Coffee bean quality analysis using convolutional neural networks

Coffee grading is the main procedure in its production. One of the processes to grade coffee is done manually, requiring a lot of training and experience from experts. The main objective of this work is to use technologies based on Artificial Intelligence, with Convolutional Neural Network models, and together with the application of image processing techniques, to improve the quality analysis of coffee beans. We performed a comparison of pre-trained models, namely AlexNet and DenseNet, using a coffee bean dataset. A total of 4272 coffee bean images from the USK-COFFEE database were used in this work. The model training, testing and validation processes were performed with an 80/10/10 division of the obtained images. Classification metrics such as Recall, F-1 Score, were used for the detailed analysis of the performance models. ROC curves were used to analyze their distinction.