Comparative Performance Analysis of CNN Architectures for Coffee Bean Image Classification Based on Accuracy and Training Time

Coffee bean image classification is an application of computer vision that can support the automated identification and quality categorization of coffee beans. This study aims to compare the performance of several Convolutional Neural Network (CNN) models, namely MobileNetV2, ResNet50, EfficientNetB0, VGG16, and DenseNet121, for coffee bean image classification. The dataset consists of coffee bean images that underwent preprocessing and data splitting for model training and testing. Model performance was evaluated based on classification accuracy and training time. The results show that MobileNetV2 achieved the best performance, with an accuracy of 98.75% and the shortest training time of 97.69 seconds. DenseNet121 also achieved high accuracy of 98.25%, with a training time of 230.96 seconds. Meanwhile, VGG16 achieved an accuracy of 89.00% but required the longest training time of 491.79 seconds. ResNet50 and EfficientNetB0 demonstrated comparatively lower performance, with accuracies of 63.75% and 25.00%, respectively. Based on these findings, MobileNetV2 can be recommended as the most effective model for coffee bean image classification, as it provides a favorable balance between high classification accuracy and training efficiency.