Evaluation of dielectric spectroscopy in fusion with Vis-SWNIR spectroscopy for measurement of sugar concentration on sugarcane stalk samples
Development of rapid instruments for measuring the quality attributes of sugarcane such as sugar concentration and water content is one of the important needs in agricultural and industrial technology of sugarcane production and processing. Near infrared (NIR) spectroscopy has been one of the techniques widely applied for quality evaluation of sugarcane. However, NIR spectrophotometers in the full NIR wavelength range (up to 2500 nm) are expensive devices which are not readily available for portable applications. Short-wave NIR devices in the range of 1100 nm are available with lower costs but need to be evaluated for any specific application. On the other hand, dielectric spectroscopy has attracted the attention of researchers for quality evaluation of agricultural and food products. In a previous study, a parallel-plate capacitance sensor was developed and evaluated for non-destructive measurement of sugarcane Brix (total soluble solids) and Pol (sucrose concentration) as well as water content in the frequency range of 0-10 MHz. The results showed excellent prediction models with root mean square errors smaller than 0.3 for Brix and Pol. This study was aimed at development and evaluation of a dielectric sensor to be extendable for portable measurements on standing sugarcane stalk in comparison with short-wave NIR (SWNIR) spectroscopy to address how the fusion of the two methods may improve the accuracy of models for prediction of sugarcane Brix.
Materials and Methods
A dielectric senor in the form of a gadget was developed with metallic electrodes to encompass the sugarcane stalk samples. The dielectric sensor was excited with sinusoidal voltage within 0-150 MHz frequency by a function generator and the conductive power through the electrodes was measured with a spectrum analyzer. 105 sugarcane stalk samples were prepared from seven sugarcane varieties and scanned by the dielectric sensor. The samples were also subjected to Vis-SWNIR radiation in the wavelength range of 400-1100 nm and the reflectance spectra were captured. Reference Brix and water content of the samples were determined using a portable refractometer and oven-drying method, respectively. Regression analyses and artificial neural networks were performed on independent and combined data of dielectric and Vis-SWNIR spectroscopy to develop prediction models for Brix and water content.
Results and discussion
Partial least square regression on independent data sets of each instrument resulted in RMSEP= 1.14 and RMSEP= 1.88 for Brix using Vis-SWNIR and dielectric spectroscopy, respectively. Moreover, data fusion of dielectric and Vis-SWNIR spectroscopy in low level for prediction of Brix significantly improved the prediction accuracy to R2P= 0.94 and RMSEP= 0.74. The medium level data fusion resulted in R2P= 0.89 and RMSEP= 0.93 for prediction of water content.
In this study, Vis-SWNIR spectroscopy and dielectric spectroscopy methods were evaluated for measurement of sugarcane Brix and water content on stalk samples. To develop prediction models for Brix and water content, partial least squares (PLS) regression and artificial neural network (ANN) were compared. First, the prediction models were developed based on independent measurement of Vis-SWNIR and dielectric spectroscopy and then the fusion of the two techniques to improve the prediction accuracy was investigated. With fusion of the two methods at an intermediate level, the RMSE of Brix was obtained to be 0.74 which is a noticeable improvement as compared with the other studies. Based on the results achieved, development of a fusion probe of SWNIR and dielectric spectroscopy and measuring system could be the aim of future studies for in-situ evaluation of standing sugarcane quality parameters. Regarding the importance of measuring the quality properties of sugarcane during growth and maturity, the results of this study can have a significant role in development of a combined portable measurement system of NIR and dielectric spectroscopy methods for fast and non-destructive indication of sugarcane quality parameters.
This article was extracted from a research project financially supported by the research deputy of Shahrekord University. The Grant number was 0GRD34M1614. The authors would like to appreciate the support of the Amir-Kabir Sugarcane Agro-Industry Co., Khuzestan, Iran for providing the sugarcane stalk samples.
Keywords: Dielectric spectroscopy, Visible-Near infrared spectroscopy, Data fusion, Brix, Sugarcane