Analisis Penerapan Deep Learning CNN untuk Kepentingan Klasifikasi Tingkat Roasting Biji Kopi dengan Citra Digital

Abstract

Indonesia is an agrarian country with a significant population working in agriculture, notably coffee cultivation. Coffee quality heavily depends on the roasting level, making the classification of roasting levels vital. This study uses Convolutional Neural Networks (CNN) to classify coffee bean roasting levels into light, medium, and dark. CNN involves feature extraction and classification stages. Images of roasted coffee beans undergo processing through convolutional, activation (ReLU), and pooling layers for feature extraction, followed by fully connected layers for classification. The final classification uses the softmax function for optimization. The model’s performance is evaluated using different data schemes and epoch numbers. The best performance, with 100% accuracy, recall, and precision, was achieved with an 80:20 data scheme and 50 epochs. A 80:20 scheme with 30 epochs also performed well, showing high accuracy and consistency. Accuracy reflects epoch precision, precision indicates misclassification rates, and recall measures classification accuracy of individual classes. Higher values in these metrics indicate fewer errors and better model performance.