CLASSIFICATION OF GENETICALLY VARIABLE WOLLEGA COFFEE BEANS USING IMAGING TECHNIQUES AND ARTIFICIAL NEURAL NETWORK

Abstract

In this research work, a digital image analysis technique was used to classify Wollega coffee beans genotype based on their color and morphological features and performance evaluation (Accuracy, Specificity and Sensitivity) were done. Artificial neural network (ANN) was used to classify WCBs that are grown in Haru Coffee Research Center (under the same agronomical management), whether correctly classifies to their origin of collection or not. Three classification set-ups were used, which are classification based on color feature, morphological feature, and combination of color and morphological features. For WCBs genotype, all set up features of 500 images were used as inputs of the ANNs. From those data sets 60% (300 images), 20% (100 images) and 20% (100 images) were used in the network for training, testing and validating, respectively. The accuracy of classification using color, morphological, and combination of color and morphological features were respectively, 69%, 71% and 69%. The evaluation performance of the above samples (Accuracy, Specificity and Sensitivity) was respectively, 70.2%, 0% and 100%. The best validation was found at epoch 45 and its best validation performance was 3.0739e-06.