Semi-Supervised Learning with Attention-Based CNN for Classification of Coffee Beans Defect

Abstract

As the global demand for coffee rises, coffee has become a part of the daily lives of many. The taste of the brewed coffee is closely related to the quality of coffee beans, which has led to many researchers developing automated methods to accurately distinguish good coffee beans from bad ones. The research often used supervised learning technology by utilizing large sets of labeled data for training, but the labeling requires a substantial amount of manpower that is impractical for real production line usage. To solve this problem, we proposed a method that the combines semi-supervised learning and attention mechanism to classify the two types of coffee beans. Through explainable consistency training and directional attention algorithm, we solve the high-cost problem of labeling data and strengthen the prediction ability of the model. The experimental results show that the study has high classification performance and can achieve an F1-score of 97.21%.