Enhancing green coffee quality assessment through deep learning

Abstract

Coffee is the world’s most consumed commodity beverage, vital for the Brazilian market. Assessing coffee bean quality through visual features is essential for market value. However, human-based visual analysis has limitations. Deep neural networks, particularly CNNs, offer a promising solution by automating this process. In this work, we propose an evaluation of deep learning models and training strategies to classify green coffee beans automatically. We evaluate four CNN architectures: AlexNet, ResNet-50, MobileNet V3, and EfficientNet B4. After a hyperparameter optimization step, the models were fine-tuned, and we evaluated the impact of data augmentation strategies on the classification performance through the USK-Coffee dataset. EfficientNet B4 excels, achieving 0.8844 accuracy when trained with data augmentation. Our findings showcase deep learning’s potential for coffee quality assessment, aiding professionals in classifying and guaranteeing coffee quality and value.