Analysis of shape features by applying gain ratio and machine learning for coffee bean classification

Abstract

Coffee is one of the daily consumed beverages in many countries. It is yielded from coffee beans, which have proceeded through several processes. Several common coffee beans have been produced in Indonesia, such as Arabica, Robusta, Liberica, and Excelsa. Nevertheless, many coffee fanatics are unable to distinguish the various coffee bean types visually based on those shapes. Accordingly, it is necessary to classify the types of coffee beans. The work applied training and testing steps. Both involved ROI detection, pre-processing, segmentation, feature extraction, selection, and classification. Image processing was used in ROI detection, pre-processing, and segmentation to simplify the procedure and separate the coffee bean from the background. The feature extraction produced 14 shape features to distinguish the coffee bean’s class, but the proposed method’s performance has yet to reach the optimal result. The gain ratio was used to reduce the features; hence, only 4 features were selected, including aspect ratio, eccentricity, equivalent diameter, and area. These features were utilized as input data for classification using Naive Bayes, Artificial Neural Network (ANN), Support Vector Machine (SVM), C4.5, and decision tree. The proposed method used 4 features and a decision tree classifier. The local dataset has 400 coffee bean photos in four classes of 100 images each. The photos were divided for training and testing using k-fold 10 cross-validation. The accuracy evaluation parameter reached 0.995.