论文部分内容阅读
为研究能否通过对算法参数的调整和算法的组合来减弱甚至消除便携式近红外仪和样品组织结构等对样品光谱信息的影响,提高模型的预测准确性和稳健性,实现现场快速无损检测生鲜羊肉挥发性盐基氮(total volatile basic nitrogen,TVB-N)的目的。本研究应用不同参数组合的单一算法和不同算法组合对样品的光谱信息进行预处理并建模,从模型的预测准确性和稳健性2个方面探讨算法参数和算法组合对模型性能的影响,找出针对检测生鲜羊肉中TVB-N含量的最佳预处理方法。结果表明,不同的算法参数和算法组合对模型性能的影响差别很大,对样品的近红外光谱信息进行差分求导(窗口数为6,求导阶次为1)后,模型性能最佳。模型的校正标准差和验证标准差分别为1.21和1.31,校正标准差和验证标准差的比值为1.08小于1.2,主成分数为10,校正集相关系数和验证及相关系数分别为0.94和0.92。说明通过对算法参数的调整和对算法的组合可以有效提高模型性能,满足应用便携式近红外仪现场快速无损检测生鲜羊肉TVB-N含量的要求。
In order to study whether the influence of portable NIR and sample structure on the spectral information of the sample can be reduced or even eliminated by adjusting the algorithm parameters and the algorithm, the prediction accuracy and robustness of the model can be improved and the on-site rapid non-destructive test Fresh mutton volatile basic nitrogen (total volatile basic nitrogen, TVB-N) purposes. In this study, the spectral information of the sample was preprocessed and modeled using a single algorithm with different combinations of parameters and different combinations of algorithms. The effects of the algorithm parameters and algorithm combination on the performance of the model were discussed from the prediction accuracy and robustness of the model. The best pretreatment method for detecting the content of TVB-N in fresh mutton is given. The results show that the influence of different algorithm parameters and algorithm combination on the performance of the model is very different. The best performance of the model is obtained when the near-infrared spectral information of the sample is differentially derived (the number of windows is 6 and the order of derivative is 1). The standard deviation of calibration and standard deviation of validation were 1.21 and 1.31, respectively. The ratio of calibration standard deviation to standard deviation of validation was 1.08 less than 1.2, the principal component was 10, correlation coefficient of calibration set and validation and correlation coefficient were 0.94 and 0.92 respectively. It indicates that the performance of the model can be effectively improved through the adjustment of the algorithm parameters and the combination of algorithms to meet the requirements of rapid and nondestructive detection of the TVB-N content of fresh mutton in the portable near-infrared instrument.