GENOTYPIC CLASSIFICATION OF HARARGHE COFFEE BEANS USING IMAGING TECHNIQUES

Abstract

In this research work, a digital image analysis technique was used to classify Hararghe coffee beans of six Weredas (Kombolcha, Meta, Deder, Tulo, Bultum and Mechara) based on their color, morphological and textural features. From each wereda, 20 images were captured.Thirty six images were captured from samples collected from mechara research center. Support vector machine was used to classify HCBs and to cheek weather the genotype HCBs correctly classifies to their origin of collection or not. For the analysis, 18 color, 8 morphological and 6 textural features, totally 32 features were extracted from images of coffee beans. Four experimental classification setups were applied. Setups 1, 2, 3 and 4 are classification designs by using morphology, color, texture and combination of morphology and color features, respectively. For all setups 156 images were used as inputs of the machine. From these datasets 84 images were used for training the machine while the remaining 36 images were used for testing. The accuracy of classification using morphological feature was 100%, 100%,color features was 94.4%, 83.3%, texture features was 94.45%, 88.9% and combination of morphological and color features was 88.9%, 100% for Harar A and Harar B, respectively. The classification performance of the machine was best for morphological and mixed features. Genotypic HCBs were not correctly classified to their origin of collection. From experiment 1, 2, and 4, the percentage of images which were not classified to their origin of collection were 33.3%, 77.8% and 72.2%, respectively and all the images of Harar B genotype coffee beans were classified to Harar A genotype coffee beans and 27.78% image of Harar A cofee genotype were classified to Harar B CBs