Computer vision and supervised machine learning techniques for classifying seeds of chickpea cultivars

Abstract

The morphological similarity among chickpea (Cicer arietinum) cultivars makes their correct identification challenging, compromising the varietal purity of the seeds. This study aimed to evaluate models for the classification of chickpea varieties using computer vision and supervised machine learning, analyzing the morphometric attributes of chickpea seeds extracted from digital images. In total, 21 color, shape and size attributes were determined from digital seed images of nine chickpea cultivars, corresponding to five kabuli and four desi cultivars. For the varietal classification, supervised learning models including Support Vector Machine, Multilayer Perceptron, Random Forest and k-Nearest Neighbors were employed. The evaluation of the models was performed through stratified k-fold cross-validation to determine the metrics performance for each model. The models with the best performances were Support Vector Machine and Random Forest, which showed high accuracy (95.37 and 94.26 %) and discriminatory capacity, according to the Matthews correlation coefficient (95.10 and 94.24 %), being considered adequate methods for the varietal differentiation of chickpea seeds.

KEYWORDS: Cicer arietinum, image analysis in seeds, cultivar identification, digital phenotyping.