Cultivar discrimination of stored apple seeds based on geometric features determined using image analysis

Abstract

Digital image analysis can allow determining the features of seeds and their interpretation in a fast, inexpensive, and non-destructive way. The research aimed to develop discriminative models based on geometric features to distinguish seeds belonging to different apple cultivars. Images of seeds of apples ‘Gala’, ‘Jonagold’ and ‘Idared’ were acquired using a flatbed scanner. The linear dimensions and shape factors were calculated, and after attribute selection, they were used for the discriminant analyzes of seeds for ‘Gala’ with ‘Idared’, ‘Gala’ with ‘Jonagold’, ‘Idared’ with ‘Jonagold’ and ‘Gala’ with ‘Idared’ with ‘Jonagold’ with the use of discriminative classifiers from Bayes, Function, Meta, Rules and Decision Trees groups. In the case of models build based on selected linear dimensions, the accuracy of discrimination was equal up to 84% for distinguishing seeds of all three apple cultivars for the J48 classifier from Decision Trees and 93% for analysis of ‘Gala’ and ‘Idared’ for the J48 from Decision Trees. The slightly lower correctness of up to 82% for discrimination of ‘Gala’, ‘Idared’ and ‘Jonagold’, Multi Class Classifier) and 90% (‘Gala’ and ‘Jonagold’, J48) were obtained for models build based on selected shape factors. The analyzes performed based on sets including selected combined linear dimensions and shape factors of seeds provided the correctness of up to 86% for the discrimination of three apple seed cultivars (J48) and 91% (‘Gala’ and ‘Jonagold’, J48). The results can be used in practice for the assessment of the authentication of seeds of apple cultivars with high probability.