连续投影算法在猪肉pH值无损检测中的应用

利用鲜肉的近红外光谱中少量特征波长对其pH值进行预测，可以大幅度降低模型复杂性和计算量，对开发无损检测装置, 实施肉品生产加工过程中pH值监测有重要意义。该文通过连续投影算法（SPA）选择特征波长建立简单多元线性回归模型（SPA-MLR），并对比了SPA-MLR模型与全波段（5 000～10 440 cm-1）偏最小二乘回归模型（PLSR）及逐步线性回归（SMLR）、遗传算法（GA）选择特征波长所建模型的性能。结果表明经连续投影算法提取37个特征波长建立的模型，所用变量数仅占全波段的2.6%，校正集相关系数0.870，校正集均方根误差为0.094，验证集相关系数0.892，验证集均方根误差为0.085；性能与经多元散射校正预处理的PLSR模型接近，但采用变量数明显减少，优于逐步线性回归和遗传算法选择特征波长建立的模型，表明该方法可较好的选择特征波长，建立简单的预测模型。

Application of successive projections algorithm to nondestructive determination of pork pH value

Using a few variables from the fresh pork spectra to construct the model of pH prediction is capable of decreasing of calculated amount, which has a vital significance to determine and monitor pork pH values. In this study, successive projection algorithm (SPA) was proposed to select feature wavelength from pork to determine the pH values. The performance of the model which constructed by variables selected of SPA was compared with various models including the model of partial least squares regression (PLSR) based on full spectrum (5 000-10 440 cm-1), the model constructed by variables selected of stepwise multiple linear regression (SMLR) and the model constructed by variables selected of genetic algorithm (GA). A total of 37 variables, only 2.6 percent in the full spectrum, selected by SPA were employed to construct the model with 0.870 as the correlation coefficient and 0.094 as the root of mean square error of calibration set and 0.892 as the correlation coefficient and 0.085 as the root of mean square error of validation set. While it was nearly to the PLSR model with preprocess of multiplicative signal correction, the SPA-MLR model is more accurate than SMLR model and GA model. The results confirm that SPA can be applied to select few variables from huge information space of NIR spectroscopy to build simple model to determine fresh pork pH values.