Detection of Cacao Trees in Orthomosaic Images Using YOLOv8 and SAHI

This study explores the application of drone-based orthomosaic imagery combined with the YOLOv8x segmentation model and the Slicing Aided Hyper Inference (SAHI) method for detecting cacao trees. High-resolution orthomosaic images were processed by slicing them into smaller tiles to facilitate object segmentation. Although the YOLOv8x model was trained for segmentation, the evaluation was conducted using bounding boxes for ease of comparison. The model achieved a precision of 93.49% and an accuracy of 81.10%, demonstrating its potential for monitoring large-scale plantations. However, the system is best suited for estimation purposes rather than precise, real-time decision-making.