Prototype Development of Sorting System for Coffee Cherries Using Computer Vision

Abstract

Thailand's coffee market has experienced remarkable growth, with a consistent annual increase of 7.3% from 2017 to 2023, reaching a market valuation of 34 million Baht. Given this growth, assuring consistent coffee quality has become increasingly critical, particularly at the initial stage of coffee cherry selection. The quality of coffee cherries directly impacts the final coffee bean quality, making efficient and accurate sorting essential for premium coffee production. This research explores an innovative approach to coffee cherry classification using image processing techniques. We developed and compared four machine learning models: Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Artificial Neural Network (ANN), and Convolutional Neural Network (CNN). Each model was evaluated based on classification accuracy and processing speed. The experimental results revealed that the CNN model significantly outperformed other approaches, achieving an overall operational accuracy of 81.73% and a throughput of approximately 120 cherries per minute (about 2 kg per hour). These findings suggest that automated visual inspection using CNN-based methods can substantially improve the efficiency and consistency of coffee cherry sorting, demonstrating strong potential for practical implementation in small- to medium-sized specialty coffee community enterprises.