Selection of NIR variables for online detecting sugar content of navel orange | 脐橙糖度近红外光谱在线检测的建模变量优选

English. In order to improve the precision of detecting sugar content of navel orange by online near infrared (NIR) spectroscopy, wavelet transform and genetic algorithm were applied to select the NIR variables. Spectra was measured in near infraed diffuse reflectance mode using the dynamic spectra detecting system. In the wavelength range of 700.28-933.79 nm, the first derivative spectra were compressed into the variables of wavelet coefficient by wavelet transform (WT). The partial least squares (PLS) models were developed with the variables selected by genetic algorithm (GA). The prediction was used to evaluate the predictive ability of the models. By comparison the predictive performance of the PLS model for navel orange SSC that developed with the variables using WT and GA variables was the best. The correlation coefficient (r) of predictive mode was 0.759, and the root mean square error of prediction (RMSEP) was 0.468 °Brix. The experiment showed that the precision for detecting sugar content of navel orange by WT, GA and online NIR technique was improved.

Chinese. 采用小波压缩结合遗传算法，优选脐橙糖度近红外光谱在线检测的建模变量，提高在线检测精度。利用近红外光谱检测装置采集脐橙样品的光谱，并将其转换为反射比光谱，在700.28~933.79 nm波段，利用小波变换将一阶微分处理后的近红外反射比光谱变量压缩成小波系数变量。经遗传算法优选后，建立偏最小二乘法（PLS）模型，并对该模型的预测结果进行评价。利用小波压缩结合遗传算法优选变量建立的脐橙糖度PLS模型，预测效果最优，模型的相关系数为0.759，模型预测均方根误差为0.468°Brix。采用小波压缩结合遗传算法对变量进行优选，可提高脐橙糖度近红外光谱在线检测的精度。