Classification Of Jimma Coffee Beans Using Morphological and Color Features

Abstract

Image processing techniques have become increasingly useful in the vegetable industry, especially for applications in quality inspection and evaluation of food grains, fruits, vegetables, processed foods, and defect sorting. In this research work, we focused on the classification of Jimma coffee beans using morphological features (Limmu, Abba Buna, Yachi, and Buno wash), color, and morphological features extracted from 600 images. For each category, 150 images were captured. We used Artificial Neural Networks (ANN) to classify Jimma coffee beans based on the extracted features. Specifically, 15 color features and 8 morphological features, totaling 23 predominant features, were extracted from images of the four classes of coffee beans. We implemented a three-classification setup using color, morphology, and a combination of color and morphological datasets as inputs to the network. For the dataset considered, we allocated 70% (420 images) for training, 20% (120 images) for testing, and 10% (60 images) for validation. The accuracy of classification using color, morphology, and their combination were 74.8%, 75.0%, and 74.8%, respectively. The performance of the network was optimal for all classifications