Classifying Wheat Genotypes using Machine Learning Models for Single Kernel Characterization System Measurements

Abstract

The properties related to market value, milling, classification, storage, and transportation of bread wheat are determined by using some important physical quality characteristics such as weight, shape, dimensions, and hardness of wheat kernels. It is possible to measure all these features using single kernel characterization system (SKCS). Classification of wheat genotypes using computer-based algorithms is crucial to determine the most accurate physical quality classification for breeding studies. In this paper, four commercial wheat cultivars (Altay-2000, Bezostaja-1, Harmankaya-99, and Kate A-1) and six doubled haploid (DH) wheat genotypes are studied to classify wheat cultivars and DH wheat genotypes separately. In the classification stage, feature vectors constructed from measured characters namely, kernel weight, diameter, hardness, and moisture are applied to well-known classifiers such as Common Vector Approach (CVA), Support Vector Machines (SVM) and K-Nearest Neighbor (KNN). Satisfactory results especially for the training set are obtained from the experimental studies. Classification results are compared with single linkage hierarchical cluster (SLHC) analysis, which is the most widely used in breeding studies. Recognition of clustered genotypes in all three classification methods and dendrograms present similar results. The SVM model is found to be outperformed over other methods for studied characters and could therefore effectively be utilized for characterizing, classifying and/or identifying the wheat genotypes.