High-throughput olive germplasm classification using morphological phenotyping and machine learning.

Abstract

This study presents a robust framework for high-throughput olive germplasm classification, addressing the phenotyping bottleneck that currently limits breeding programs. Unlike prior research restricted to narrow genotypic ranges or single-image modalities, we analyzed 65 genetically diverse olive cultivars from the Tarom Olive Research Station (Zanjan, Iran). We employed a dual-image phenotyping approach, integrating high-resolution imagery of both fruits and kernels with quantitative weight metrics. This methodology enabled the extraction of critical morphological traits—including eccentricity, solidity, and shape factors—to train and validate seven Machine Learning (ML) algorithms. Our comparative analysis of Discriminant Analysis (DA), Support Vector Machine (SVM), Neural Networks (NN), and ensemble methods reveals that the DA model achieves superior performance, attaining a recall and precision of 0.98 when integrating fruit, kernel, and weight data. This significantly outperforms standard models like KNN and Naive Bayes in this domain. These findings demonstrate that combining multi-view imaging with morphological feature extraction provides a highly accurate, cost-effective tool for managing olive genetic resources and accelerating crop improvement.

Plant phenotyping relevance

果実・核の画像から形態形質を抽出し、機械学習モデルを比較検証する高スループット植物フェノタイピング手法が研究の中心である。

Code and data availability

The paper describes olive fruit/kernel image phenotyping and ML classification, but provides no public dataset, image repository, or author code. Data availability states only that all data are included in the article, and the sole supplement (Supplementary Material 1) is described only as containing hyperparameter/Bay

No evidence-backed public reproduction asset is currently recorded.