Going for 2D or 3D? Investigating Various Machine Learning Approaches for Peach Variety Identification

Machine learning-based pattern recognition methods are about to revolutionize the farming sector. For breeding and cultivation purposes, the identification of plant varieties is a particularly important problem that involves specific challenges for the different crop species. In this contribution, we consider the problem of peach variety identification for which alternatives to DNA-based analysis are being sought. While a traditional procedure would suggest using manually designed shape descriptors as the basis for classification, the technical developments of the last decade have opened up possibilities for fully automated approaches, either based on 3D scanning technology or by employing deep learning methods for 2D image classification. In our feasibility study, we investigate the potential of various machine learning approaches with a focus on the comparison of methods based on 2D images and 3D scans. We provide and discuss first results, paving the way for future use of the methods in the field.