Influence of tomato storage period on the generalization of a near-infrared spectroscopy-based brix prediction mode

Abstract

To mitigate the impact of storage period variations on fruit sugar content prediction models and further enhance the universality of sorting models, this study investigated the influence of different storage periods on tomato brix prediction using near-infrared (NIR) spectroscopy and established a partial least squares (PLS) brix content prediction model. Experiments revealed that when the storage periods of the calibration set and the prediction set differed, the predictive performance of the PLS model significantly declined. To address this issue, the study found that optimizing spectral data with standard normal variate (SNV) transformation and adopting a mixed-modeling strategy incorporating multiple storage periods substantially improved the accuracy of the universal model: the correlation coefficient of the prediction set (Rp) increased from 0.803 to 0.934, the root mean square error of prediction (RMSEP) decreased from 0.476 to 0.375, and the residual predictive deviation (RPD) rose from 2.11 to 3.26. Finally, the competitive adaptive reweighted sampling (CARS) algorithm was employed to screen key wavelengths, effectively reducing data dimensionality while minimizing interference from storage period differences. Compared with the successive projections algorithm (SPA), the CARS method demonstrated superior performance, ultimately establishing a highly robust universal prediction model for tomato brix.

Plant phenotyping relevance

トマト果実の糖度（Brix）をNIR分光で推定する予測モデルを開発・比較・検証しており、表現型取得手法が研究の中心である。

Code and data availability

公開状態または取得可能な本文経路を確認できませんでした。

No evidence-backed public reproduction asset is currently recorded.