草莓可溶性固形物含量是评价草莓内部品质的关键指标。为了实现对该指标的快速、无损评估,本研究基于近红外光谱技术,构建了线性偏最小二乘PLS和非线性最小二乘支持向量机LS-SVM模型,联合蒙特卡罗无信息变量消除和连续投影算法MC-UVE-SPA从原始光谱4 254个变量中提取了27个有效变量并构建了基于有效变量的定量分析模型。同时,考虑到草莓表面颜色的影响,基于草莓RGB图像各分量获取了颜色特征参数,进一步融合光谱和颜色特征构建了多参数融合PLS和LS-SVM模型。基于相同的校正集和预测集,比较了所有模型对草莓内部SSC的预测性能,结果表明,MC-UVE-SPA是一种有效的草莓光谱变量选择算法,且多参数融合非线性LS-SVM模型是草莓内部SSC定量预测的最优模型。针对预测集样本,该模型相关系数RP和预测均方根误差RMSEP分别为0.9885和0.1532。本研究为基于近红外光谱技术的草莓可溶性固形物含量检测便携式仪器和在线检测设备研发奠定了基础。
Soluble solids content (SSC) is a key index to evaluate the quality of strawberries. In order to achieve the non-destructive evaluation of SSC, the near infrared spectroscopy was used to build the linear partial least squares (PLS) and nonlinear least squares support vector machine (LS-SVM) models. Twenty-seven effective variables were selected from the original 4 254 variables by combining both successive projections algorithm and Monte-Carlo uninformative variable elimination. The quantitative analysis models were then established by using the selected effective variables. At the same time, color feature parameters were obtained based on components of RGB images of samples considering the influence of surface color on strawberries, and the multi-parameter PLS and LS-SVM models were constructed by further integrating spectra and color features. Based on the same correction set and prediction set, the prediction performance of all models for SSC was compared. The results showed that MC-UVE-SPA was an effective spectral variable selection algorithm, and the multi-parameter fusion nonlinear LS-SVM model was the optimal model for the quantitative prediction of SSC in strawberries. For samples in the prediction set, the correlation coefficient (RP) and root mean square error of prediction (RMSEP) of the model were 0.9885 and 0.1532, respectively. This study lays a foundation for the development of portable instruments and online detection equipments for the detection of soluble solids content in strawberries based on near infrared spectroscopy.