Automated Coffee Roast Level Classification Using Machine Learning and Deep Learning Models.

Area of Science:

• Agricultural Science
• Computer Science
• Food Science

Background:

• Coffee quality is heavily influenced by the roasting process, traditionally assessed manually.
• Manual roast classification is subjective, inconsistent, and time-consuming.
• Advancements in machine learning (ML) and computer vision offer automated solutions.

Purpose of the Study:

• To evaluate multiple ML models for automated coffee roast level classification.
• To compare the performance of CNNs with traditional ML algorithms.
• To develop a reliable and scalable solution for coffee quality control.

Main Methods:

• Trained and tested ML models, including a CNN with Xception, AdaBoost, random forest (RF), and support vector machine (SVM).
• Utilized a public dataset of 1,600 images across four roast levels (green, light, medium, dark).
• Applied image augmentation techniques to enhance model generalizability.

Main Results:

• All evaluated models achieved 100% accuracy and F1-scores in classifying coffee roast levels.
• The proposed automated approach demonstrated strong performance compared to previous studies.
• Image augmentation improved the generalizability of the models.

Conclusions:

• ML and computer vision provide a highly accurate and automated method for coffee roast classification.
• This technology offers significant improvements in quality control for the coffee industry.
• The developed solution is reliable, scalable, and contributes to consistent coffee production.