The application of digital image processing in grading of begonia pot plants

Abstract

Digital image processing (DIP) is a potential tool for measuring and classifying pot plants in various growth stages in an automatic, objective and consistent way, with a high capacity and low labour input. In this study, features of begonia cuttings which could be relevant for grading were analysed. Images of unrooted and rooted Begonia elatior [B. hiemalis] cuttings were acquired and analysed with DIP. The various parts of the cuttings were identified and measured using knowledge based image processing. These measurements were shown to be consistent and to be well correlated with the features measured in conventional ways. Experts graded the rooted cuttings into 3 classes: small, medium and large. The effect of grading unrooted cuttings and growing them with similar-sized cuttings was still apparent 4 weeks later - the rooted cuttings in graded units were more uniform than those in random units. Two models were constructed to determine the quality of the rooted cuttings based on DIP measurements, one based on multiple linear regression and one based on a neural network. Both were able to grade at least 75% of the rooted cuttings in the same class as the expert. The neural network-based model performed slightly (5%) better, especially for the classification of small and large plants. The lack of objective quality criteria is a major obstacle for the development of grading models for pot plants.