Hyperspectral Imaging for Predicting the Internal Quality of Kiwifruits Based on Variable Selection Algorithms and Chemometric Models

Hongyan Zhu,Bingquan Chu,Xiaoya Tao,Yangyang Fan,Wenxin Yin,Yong He

doi:10.1038/s41598-017-08509-6

Abstract

We investigated the feasibility and potentiality of determining firmness, soluble solids content (SSC), and pH in kiwifruits using hyperspectral imaging, combined with variable selection methods and calibration models. The images were acquired by a push-broom hyperspectral reflectance imaging system covering two spectral ranges. Weighted regression coefficients (BW), successive projections algorithm (SPA) and genetic algorithm–partial least square (GAPLS) were compared and evaluated for the selection of effective wavelengths. Moreover, multiple linear regression (MLR), partial least squares regression and least squares support vector machine (LS-SVM) were developed to predict quality attributes quantitatively using effective wavelengths. The established models, particularly SPA-MLR, SPA-LS-SVM and GAPLS-LS-SVM, performed well. The SPA-MLR models for firmness (Rpre = 0.9812, RPD = 5.17) and SSC (Rpre = 0.9523, RPD = 3.26) at 380–1023 nm showed excellent performance, whereas GAPLS-LS-SVM was the optimal model at 874–1734 nm for predicting pH (Rpre = 0.9070, RPD = 2.60). Image processing algorithms were developed to transfer the predictive model in every pixel to generate prediction maps that visualize the spatial distribution of firmness and SSC. Hence, the results clearly demonstrated that hyperspectral imaging has the potential as a fast and non-invasive method to predict the quality attributes of kiwifruits.

Highlights

The conventional methods for quality measurements using standard instrumental methods include a Magness-Taylor penetrometer or texture analyser[8] for flesh firmness, a digital hand-held pocket refractometer for solids content (SSC), and a pH-meter for pH
We have demonstrated the feasibility and usefulness of hyperspectral imaging in the Vis/NIR spectral region (380–1023 nm) and the NIR spectral window (874–1734 nm) for the rapid prediction of quality parameters and mapping the spatial distributions of SSC and firmness in kiwifruits
This work demonstrated that (1) the spectral profiles had four broadband absorption regions of approximately 673, 970, 1200, and 1460 nm, in addition to the small absorption region at 835 nm, which were mainly due to the pigments, carbohydrates and water in kiwifruits; (2) firmness and SSC were better predicted in the region of 450–1000 nm, and the improved pH results were obtained using the range of 951–1670 nm by means of the full-spectral wavelengths with the partial least squares regression (PLSR) model; (3) successive projections algorithm (SPA) and GA-PLS were more powerful than BW, which was the most common method used in previous studies[28, 29] for the selection of effective wavelengths (EWs); and (4) the linear and nonlinear calibration models were established for spectral analysis using the EWs

Summary

Introduction

The conventional methods for quality measurements using standard instrumental methods include a Magness-Taylor penetrometer or texture analyser[8] for flesh firmness, a digital hand-held pocket refractometer for SSC, and a pH-meter for pH. We sought to explore the feasibility and potentiality of determining firmness, SSC, and pH in kiwifruits based on HSI This utmost goal was achieved by meeting the following specific objectives: (i) establishing a hyperspectral imaging system and choosing a better spectral range for determining the quality attributes by means of full-spectral wavelengths with partial least squares regression (PLSR) model; (ii) comparing and evaluating the superior variable selection method from the weighted regression coefficients (BW), successive projections algorithm (SPA) and genetic algorithm–partial least square (GAPLS) and determining the corresponding optimal wavelengths, which give the highest correlation between the spectral data and the three quality parameters; (iii) developing robust and accurate calibration models [PLSR, multiple linear regression (MLR), least squares support vector machine (LS-SVM)] to quantitatively predict the flesh firmness, SSC, and pH using spectral responses from only the optimal wavelengths; and (iv) applying the optimal model to predict the quality attributes of each pixel in samples and generate spatial distribution maps for the whole kiwifruit

Methods

Results

Conclusion