Automated Segmentation and Counting of different varieties of Mango Trees in UAV-videos

Accurate segmentation and counting of different mango tree varieties of UAV-captured videos is critical for variety-specific yield estimation, resource allocation, and targeted orchard management. This supports sustainable practices, enhances economic outcomes, and facilitates data-driven decision-making in agriculture. To the best of our knowledge, no prior study has focused on the segmentation and tracking of multiple mango tree varieties in UAV-captured videos, making this work a novel contribution to the field. This study introduces an advanced system for segmentation, detection, tracking, and yield estimation of mango trees in unstructured orchard environments. Leveraging UAV video data of three mango varieties—“Rumani,” “Banganapalle,” and “javeri”, the system employs the YOLOv8 segmentation model to achieve precise segmented region of the tree. For robust tracking, a ORB (Oriented FAST and Rotated BRIEF) feature detector addresses challenges like overlapping canopies and variable lighting. The proposed approach delivers accurate tree counts, achieving a Mean Average Precision (mAP) of 97.9% for segmentation and a tracking accuracy of 82.9%, demonstrating its effectiveness in unstructured agricultural settings.