Assessing yield quality parameters in bush bean via RGB imagery

We observe an increasing demand for rapid and objective assessment of yield quality parameters in horticultural crops, which are featured by well-defined breeding targets and a multitude of parameters relevant for breeding processes.
In this context, this paper addresses hardware and algorithmic requirements for a robust image-based assessment of bush bean traits according to the breeders’ evaluation practice and demonstrates this with a new method for the estimation of pod length and caliber.
To achieve this we built a customized acquisition box with a calibrated camera for maximally controlled imaging conditions.
We trained a random-forest classifier with six different features extracted from 140 bean pod images for automatic identification of apical and basal pod ends including a check for the presence of peduncles, which together with the distal tips need to be excluded from the pod length measures with respect to marketing requirements.
We validated estimations of metric pod length and caliber with manual measurements from an expert breeder and achieved in both cases a high concordance with correlations of 96%. The method developed in this study will offer new opportunities for automated sensor-based assessment of quality traits in bush beans with the potential to considerably reduce the breeding effort, in particular the manual labor-intensive scorings.
The approach comes along with higher objectivity and an increased scoring resolution.

IV International Symposium on Horticulture in Europe – SHE2021

D. Jollet, U. Rascher, M. Müller-Linow

horticulture, vegetables, image analysis, plant phenotyping, plant breeding, bush bean, shape analysis

https://doi.org/10.17660/ActaHortic.2021.1327.56