Prediction of Defect Coffee Beans Using CNN
Jiyoon Lee, Young-Seob Jeong
Abstract
The growing demand for coffee has led to the development to the coffee industry. Since defect coffee beans affect the taste of coffee, it is essential to select them to improve the quality of coffee. This is basically a classification task that predicts appropriate types of defect in given coffee beans. When a person is working manually for this classification, it can be affected by human condition and has the disadvantage of taking a long time. There have been few studies that utilized data-driven method to predict defect coffee beans by analyzing images, and they commonly used convolutional neural network (CNN) model as it has shown its superior performance and efficient in image classification area. This paper proposes a method to predict defects of coffee beans by applying the CNN model to the images, so we basically solve a problem of binary image classification. We achieved the accuracy of 90.44%, and we believe our model can be used systematically and efficiently in the coffee industry.