A Cascading of YOLOv8 and Random Forest Regression in Oil Palm Fresh Fruit Bunch Mass Estimation System using Unmanned Aerial Vehicle Imagery

Abstract

Efficient management of oil palm farms requires accurate pre-harvest planning to maximize productivity. Traditional methods for estimating the mass of Fresh Fruit Bunches (FFBs) typically involve manual sampling and weighing, which are time-consuming and prone to errors. This study presents a novel system combining unmanned aerial vehicle (UAV) photography with geometric feature extraction using YOLOv8-Segmentation and machine learning models—Random Forest Regression (RFR)—to estimate FFB mass. The system addresses challenges posed by dynamic drone imagery, including environmental variations and frond occlusions. Instead of directly integrating YOLOv8 with the regression models, geometric features such as the minor axis, perimeter, and eccentricity are extracted from the segmented images and used to train the RFR for mass estimation. The top-performing model, using features extracted from YOLOv8-Small-Segmentation with the minor axis and eccentricity, achieved a Root Mean Square Error (RMSE) of 3.95 and a Mean Absolute Error (MAE) of 2.87 for frond-covered FFBs. For frond-uncovered FFBs, the best-performing features were the minor axis, perimeter, and area extracted using YOLOv8-Large-Segmentation, resulting in an RMSE of 3.91 and MAE of 2.91. These results demonstrate the system's capability to accurately estimate FFB mass based on UAV-captured imagery and feature extraction. This approach offers a scalable and efficient solution for pre-harvest planning in oil palm plantations, addressing the limitations of traditional methods while improving operational efficiency and accuracy in yield estimation.

Access to Document

Other files and links

• Link to publication in Scopus
• Open Access Version Available

Fingerprint