Automatic Detection of Olive Tree Canopies for Groves with Thick Plant Cover on the Ground

Abstract

Marking the tree canopies is an unavoidable step in any study working with high-resolution aerial images taken by a UAV in any fruit tree crop, such as olive trees, as the extraction of pixel features from these canopies is the first step to build the models whose predictions are compared with the ground truth obtained by measurements made with other types of sensors. Marking these canopies manually is an arduous and tedious process that is replaced by automatic methods that rarely work well for groves with a thick plant cover on the ground. This paper develops a standard method for the detection of olive tree canopies from high-resolution aerial images taken by a multispectral camera, regardless of the plant cover density between canopies. The method is based on the relative spatial information between canopies.The planting pattern used by the grower is computed and extrapolated using Delaunay triangulation in order to fuse this knowledge with that previously obtained from spectral information. It is shown that the minimisation of a certain function provides an optimal fit of the parameters that define the marking of the trees, yielding promising results of 77.5% recall and 70.9% precision.

Plant phenotyping relevance

オリーブ樹冠を高解像度UAV画像から自動検出・標識する手法の開発が中心であり、植物の樹冠形態・位置を抽出する画像ベースのフェノタイピング手法に該当する。

Code and data availability

The paper's multispectral UAV captures, metadata, and labelled data used for olive tree canopy detection are not publicly available; the Data Availability Statement says they are available only on request from the corresponding author due to confidentiality. No public code, model, or dataset URL is provided.

No evidence-backed public reproduction asset is currently recorded.