Identifikasi Kualitas Biji Kopi Hijau Menggunakan CNN- VGG19 dan Random Forest untuk Meningkatkan Akurasi Sortir Pra-Produksi pada Industri Kopi

Abstract

The quality of green coffee beans is a fundamental factor in determining the final product quality, which directly influences the selling price and market preference. The current manual sorting process is prone to inconsistencies due to the subjectivity of the assessors. This study aims to develop an automated system for identifying the quality of green coffee beans using deep learning and machine learning approaches. The CNN-VGG19 architecture is used to learn visual patterns in coffee bean images, which are then utilized to identify and classify the beans into four quality categories: premium, longberry, peaberry, and bad beans. These patterns are learned by the model from training data without human intervention in selecting relevant features. After the features are automatically extracted by CNN-VGG19, the Random Forest algorithm is employed to classify the coffee bean types based on the resulting feature vectors. The dataset consists of a combination of primary data collected directly from Guru Patimpus Coffee Shop by the researcher and secondary data from public repositories to improve model robustness. The system is implemented as an interactive web-based application, enabling real-time identification through device camera input. The testing results demonstrate that this hybrid approach can classify the quality of coffee beans with a validation accuracy of up to 95%, proving to be both efficient and consistent under various environmental conditions.