多模式可调节光学机构的苹果可溶性固形物近红外光谱检测

Research on Vis/NIR Detection of Apple’s SSC Based on Multi-Mode Adjustable Optical Mechanism

利用多模式可调节的光学机构采集了苹果漫透射、全透射和漫反射三种检测方式的光谱,研究在不同的检测方式下苹果的光谱特征并采用PLS建立苹果可溶性固形物含量SSC预测的模型。首先分别采集每个样品赤道上四点的漫透射、全透射和漫反射光谱,然后分别使用多元散射校正MSC、基线偏移校正BOC、归一化Normalize和高斯滤波平滑GFS等方法对平均后的120个光谱做预处理,并结合竞争性自适应权重取样CARS法对漫反射光谱进行特征波长筛选,最后采用偏最小二乘法PLS回归建立预测苹果SSC的模型,并另购30个苹果验证模型性能。结果表明,苹果在三种检测方式下采集的光谱吸收峰和波谷所处的波段大致相同,但光谱强度有差异。三种光谱经3点高斯滤波平滑GFS预处理后建立预测苹果SSC含量的模型均取得很好的结果。漫透射的模型性能为R_(cal)=0.972,R_(pre)=0.967和RMSEC=0.436%,RMSEP=0.507%;全透射的模型性能为R_(cal)=0.964、R_(pre)=0.957和RMSEC=0.5%,RMSEP=0.574%;漫反射的模型性能为R_(cal)=0.963,R_(pre)=0.949和RMSEC=0.522%,RMSEP=0.536%;三种光谱经归一化预处理后融合建模的模型性能为R_(cal)=0.894,R_(pre)=0.857和RMSEC=0.836%,RMSEP=0.966%。进一步将漫反射光谱结合CARS算法筛选特征波长,使用119个变量建立模型的性能为R_(cal)=0.986,R_(pre)=0.977和RMSEC=0.323%,RMSEP=0.362%。最后将该模型导入新型多模式可调节的水果检测系统中,使用30个未参与建模的冰糖心苹果检验该模型预测苹果SSC的性能。结果显示30个外部验证集的相关系数为0.906,验证均方根误差为0.707%。进一步表明使用多模式可调节的水果内部品质检测系统采集的漫反射光谱结合光谱预处理、波段筛选算法和偏最小二乘回归方法可以建立较好的模型预测苹果可溶性固形物SSC含量。本研究为苹果的品质检测提供新的技术支持。

A multi-mode adjustable optical mechanism was used to collect the spectra of apples in three detection methods:diffuse transmittance,total transmittance and diffuse reflection.The spectrum characteristics of apples under different detection methods were studied and PLS established the apple soluble solid content prediction model.First,the system will collect the spectra of four points on the equator of each sample under the way of diffuse transmittance,total transmittance and diffuse reflection respectively,and then MSC(Multiplicative Scatter Correction),BOC(Baseline offset correction),Normalize,Gaussian filter smoothing and other methods will be applied to preprocess the 120 averaged spectra combined with the CARS method to filter the characteristic wavelength of the diffuse reflection spectrum.Finally,PLS will be used to establish a model for predicting apple’s SC,and another 30 apples will be purchased to verify the performance of the model.The results show that the spectra collected by the self-designed fruit quality detection system under the three detection methods have good result in predict SSC content in apples after 3-point Gaussian filtering smoothing pretreatment.The performance of diffuse transmittance model is R_(cal)=0.972,R_(pre)=0.967 and RMSEC=0.436%,RMSEP=0.507%.The performance of total transmittance model is R_(cal)=0.964,R_(pre)=0.957 and RMSEC=0.5%,RMSEP=0.574%.The performance of diffuse reflection model is R_(cal)=0.963,R_(pre)=0.949 and RMSEC=0.522%,RMSEP=0.536%.The fusion modeling performance of the three spectra after normalization pretreatment is R_(cal)=0.894,R_(pre)=0.857 and RMSEC=0.836%,RMSEP=0.966%.Further,the diffuse reflection spectrum is combined with the CARS algorithm to filter the characteristic wavelengths.The performance of the model established with 119 variables is R_(cal)=0.986,R_(pre)=0.977 and RMSEC=0.323%,RMSEP=0.362%.Finally,the model is imported into this new multi-mode adjustable fruit detection system,and 30 non-model apples are used to test the model to predict the performance of apple’s SSC.The results show that the correlation coefficient of the 30 external validation sets is 0.906,and the root means square error of validation is 0.707%.It further shows that the diffuse reflection spectrum collected by the multi-mode adjustable fruit internal quality detection system,which is combined with spectral pretreatment,CARS and PLS can establish a better model to predict the solid soluble content of apple.This research provides new technical support for Apple’s internal quality testing.