ESTIMATION AND VISUALIZATION OF SOLUBLE SUGAR CONTENT IN OILSEED RAPE LEAVES USING HYPERSPECTRAL IMAGING

A novel technology for hyperspectral imaging was used to estimate the soluble sugar content of rape leaves. Rape leaves collected at four different growth stages (seedling stage, bolting stage, florescence stage, and pod stage) were analyzed. Sample regions in the hyperspectral images were extracted by removing the background, and spectral data of all pixels in the sample region were extracted and averaged as the spectrum of the sample. Eleven of 128 samples were defined as outliers. The remaining samples were split into a calibration set and a prediction set in a 3: 1 ratio using two methods: considering the difference in growth stage (method 1) and not considering the growth stage (method 2). Eight and seven sensitive wavelengths were selected by a successive projections algorithm (SPA) for methods 1 and 2, respectively. Partial least squares (PLS) was applied to build calibration models using full spectra and sensitive wavelengths, and multiple linear regression (MLR) and back-propagation neural network (BPNN) models were built using selected wavelengths. The calibration models for method 1 performed better than the models for method 2. The BPNN for method 1 using sensitive wavelengths obtained the optimal prediction result, with a correlation coefficient of prediction (r(p)) of 0.885, root mean square error of prediction (RMSEP) of 2.010 mg g(-1), and residual prediction deviation (RPD) of 2.177. The overall results indicated that hyperspectral imaging could be used for detection of soluble sugar content in rape leaves. In addition to building calibration models for soluble sugar content estimation, visualization of soluble sugar distribution in rape leaves was achieved by predicting the soluble sugar content of each pixel within the image using the optimal SPA-BPNN model.