<![CDATA[Freshness assessment of tuna (Auxis thazard) in traditional markets still relies mainly on subjective visual inspection, which introduces inconsistency and food-safety risk. This study proposes an objective, low-cost classification pipeline for tuna freshness based on eye images. Six hundred forty fish-eye images were acquired at Kedonganan Fish Market and labeled through a 30-panelist organoleptic test based on SNI 2729:2013. The pipeline segments the eye Region of Interest using U-Net, extracts HSV color features from the segmented eye, reduces dimensionality using Principal Component Analysis (PCA), and classifies freshness with a Support Vector Machine (SVM). A Group K-Fold (k=8) validation scheme and per-fold standardization are used to prevent data leakage across acquisition groups. Grid search over target cumulative variance (50%-95%) and SVM regularization (C=0.1, 1, 10) yields a best configuration at cumulative variance of 55% and C=1, achieving 96.72% accuracy, 97.76% precision, 96.72% recall, and 96.51% F1-score. Compared with SVM without PCA (95.47% accuracy, 215.82 s), the PCA-SVM model reaches equivalent or higher accuracy with 47% lower classification time, supporting deployment on resource-constrained devices.]]>
Copyrights © 2026
