<![CDATA[Evaluating coffee beans through a computer vision system (CVs) requires a large number of visual attributes to be extracted, but may affect prediction accuracy. Therefore, it is essential to reduce the large features to gain better prediction accuracy by generating new data that represents the most informative dimensions of the original data. Previous studies are limited to comparing different methods of feature extraction. The objective of this research was to explore the comparison of six feature extraction methods (PCA, EFA, LDA, SVD, ICA, and PLS) combined with support vector machine (SVM) as a supervised approach to predict three groups of coffee beans, namely long-berry, normal, and peaberry, for grading issues. SVM with three kernel functions (linear, RBF, and sigmoid) was used to construct a superior classification model. Data were acquired from coffee images processed to generate shape-based features. The results show that LDA provides a better visualization in separating sample classes according to the score plot with 2 variables obtained. The combination of SVM and LDA has a better recognition of coffee beans for grading, which is higher than that of other combinations. A combination of SVM-sigmoid with EFA gave mostly the worst recognition. Our findings proved that the investigation of feature extraction methods and SVM successfully achieve accurate results on grading coffee beans.]]>
Copyrights © 2026
