Learning-Based versus Geometric Three-Dimensional Point Cloud Analysis in Precision Viticulture: A Systematic Review of Deployment Evidence
Smart Agricultural Technology · 2026-11 · 102714
DOI: 10.1016/j.atech.2026.102714 ↗Kurz gefasst
Systematische Übersicht in der Zeitschrift Smart Agricultural Technology, 2026-11. Verfasst von Dimitrios Tilemachos Davarakis und 1 weiteren Autor:innen. Die Arbeit nennt 186 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0).
Abstract (Originalsprache)
Three-dimensional point cloud methods are central to precision viticulture, measuring vineyard structures from individual berries to entire blocks. This review synthesises 183 articles retrieved from Scopus (2018–April 2026): 167 primary studies (54 learning-based (LB) and 113 non-learning-based (NLB)) classified across seven task categories spanning organ detection, yield estimation, robotic perception, canopy management and phenotyping, disease assessment and vineyard mapping, plus 16 supporting works comprising 3D vineyard datasets, synthetic-data generators, and prior reviews. Both streams are analysed in parallel and profiled against the evidence each reports for deployment. No included study documents a system in operational use, and the field sits predominantly at evaluation in a working vineyard; evaluation beyond the development dataset was recorded for 14 of 54 learning-based and 24 of 113 non-learning-based studies. Organ detection and segmentation is the one category in which the two streams separate on the evidence recorded, in favour of learning-based methods. Five architecture families structure the LB corpus: 72% of LB articles follow a 2D-to-3D fusion paradigm, and one-sixth use native 3D architectures operating directly on point clouds. Three challenges constrain progress: dataset scarcity (only five publicly available 3D vineyard datasets; 70% of LB articles rely on private training sets); benchmarking fragmentation (54 distinct primary metrics prevent cross-study comparison); and limited generalisation across varieties, growing seasons, and vineyard sites. Priorities include a shared annotated 3D vineyard benchmark, native 3D architectures, cross-variety and cross-geography evaluation, and a shift to agronomic outcome measures.
Zitieren
APA
Davarakis, D. T., & Argyropoulos, D. (2026). Learning-Based versus Geometric Three-Dimensional Point Cloud Analysis in Precision Viticulture: A Systematic Review of Deployment Evidence. Smart Agricultural Technology, 102714. https://doi.org/10.1016/j.atech.2026.102714
Vancouver
Davarakis DT, Argyropoulos D. Learning-Based versus Geometric Three-Dimensional Point Cloud Analysis in Precision Viticulture: A Systematic Review of Deployment Evidence. Smart Agricultural Technology. 2026:102714. doi:10.1016/j.atech.2026.102714
BibTeX
@article{Davarakis2026,
author = {Davarakis, Dimitrios Tilemachos and Argyropoulos, Dimitrios},
title = {Learning-Based versus Geometric Three-Dimensional Point Cloud Analysis in Precision Viticulture: A Systematic Review of Deployment Evidence},
journal = {Smart Agricultural Technology},
year = {2026},
pages = {102714},
doi = {10.1016/j.atech.2026.102714}
}
EUVIMED