Classification of the seeds of traditional and double‐low cultivars of white mustard based on texture features

Abstract

AbstractThe aim of this study was to develop classification models based on the texture features from the images acquired with a flatbed scanner to distinguish between the seeds of traditional (Radena) and double‐low (Warta) cultivars of white mustard. Texture features were calculated in MaZda software. The classification models for RGB, Lab, and XYZ color space and individual color channels for distinguishing between the seeds of traditional and double‐low cultivars of white mustard were built in the WEKA application with the use of selected Decision Trees (J48), Rules (J Rip), Bayes (Naive Bayes), Lazy (IBk), Meta (Multi Class Classifier), and Functions (FLDA) classifiers. The classification accuracy of the evaluated color models ranged from 77% (XYZ, IBk classifier) to 83% (RGB, JRip classifier). In an evaluation of individual color channels, classification accuracy was highest for channels R, L, and X, and it ranged from 85% for color channels L and X when the IBk classifier was used to 93% for the texture features from color channel R when the JRip classifier was used.Practical applicationsThe developed models support fast, cheap, nondestructive, and reliable discrimination of the seeds of white mustard. The proposed classification models based on texture features effectively distinguished between the seeds of traditional (Radena) and double‐low (Warta) cultivars of white mustard. Therefore, they can be reliably used to examine the authenticity of seeds and to detect seed adulteration.