Distinguishing Seed Cultivars of Quince (Cydonia oblonga Mill.) Using Models Based on Image Textures Built Using Traditional Machine Learning Algorithms

Abstract

Different cultivars of seeds may have different properties. Therefore, distinguishing cultivars may be important for seed processing and product quality. This study was aimed at revealing the usefulness of innovative models developed based on selected image textures built using traditional machine algorithms for cultivar classification of quince seeds. The quince seeds belonging to four cultivars ‘Uspiech’, ‘Leskovac’, ‘Bereczki’, and ‘Kaszczenko’ were considered. In total, 1629 image textures from different color channels for each seed were extracted from color images acquired using a flatbed scanner. Texture parameters were used to build models for a combined set of selected textures from all color channels, sets of selected textures from color spaces RGB, Lab, and XYZ, and individual color channels R, G, B, L, a, b, X, Y, and Z using algorithms from different groups. The most successful models were developed using the Logistic (group of Functions), IBk (Lazy), LogitBoost (Meta), LMT (Trees), and naïve Bayes (Bayes). The classification accuracy reached 98.75% in the case of a model based on a combined set of textures selected from images in all color channels developed using the Logistic algorithm. For most models, the greatest misclassification of cases was observed between seeds ‘Bereczki’ and ‘Kaszczenko’. The developed procedure can be used in practice to distinguish quince seeds in terms of a cultivar and avoid mixing seed cultivars with different properties intended for further processing.

An RGB image dataset for seed germination prediction and vigor detection - maize

“…The digital image of soybean was obtained by using RGB, and the character of soybean was evaluated automatically by using Python Algorithm ( Ghimire et al., 2023 ). The performance of a neural network-based model to identify plant species from paramo seeds via optical RGB images ( Ropelewska et al., 2023 ). High-quality datasets are crucial for accurate machine vision algorithms in seed vigor detection and classification.…”