CLASSIFICATION OF COFFE FRUIT DRYING USING VGG16

Abstract

The drying process is a crucial stage in coffee post-harvest handling that directly affects the final product quality, especially in the specialty coffee segment. Assessment of the coffee fruit drying level in the field is still largely carried out visually and subjectively, which can potentially lead to inconsistent quality. This study aims to develop an automatic classification system for coffee fruit drying levels based on digital images using a deep learning method with the Convolutional Neural Network (CNN) VGG16 architecture. The dataset used consists of 561 coffee fruit images classified into three classes: Wet, Medium, and Dry. The preprocesssing stages include background removal, auto-cropping, and image standardization. Two models were developed: a baseline model without data augmentation and a model with data augmentation and selective fine-tuning on the final layers of VGG16. The evaluation results show that the baseline model achieved a validation accuracy of 83%, while the model with augmentation and fine-tuning improved the accuracy to 94%, accompanied by significant increases in precision, recall, and F1-score values. The proposed model also demonstrates a high and stable level of prediction confidence. These results prove that the VGG16 approach is effective for classifying coffee fruit drying levels and has the potential to be applied as an objective post-harvest quality control support system.

GS Cited Analysis