Identification of acidic tolerance in rapeseed varieties based on hyperspectral imaging

Abstract

With increasingly severe soil acidification, it is essential to screen and identify acid-tolerant crop varieties to safeguard agricultural production. Integration of hyperspectral imaging with machine learning models has been extensively used in high-throughput crop phenotyping. Here, we established a multi-indicator evaluation system for quantification of acidic tolerance in rapeseed based on hyperspectral data collected from 65 rapeseed varieties under pH = 5.2 (acidic) and pH = 6.4 (normal) soil conditions. After denoising and smoothing, 34 existing vegetation indices and band combination indices were derived from which eight growth-sensitive indices were selected based on their correlations with the actual growth scores. Six key spectral bands exhibiting significant changes under acidic stress were identified, with the feature importance outputs from three machine learning classification models. Comprehensive sensitivity coefficients (SC) were derived by integrating growth-sensitive indices and key band data using principal component analysis weighted-sum (PCA-WS). Hierarchical clustering classified the 65 tested varieties into strongly-tolerant (four varieties), moderately-tolerant (26 varieties), and weakly-tolerant (35 varieties). Physiological validation based on yield and branch number showed that the strongly-tolerant varieties had 10.03 % and 17.15 % higher relative yields and 13.24 % and 11.14 % higher relative branch numbers than moderately and weakly-tolerant varieties, respectively. These results have established a hyperspectral evaluation system that can accurately evaluate the acidic tolerance of rapeseed, providing a reliable basis for screening acid-tolerant rapeseed varieties.

Plant phenotyping relevance

ラペシードの酸性耐性をハイパースペクトル画像から定量評価する評価システムを構築し、特徴量選択・機械学習・検証まで行っており、植物表現型取得・抽出手法が中心である。

Code and data availability

公開論文であることは確認できましたが、現在の公式API・許可済み取得経路では本文を自動取得できませんでした。

No evidence-backed public reproduction asset is currently recorded.