近红外光谱法检测李果坚实度的研究

Improvement of Models for Quantitative Analysis of Plum Fruit Firmness Based on NIR

为了实现准确无损检测"安哥诺"李果实的坚实度,试验利用MPA近红外光谱仪在4 000~12 500 cm^(-1)光谱范围采集了515个李果实样品的漫反射光谱,采用偏最小二乘法和反向传播人工神经网络建立"安哥诺"李果实坚实度的定量分析模型,使用波段筛选和多种光谱预处理方法优化了偏最小二乘模型。结果表明,4 000~7 267 cm^(-1)波段光谱数据经MSC校正的预处理方法处理后,偏最小二乘定量模型的校正集相关系数和均方根误差分别为0.878 1和1.22 kg/cm^2,预测集相关系数和均方根误差分别为0.836 5和1.51 kg/cm^2,优于BP-ANN模型。因此认为试验所建立的定量模型可为实现近红外无损检测"安哥诺"李果实坚实度提供技术支持和理论依据。

In order to evaluate firmness of plum nondestructively, the diffuse reflectance spectra of 515 plum fruit was collected by near infrared spectroscopy（NIR） in the band of 4 000-12 500 cm^-1. The quantitative models of firmness in plum fruit were established using methods of partial least squares（PLS） and back propagation-artificial neural networks（BPANN）. And the variables selection and different spectral pretreatments were performed to optimize the PLS models. The results showed that the parameters of PLS model with spectral data of 4 000-7 267 cm^-1 and the pretreatment of multiplicative scatter correction were better than those from BP-ANN model. The correlation coefficient and root mean square error of the calibration set in optimal PLS model were 0.878 1 and 1.22 kg/cm^2, respectively; The parameters of the prediction set were 0.836 5 and 1.51 kg/cm^2, which was higher in prediction accuracy than results reported by previous studies. In a conclusion, NIR combined with chemometrics could provide a technical support and theoretical basis for evaluate the firmness of plum fruit nondestructive examination.