Lightness in chips and stem end browning in fries were measured using sample images from the University of Minnesota and USDA. Chip lightness scores were compared to Hunter Lab Lightness scores acquired using a colorimeter, while fry lightness scores were compared to reflectance values from a Photovolt device. The protocol used here for the creation of classification models can potentially be used for an even wider variety of tuber skin traits. Furthermore, the limited changes to the package necessary to measure chip and fry traits suggest potential for TubAR or derivative programs to be used in applications beyond potato tubers.

Session Type
Parent Session
7/19 - Concurrent Sessions F