Automatic recognition of coffee bean varieties based on pre-trained architectures

Abstract

Coffee is an agricultural commodity of fundamental and considerable economic importance on the global market. In this study, the coffee bean varieties were examined from images via artificial intelligence due to their quality and value on the market. This study aims to create an automated system that can efficiently identify coffee beans without requiring a significant amount of time. In this study, five pre-trained Convolutional Neural Network (CNN) architectures were performed to detect four varieties of coffee beans through images. Extracting features from images is a challenging and specialized task. However, CNN possesses the ability to extract features automatically. Therefore, these architectures were employed as both deep feature extractors and classifiers. Primarily, 1600 coffee beans' images were split into 75:25 training and testing sets. Next, 5-fold cross-validation was applied during the training process. This study presented both validation and testing results. Eventually, ShuffleNet achieved the best classification performance with 99.33% and 99.75% accuracy rates in identifying types of coffee beans for the training and testing sets, respectively. As a result, this study has demonstrated that deep learning technologies can automatically recognize the different types of coffee beans.