Comparison of Linear and Radial Basis Function Kernels in Support Vector Machine for Rice Variety Classification

Rice variety classification plays a pivotal role in quality control and market segmentation, particularly in regions where rice is a dietary staple. This study compares the performance of Support Vector Machine (SVM) classifiers utilizing linear and Radial Basis Function (RBF) kernels, integrated with Principal Component Analysis (PCA) for dimensionality reduction. A dataset comprising <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{7 5, 0 0 0}$</tex> rice grain images representing five varieties-Arborio, Basmati, Ipsala, Jasmine, and Karacadag-was employed. PCA was applied to the extracted RGB histogram features, reducing dimensionality while preserving 90 % of the dataset's variance. The RBF kernel achieved a higher accuracy of 84.80 %, outperforming the linear kernel's <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{8 3. 2 8 \%}$</tex>, particularly in cases with overlapping features, such as Basmati and Jasmine. Despite the linear kernel's superior computational efficiency, the RBF kernel excelled in capturing non-linear patterns. These results underline the trade-offs between computational demands and classification precision. This study recommends the linear kernel for resource-constrained applications requiring real-time processing and the RBF kernel for tasks demanding higher accuracy. Future research will explore advanced feature extraction techniques and kernel optimizations to enhance classification performance.