COMPARISON OF CONVOLUTIONAL NEURAL NETWORK ARCHITECTURE FOR CLASSIFICATION OF COFFEE BEAN SPECIES BASED ON DAMPIT COFFEE BEAN VARIATIONS

One of the differences between types of coffee is the visual appearance of the coffee beans, but it takes long experience and extra precision to be able to distinguish from visual appearance. In this study, a deep learning architecture was developed for image classification based on the Convolutional Neural Network (CNN). The system was developed to distinguish the types of coffee in the Dampit area, Malang Regency. The first dataset for Dampit coffee beans consists of four classes, namely Kopi Lanang, Robusta Wine, Lanang Peaberry, and Arabica Semeru. The classification system that was built on the basis of a convolution layer and a classification layer based on Artificial Neural Networks (ANN) but experienced overfitting. So the solution is to use the GLCM method. The Kopi Dampit classification system can be carried out with application accuracy performance reaching 60% using the GLCM and ANN algorithms. In this study, the CNN method has not been able to optimize training data for making the Dampit's Coffee Bean classification system. The GLCM method can be a problem solution for classification cases with a minimum amount of data and pixel size. The computational load and accuracy are obtained so that the overfitting problem can be solved by the GLCM method.