Optimizing Pomegranate Growth Stage Detection Using YOLOv8 and Hybrid Machine Learning Approaches

Pomegranates, valued for their nutritional and antioxidant properties, require precise growth stage monitoring to optimize yield and quality. This study aims to detect pomegranate growth stages using the YOLOv8 deep learning model integrated with hybrid machine learning techniques. A dataset of 5857 images depicting five stages of pomegranate growth, including Bud, Flower, Early-Fruit, Mid-Growth, and Ripe, was pre- processed with Region of Interest (ROI) extraction, resizing, and normalization. The YOLOv8 model was fine-tuned to serve as a direct classifier and a feature extractor. Extracted features were refined using the Random Forest (RF) model and classified with Logistic Regression (LR), Support Vector Machine (SVM), and Extreme Gradient Boosting (XGB). Ensemble methods such as stacking and voting further enhanced classification accuracy. Results showed significant improvements in performance metrics, with the hybrid approach outperforming the individual YOLOv8 model. F1 Score values for YOLOv8, LR, SVM, XGB, Stacking, and Voting were 90.0%, 93.1%, 93.4%, 93.8%, 94.7<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup>, and 95.0%, respectively. The voting technique achieved the highest F1-Score, showing a 5% improvement, demonstrating that combining YOLOv8 with machine learning models enhances accuracy and robustness. This research provides a scalable solution for agricultural monitoring, offering valuable insights into integrating deep learning and machine learning for effective crop management.