Colour detection method of Korla fragrant pear based on dielectric spectroscopy technology

TL;DR

This paper introduces a new method to detect the color of Korla fragrant pears using dielectric spectroscopy, which could improve quality control and commercial value.

Contribution

A novel color prediction method for Korla fragrant pears using dielectric spectroscopy and feature variable extraction techniques is proposed.

Findings

UVE and SPA feature extraction improved color prediction accuracy of fragrant pears.

SPA-PLSR model achieved best prediction for L* color parameter with R2 = 0.83.

UVE-PLSR model performed best for a* and b* color parameters.

Abstract

Accurate control of fruit quality determines the commercial value of Korla fragrant pear. The rapid and accurate detection of the colour of fragrant pear is crucial for improving its commercial value.

In this study, a vector network analyser and coaxial probe were applied to detect the dielectric constant ϵ’ and dielectric loss factor ϵ″ of fragrant pear samples in the frequency range of 0.1–26.5GHz, and to analyse the linear relationship between the colour of fragrant pear and the dielectric parameter. Uninformative variables elimination (UVE) and the successive projections algorithm (SPA) were used to extract feature variables from the dielectric spectroscopy data; partial least squares regression (PLSR), support vector regression (SVR), and least squares support vector regression (LSSVR) were used to establish the colour prediction models of Korla f.agrant pear, respectively. The prediction results of color prediction model with full frequency band of dielectric spectrum and feature variable extraction were compared, facilitating the identification of the best prediction model.

The results showed that the linear correlation between ϵ’, ϵ’’ and L*, a*, b* at a single frequency was weak. Both feature variable extraction methods, UVE and SPA, were able to improve the prediction accuracy of the colour of fragrant pear. The SPA-PLSR model showed the best prediction for L* (R2 = 0.83, RMSE = 0.866, RPD = 2.477), while the UVE-PLSR model showed the best prediction for both a* (R2 = 0.85, RMSE = 0.901, RPD = 2.523) and b* (R2 = 0.73, RMSE = 0.895, RPD = 1.973).

The results can provide a new method for the accurate detection of the quality of Korla fragrant pear.

Figures