Scholarlink

<p>This work presents a spatially resolved Vis–NIR hyperspectral imaging workflow (Vis/NIR-HSI; 400–1000&#xa0;nm) offers an alternative to point-based NIR spectroscopy for mapping inter-fruit variability in mangoes critical for accurate total soluble solids (TSS, °Brix) assessment, leveraging a customised three stage PCA–Masking–Glare (PMG) pipeline built to extracts only valid fruit spectra. Eight spectral preprocessing techniques is clustered into three categories - denoising (Savitzky-Golay filtering (SGF), wavelet transform (WT)), scatter correction (multiplicative scatter correction (MSC), extended MSC (EMSC), standard normal variate (SNV)), and baseline correction (first derivative (D1), second derivative (D2), detrending (DT)) gets parameterized into <i>68</i> single-step profiles. Each profile evaluated using four machine learning regression models - Support Vector Regression (SVR), K-nearest neighbours (KNN), Random Kitchen Sinks Ridge (RKS), and Extreme Gradient Boosting (XGBoost) were trained on mean-centred spectra, with Optuna-powered stratified five-fold hyperparameter optimization. The top-ranked profiles are then paired into <i>88</i> dual-step profiles, re-assessed under the same modelling sequence and ranking strategy to isolate the best profiles. These leading dual profiles are further merged into <i>7</i> triple-step profiles, again undergoing hyperparameter optimization and performance ranking. The best triple-step profile using KNN achieves <i>R²ₚ = 93.83% and RMSE</i><sub><i>P</i></sub> <i>= 1.22 °Brix</i>, but residual‐error analysis demonstrated that a dual‐step sequence achieved the most consistent performance across both SVR and KNN. By integrating rigorous data preparation with systematic preprocessing spectral profile evaluation and workflow optimization, this approach significantly accelerates non-destructive TSS prediction in mangoes and lays the groundwork for hyperspectral quality assessment across diverse horticultural products.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Optimizing chemometric spectral preprocessing profiles for hyperspectral non-destructive prediction of mango total soluble solids

  • 期刊论文

  • 发表日期：

    2025年11月25日

查看全文
  • Kripa S,
  • Jeyalakshmi V

摘要

This work presents a spatially resolved Vis–NIR hyperspectral imaging workflow (Vis/NIR-HSI; 400–1000 nm) offers an alternative to point-based NIR spectroscopy for mapping inter-fruit variability in mangoes critical for accurate total soluble solids (TSS, °Brix) assessment, leveraging a customised three stage PCA–Masking–Glare (PMG) pipeline built to extracts only valid fruit spectra. Eight spectral preprocessing techniques is clustered into three categories - denoising (Savitzky-Golay filtering (SGF), wavelet transform (WT)), scatter correction (multiplicative scatter correction (MSC), extended MSC (EMSC), standard normal variate (SNV)), and baseline correction (first derivative (D1), second derivative (D2), detrending (DT)) gets parameterized into 68 single-step profiles. Each profile evaluated using four machine learning regression models - Support Vector Regression (SVR), K-nearest neighbours (KNN), Random Kitchen Sinks Ridge (RKS), and Extreme Gradient Boosting (XGBoost) were trained on mean-centred spectra, with Optuna-powered stratified five-fold hyperparameter optimization. The top-ranked profiles are then paired into 88 dual-step profiles, re-assessed under the same modelling sequence and ranking strategy to isolate the best profiles. These leading dual profiles are further merged into 7 triple-step profiles, again undergoing hyperparameter optimization and performance ranking. The best triple-step profile using KNN achieves R²ₚ = 93.83% and RMSEP = 1.22 °Brix, but residual‐error analysis demonstrated that a dual‐step sequence achieved the most consistent performance across both SVR and KNN. By integrating rigorous data preparation with systematic preprocessing spectral profile evaluation and workflow optimization, this approach significantly accelerates non-destructive TSS prediction in mangoes and lays the groundwork for hyperspectral quality assessment across diverse horticultural products.