Korkmaz Adem, Talan Tarık, Koşunalp Selahattin, Iliev Teodor
Department of Computer Technologies, Bandırma Onyedi Eylül University, Bandırma, Turkey.
Department of Computer Engineering, Gaziantep Islam Science and Technology University, Gaziantep, Turkey.
PeerJ Comput Sci. 2025 Apr 7;11:e2759. doi: 10.7717/peerj-cs.2759. eCollection 2025.
As one of the most widely consumed beverages worldwide, coffee is characterized by its diverse flavor profiles and complex production processes. In this study, deep learning-based image processing techniques are employed for the automatic classification of coffee bean species with high accuracy. To achieve this, images of three different coffee bean species (Starbucks Pike Place, Espresso, and Kenya) were classified using five CNN-based models: Xception, DenseNet201, InceptionV3, InceptionResNetV2, and DenseNet121. The dataset comprises 1,554 coffee bean images. Cross-validation was applied to assess the models' performance, and classification accuracy was evaluated using performance metrics. Among the tested models, InceptionV3 achieved the highest classification accuracy (93%) and precision (95%), with the lowest loss rate (0.12), making it the most effective model in this study. As a result of the experiments, the average classification success rates of the models were determined as follows: 93% for InceptionV3, 92% for DenseNet121, 91% for Xception, 91% for InceptionResNetV2, and 90% for DenseNet201. These findings indicate that InceptionV3 demonstrates the highest performance. It is anticipated that this study will make significant contributions to applications in coffee bean classification.
作为全球消费最为广泛的饮品之一，咖啡具有多样的风味特征和复杂的生产工艺。在本研究中，基于深度学习的图像处理技术被用于高精度自动分类咖啡豆品种。为此，使用了五种基于卷积神经网络（CNN）的模型对三种不同咖啡豆品种（星巴克派克市场、意式浓缩和肯尼亚）的图像进行分类：Xception、DenseNet201、InceptionV3、InceptionResNetV2和DenseNet121。数据集包含1554张咖啡豆图像。采用交叉验证来评估模型性能，并使用性能指标评估分类准确率。在所测试的模型中，InceptionV3达到了最高的分类准确率（93%）和精确率（95%），损失率最低（0.12），使其成为本研究中最有效的模型。实验结果表明，各模型的平均分类成功率如下：InceptionV3为93%，DenseNet121为92%，Xception为91%，InceptionResNetV2为91%，DenseNet201为90%。这些发现表明InceptionV3表现出最高性能。预计本研究将对咖啡豆分类应用做出重大贡献。
