Classification of Arabica Coffee Beans Based on Multi-Features Using Artificial Neural Networks

Indonesian Arabica coffee beans undergo meticulous cultivation by local farmers and are subjected to various processing methods, including wet-hulled, honey, natural, and washed. Indonesia has cultivated and produced the most renowned Arabica coffees in six distinct regions. The several types of arabica coffee beans, mostly found in Indonesia, have similar sizes and shapes, making them hard to distinguish. Naturally, aroma, flavor, sweetness, roasting, and other factors vary when ingested. Thus, computer vision technology must classify Arabica coffee beans to improve coffee bean identification without consuming them first. The proposed method consists of several main processes: data acquisition, ROI detection, pre-processing, segmentation, feature extraction, and classification. Feature extraction is applied based on color, shape, and texture features, followed by implementing several machine-learning methods. The dataset was divided into training and testing sets using cross-validation with K-Fold values of 5 and 10. Performance evaluation was performed using three parameters: precision, recall, and accuracy. The Artificial Neural Network (ANN) method achieved the best results, achieving 99.75% accuracy on K-Fold 5 & 10.