ASABE Technical Library - Abstract

Click on “Download PDF” for the PDF version or on the title for the HTML version.
If you are not an ASABE member or if your employer has not arranged for access to the full-text, view access and subscription options

From Image to Pruning Decision: Segmentation and 3D Reconstruction of Cacao Tree Branches on a Mobile Decision Support System (DSS) for Pruning Branch and Points Detection and Recommendation in Cacao Orchards

Published by the American Society of Agricultural and Biological Engineers, St. Joseph, Michigan www.asabe.org

Citation:  2026 ASABE Annual International Meeting  2600173.(doi:10.13031/aim.202600173)
Authors:   Samuel Azubuine, Magni Hussain
Keywords:   Branch Segmentation, Cacao, Cocoa, Decision Support Systems, Instance and Semantic Segmentation, Photogrammetry, Pruning, RealityKit, Three-Dimensional(3D) Reconstruction, YOLO models.

Abstract. Cacao (Theobroma Cacao) is a key global agro-commodity, and pruning constitutes a crucial agronomic and post-planting activity for cacao production, tree rejuvenation, and overall orchard health management. Accurate structural representation of the cacao tree is essential for systematic pruning branch and point detection. The segmentation and three-dimensional (3D) reconstruction of tree branches enable detailed analysis of canopy geometry, branch morphology, and spatial arrangement; however, cacao‘s evergreen nature, dense foliage, cauliflorous growth habit, and variable tropical illumination make in-field branch detection particularly challenging. This paper presents the vision module of a mobile decision support system (DSS); Prunify, for cacao pruning branch and point detection. We introduce a pipeline that segments cacao tree branches from input images and reprojects the masked branch pixels into 3D space using depth maps acquired directly from the iPhone‘s integrated ARKit depth framework. The pipeline operates in two modes: a mask-guided Photogrammetry Mode for branch-specific reconstruction, and an Area Mode for full-scene capture and reconstruction. Evaluation of selected segmentation algorithms for pipeline integration under 5-fold cross-validation with progressive image resizing showed DeepLabV3+ achieved 89% mIoU, while YOLOv11n provided the optimal speed-accuracy trade-off for real-time on-device inference. Depth-guided reconstruction on laboratory and greenhouse cacao specimens produced geometrically consistent branch meshes under adequate multi-view coverage (K ≥ 6–20 frames or more, 30–40° angular spacing in less than a min and more as K grows). Future work will integrate and implement a skeletal and branch traversal algorithm for pruning branch and points recommendation and full in-field decision support integration on Prunify for in-field testing/comparisons and subsequently for an end-effector/robotic pruner

(Download PDF)    (Export to EndNotes)