Limits...
Application of Visible and Near-Infrared Hyperspectral Imaging to Determine Soluble Protein Content in Oilseed Rape Leaves.

Zhang C, Liu F, Kong W, He Y - Sensors (Basel) (2015)

Bottom Line: SPA-PLS model obtained the best performance with r(p) of 0.9554, RMSEP of 0.1538 mg/g and RPD of 3.25.Distribution of protein content within the rape leaves were visualized and mapped on the basis of the SPA-PLS model.The overall results indicated that hyperspectral imaging could be used to determine and visualize the soluble protein content of rape leaves.

View Article: PubMed Central - PubMed

Affiliation: College of Biosystems Engineering and Food Science, Zhejiang University, 866 Yuhangtang Road, Hangzhou 310058, China. chuzh@zju.edu.cn.

ABSTRACT
Visible and near-infrared hyperspectral imaging covering spectral range of 380-1030 nm as a rapid and non-destructive method was applied to estimate the soluble protein content of oilseed rape leaves. Average spectrum (500-900 nm) of the region of interest (ROI) of each sample was extracted, and four samples out of 128 samples were defined as outliers by Monte Carlo-partial least squares (MCPLS). Partial least squares (PLS) model using full spectra obtained dependable performance with the correlation coefficient (r(p)) of 0.9441, root mean square error of prediction (RMSEP) of 0.1658 mg/g and residual prediction deviation (RPD) of 2.98. The weighted regression coefficient (Bw), successive projections algorithm (SPA) and genetic algorithm-partial least squares (GAPLS) selected 18, 15, and 16 sensitive wavelengths, respectively. SPA-PLS model obtained the best performance with r(p) of 0.9554, RMSEP of 0.1538 mg/g and RPD of 3.25. Distribution of protein content within the rape leaves were visualized and mapped on the basis of the SPA-PLS model. The overall results indicated that hyperspectral imaging could be used to determine and visualize the soluble protein content of rape leaves.

Show MeSH

Related in: MedlinePlus

Original RGB image (a) and the distribution maps of soluble protein content of oilseed rape leaf (the average predicted value is on the top of the figure) (b).
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4541894&req=5

sensors-15-16576-f006: Original RGB image (a) and the distribution maps of soluble protein content of oilseed rape leaf (the average predicted value is on the top of the figure) (b).

Mentions: The spectral and spatial information of each pixel in the hyperspectral image made it possible to predict the quality parameters of each pixel by using the calibration models [18,21]. The pixels with similar spectral and spatial information would be predicted to have similar quality parameters, and then these pixels would present similar visualization. Otherwise, the pixels with different characteristics would present different visualizations with different quality parameters. In this study, SPA-PLS obtained the best results with the least sensitive wavelengths, and SPA-PLS was used to estimate the soluble protein content of each pixel. The prediction values of all pixels were calculated by the following equation:(2)Y=BX+B0where X was the selected sensitive wavelengths of each pixel, and B was the corresponding matrix of the regression coefficient, and B0 was a constant fitting this model (B0 = 2.11708 in this study). The visualization map of the predicted soluble protein was achieved and spatially smoothed (shown in Figure 6b), and the average predicted value of soluble protein content in the leaf presented in Figure 6b was 2.5765 mg/g, which was similar to the average reference value of 2.5515 mg/g. The result of the prediction map indicated that using hyperspectral imaging to estimate the soluble protein content of rape leaves was feasible. However, it could be noted that predicted soluble protein contents of some parts exceeded the range of the calibration set and the prediction set, the reason might be: (1) the reference value was the average value of three measurements, and only a small part of the sample was used in each measurement, and the average spectrum of the sample was used to build calibration models; (2) the noises of the hyperspectral image affected the spectrum of each pixel; (3) the leaf vein showed different spectral features from the leaf part, resulting the maximum values in leaf vein in Figure 6b. For future study of visualization of soluble protein content distribution, the shortcomings should be overcome by improving the model performances with wider reference value range and minimum noises of the spectrum.


Application of Visible and Near-Infrared Hyperspectral Imaging to Determine Soluble Protein Content in Oilseed Rape Leaves.

Zhang C, Liu F, Kong W, He Y - Sensors (Basel) (2015)

Original RGB image (a) and the distribution maps of soluble protein content of oilseed rape leaf (the average predicted value is on the top of the figure) (b).
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4541894&req=5

sensors-15-16576-f006: Original RGB image (a) and the distribution maps of soluble protein content of oilseed rape leaf (the average predicted value is on the top of the figure) (b).
Mentions: The spectral and spatial information of each pixel in the hyperspectral image made it possible to predict the quality parameters of each pixel by using the calibration models [18,21]. The pixels with similar spectral and spatial information would be predicted to have similar quality parameters, and then these pixels would present similar visualization. Otherwise, the pixels with different characteristics would present different visualizations with different quality parameters. In this study, SPA-PLS obtained the best results with the least sensitive wavelengths, and SPA-PLS was used to estimate the soluble protein content of each pixel. The prediction values of all pixels were calculated by the following equation:(2)Y=BX+B0where X was the selected sensitive wavelengths of each pixel, and B was the corresponding matrix of the regression coefficient, and B0 was a constant fitting this model (B0 = 2.11708 in this study). The visualization map of the predicted soluble protein was achieved and spatially smoothed (shown in Figure 6b), and the average predicted value of soluble protein content in the leaf presented in Figure 6b was 2.5765 mg/g, which was similar to the average reference value of 2.5515 mg/g. The result of the prediction map indicated that using hyperspectral imaging to estimate the soluble protein content of rape leaves was feasible. However, it could be noted that predicted soluble protein contents of some parts exceeded the range of the calibration set and the prediction set, the reason might be: (1) the reference value was the average value of three measurements, and only a small part of the sample was used in each measurement, and the average spectrum of the sample was used to build calibration models; (2) the noises of the hyperspectral image affected the spectrum of each pixel; (3) the leaf vein showed different spectral features from the leaf part, resulting the maximum values in leaf vein in Figure 6b. For future study of visualization of soluble protein content distribution, the shortcomings should be overcome by improving the model performances with wider reference value range and minimum noises of the spectrum.

Bottom Line: SPA-PLS model obtained the best performance with r(p) of 0.9554, RMSEP of 0.1538 mg/g and RPD of 3.25.Distribution of protein content within the rape leaves were visualized and mapped on the basis of the SPA-PLS model.The overall results indicated that hyperspectral imaging could be used to determine and visualize the soluble protein content of rape leaves.

View Article: PubMed Central - PubMed

Affiliation: College of Biosystems Engineering and Food Science, Zhejiang University, 866 Yuhangtang Road, Hangzhou 310058, China. chuzh@zju.edu.cn.

ABSTRACT
Visible and near-infrared hyperspectral imaging covering spectral range of 380-1030 nm as a rapid and non-destructive method was applied to estimate the soluble protein content of oilseed rape leaves. Average spectrum (500-900 nm) of the region of interest (ROI) of each sample was extracted, and four samples out of 128 samples were defined as outliers by Monte Carlo-partial least squares (MCPLS). Partial least squares (PLS) model using full spectra obtained dependable performance with the correlation coefficient (r(p)) of 0.9441, root mean square error of prediction (RMSEP) of 0.1658 mg/g and residual prediction deviation (RPD) of 2.98. The weighted regression coefficient (Bw), successive projections algorithm (SPA) and genetic algorithm-partial least squares (GAPLS) selected 18, 15, and 16 sensitive wavelengths, respectively. SPA-PLS model obtained the best performance with r(p) of 0.9554, RMSEP of 0.1538 mg/g and RPD of 3.25. Distribution of protein content within the rape leaves were visualized and mapped on the basis of the SPA-PLS model. The overall results indicated that hyperspectral imaging could be used to determine and visualize the soluble protein content of rape leaves.

Show MeSH
Related in: MedlinePlus