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, Click here for options.

Phenotyping of Pine Tree Architecture with Stereo Vision and Deep Learning

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

Citation:  2021 ASABE Annual International Virtual Meeting  2100847.(doi:10.13031/aim.202100847)
Authors:   Mousumi Akter, Nariman Niknejad, Yin Bao, Rafael Bidese Puhl, Kitt Payn, Jingyi Zheng
Keywords:   Deep Learning, Instance Segmentation, Loblolly Pine, Phenotyping, Stereo Matching, Tree Architecture

Abstract. Loblolly pine is one of the most important forest tree species in the southern U.S. for saw timber production. Its trunk and branch characteristics have a significant impact on yield potential. However, commercially important traits such as trunk straightness, branch angle, and branch diameter are currently quantified by visual grading, which is subjective and low throughput. In this study, the utility of combining stereo 3D imaging and deep learning was evaluated for pine architecture phenotyping. Stereo RGB images and manual measurements of branch diameters, branch angles, and trunk diameters were collected from individual loblolly pine trees of different families in a progeny test. A custom MS-COCO dataset was created by annotating contour polygons of trunk and branches in the images. The dataset was used to fine-tune and test a Mask R-CNN model for the instance segmentation task. Dense and patch-based stereo matching algorithms were used to reconstruct large trunks and thin branches, respectively. The resultant 3D point clouds were further processed to extract branch angle using principal component analysis (PCA), and trunk diameter. High correlation was found for the image-derived branch angle (R2 = 0.66) and trunk diameter (R2 = 0.78). The proposed system showed promising potential as a high-throughput precision phenotyping tool for tree architecture characterization of loblolly pine, facilitating the selection of tree architecture that is highly productive and resilient to climate variability and associated severe weather events.

(Download PDF)    (Export to EndNotes)