采用近红外(NIR)漫反射光谱法对新疆特色梨果库尔勒香梨的五种不同果(包括青头、 粗皮、 脱萼、 宿萼、 突顶果)的硬度进行测定。 由于近红外光谱数据量大且原始光谱噪声明显、 测定水果时散射严重等导致光谱建模时关键波长变量提取困难。 以新疆库尔勒香梨为研究对象, 为了有效地消除固体表面散射以及光程变化对NIR漫反射光谱的影响, 首先采用标准正态变量变换(SNV)和多元散射校正(MSC)对库尔勒香梨的原始光谱进行预处理。 为寻找适合近红外光谱检测库尔勒香梨硬度的最佳特征波长筛选方法, 进行香梨近红外光谱的特征波长变量选择方法的比较与研究。 研究比较了两种特征波长筛选方法对库尔勒香梨硬度偏最小二乘法(PLS)建模精度的影响。 同时使用反向偏最小二乘(BiPLS)和遗传算法结合反向偏最小二乘(BiPLS-GA)在全光谱范围内筛选香梨硬度的特征波长变量, 将校正均方根误差(RESMC)、 预测均方根误差(RESMP)以及决定系数( R²)作为模型的评价标准, 并最终确定最优波段选择方法及最佳预测模型。 基于选择的特征波长变量建立的PLS模型(BiPLS-GA)与全光谱变量建立的PLS模型进行比较发现BiPLS-GA模型仅仅使用原始变量中6.6%的信息就获得了比全变量PLS模型更好的库尔勒香梨硬度的预测结果, 其中 R², RMSEC和RMSEP分别为0.91, 1.03和1.01。 进一步与基于反向偏最小二乘算法(BiPLS)获得的特征变量建立的PLS模型比较发现, BiPLS-GA不仅可以去除原始光谱数据中的无信息变量, 同时也能够对共线性的变量进行压缩去除, 使得建模变量从301个减少到20个。 极大地简化模型的同时有效地提高了模型的预测精准度和稳定性。 因此该方法能够有效地用于近红外光谱数据变量的选择。 证明了近红外光谱分析技术结合BiPLS-GA模型能够高效地选择出建模变量, 去除与库尔勒香梨硬度无关的近红外光谱信息, 显著地提高库尔勒香梨硬度定量模型的预测精度。 这不仅为新疆地区特色梨果库尔勒香梨的快速、 精确、 无损优选分级提供一定的技术支持, 同时也为基于近红外光谱分析技术预测水果内部品质的研究提供了参考。

Near-infrared diffuse reflectance spectroscopy was used to determine the hardness of five different fruits (including green head, rough skin, dislocated, scorpion, and apex) of Xinjiang pear fruit Korla pear. Due to the large amount of data in the near-infrared spectrum, the original spectral noise is obvious, and the scattering of fruits is serious, the key wavelength variables are difficult to extract during spectral modeling. Based on this, in order to effectively eliminate the influence of solid surface scattering and optical path variation on the NIR diffuse reflectance spectrum, it is proposed to use standard normal variable transformation (SNV) and multiple scattering correction (MSC). The original spectrum of Korla pear was pretreated. In order to find the best characteristic wavelength screening method suitable for the detection of Korla pear hardness by near-infrared spectroscopy, the comparison and research on the characteristic wavelength variable selection methods of Pear near infrared spectrum were carried out. The effects of two characteristic wavelength screening methods on the modeling accuracy of Korla pear hardness partial least squares (PLS) were compared. Simultaneously using the reverse partial least squares (BiPLS) and genetic algorithm combined with reverse partial least squares (BiPLS-GA) to screen the characteristic wavelength variable of the pear hardness in the whole spectral range, the corrected root mean square error (RESMC), The prediction root mean square error (RESMP) and the decision coefficient ( R²) were used as the evaluation criteria of the model, and the optimal band selection method and the optimal prediction model were finally determined. The PLS model based on the selected characteristic wavelength variable (BiPLS-GA) was compared with the PLS model established by the full spectral variable. It was found that the BiPLS-GA model obtains better information than the full-variable PLS model by using only 6.6% of the information in the original variable. The prediction results of Korla pear hardness, where R², RMSEC and RMSEP are 0.91, 1.03 and 1.01, respectively. Furthermore, compared with the PLS model established by the feature variables obtained by the reverse partial least squares algorithm (BiPLS), BiPLS-GA can not only remove the non-information variables in the original spectral data, but also compress and remove the collinear variables, reducing the number of modeling variables from 301 to 20. The model is greatly simplified while the prediction accuracy and stability of the model are effectively improved. Therefore, the method can be effectively used for the selection of near-infrared spectral data variables. It is proved that the near-infrared spectroscopy analysis technology combined with the BiPLS-GA model can efficiently select the modeling variables, remove the near-infrared spectral information unrelated to the hardness of Korla pear, and significantly improve the prediction accuracy of the Korla pear hardness quantitative model. This not only provides a certain technical support for the rapid, precise and non-destructive optimization of the characteristic pear fruit Korla pear in Xinjiang, but also provides a reference for the research of predicting the internal quality of fruit based on near-infrared spectroscopy.