Classification of coffee bean images

The variation in green coffee beans collected from coffee trees significantly impacts coffee trade, as it affects flavor consistency and quality standards. Traditional sorting and classification processes are typically performed manually, making them time-consuming, costly, and prone to errors. Therefore, utilizing image processing, machine learning, and deep learning techniques enables the automatic classification of coffee beans, accelerating processes and enhancing accuracy. Machine learning techniques help optimize quality control by analyzing physical characteristics such as size, color, and density. In this thesis, the USK-Coffee dataset, consisting of 8,000 samples from four different green coffee bean types, was used. Feature extraction was performed using MobileNetV2, ResNet18, VGG16, and DenseNet201 models, followed by classification with machine learning algorithms such as K-Nearest Neighbors (KNN), Naive Bayes, Decision Tree, and Support Vector Machine (SVM). To analyze classification performance variations, additional feature extraction and dimensionality reduction techniques such as Principal Component Analysis (PCA) and Histogram of Oriented Gradients (HoG) were applied, and the results were compared. Among all implementations, the highest accuracy rate of 91.99% was achieved using the DenseNet201 model combined with PCA-extracted features applied to the SVM algorithm.