<![CDATA[Background: Post-harvest ripening for Hass avocados is hard because their ripening pattern is unpredictable. Recent studies have explored deep learning applications for ripening stages classification. However, it needs further development to achieve better results that can run effectively on low-end devices like smartphones. Objective: The study aims to achieve three main goals that involve developing an efficient YOLOv8 object detection model that works with local avocado varieties to precisely determine ripening stages, creating a hybrid machine learning model to improve classification performance, and predicting storage duration (shelf life). The study also aims to develop a practical mobile application for post-harvest use. Methods: The YOLOv8 model was trained first on the Hass avocado dataset for transfer learning, followed by a second training on the local avocado dataset for domain adaptation. Mean Average Precision (mAP) was used to evaluate a standalone YOLOv8 object detection model, while accuracy and F1-score were used to evaluate classification. The hybrid YOLOv8 and machine learning approach includes selecting the optimal YOLOv8 layer for feature extraction, using Random Forest for feature selection, applying SMOTE to handle data imbalance, and classifying with Logistic Regression, SVM, and XGBoost. Storage duration was performed using a standalone YOLOv8 and hybrid ML classification results, along with a linear regression formula. Results: The standalone YOLOv8 model achieved an mAP50 of 0.93 and a classification accuracy of 0.90 on the local avocado dataset. The hybrid method achieved a classification accuracy of 0.96. The storage estimation results showed that the hybrid approach produced an MAE of 0.43 days, while the standalone approach produced 0.44 days. The results were better than previous studies, achieving an error rate of 0.96 days. Conclusion: The research achieved its goal of developing an improved approach to identify avocados and determine their storage duration. The hybrid model functions as a working solution that improves post-harvest operations using Android-based applications. Keywords: YOLOv8, Machine Learning, Avocado Ripening, Shelf-life Estimation]]>
Copyrights © 2026
