基于优选波长的多光谱检测系统快速检测猪肉中挥发性盐基氮的含量

发布时间：2018-08-22 20:38

【摘要】：挥发性盐基氮(TVB-N)含量是评价猪肉新鲜度的重要理化指标。为了实现快速、无损检测猪肉的新鲜度,优选出与猪肉中TVB-N含量相关的特征波长,将包含特征波长的发光二极管(LED)光源用于多光谱检测系统,测定了猪肉中TVB-N的含量。首先利用可见-近红外(VIS-NIR)高光谱系统对猪肉进行检测,获取高光谱反射率数据,并采用一阶导数(FD)法、标准正态变量变换(SNV)以及其他预处理方法建立猪肉中TVB-N含量的偏最小二乘回归(PLSR)模型;然后分别利用逐步回归算法(SWA)、连续投影算法(SPA)、基因遗传算法(GA)筛选出与TVB-N含量相关的特征波长,利用筛选出的特征波长分别建立PLSR模型与多元线性回归(MLR)模型,比较模型结果后进一步优选特征波长;最后,将含有特征波长的LED光源用于多光谱检测系统,并建立PLSR模型与MLR模型,从而完成对猪肉中TVB-N含量的测定。实验结果表明:利用SWA、SPA、GA这3种算法筛选出的特征波长能够很好地反映全光谱的信息,建立的模型效果较好,变量数大大减少;包含优选特征波长的LED光源在多光谱检测系统中能很好地检测猪肉中的TVB-N含量;MLR模型结果好于PLSR模型结果,MLR模型的校正集相关系数和校正集均方根误差分别为0.9050和3.63×10-5,预测集相关系数和预测集均方根误差分别为0.9040和3.81×10-5。
[Abstract]:Volatile base nitrogen (TVB-N) content is an important physical and chemical index to evaluate the freshness of pork. In order to realize fast and nondestructive detection of pork freshness, the characteristic wavelengths related to the content of TVB-N in pork were selected. The light-emitting diode (LED) light source containing characteristic wavelength was used in the multispectral detection system to determine the content of TVB-N in pork. Firstly, the hyperspectral reflectance data of pork were detected by using VIS-NIR hyperspectral system, and the first derivative (FD) method was used to obtain the hyperspectral reflectance data. The partial least squares regression (PLSR) model of TVB-N content in pork was established by standard normal variable transformation (SNV) and other pretreatment methods. Then the stepwise regression algorithm (SWA), continuous projection algorithm (SPA), gene genetic algorithm (GA) was used to screen the characteristic wavelengths related to the TVB-N content. The PLSR model and the multivariate linear regression (MLR) model were established by using the selected characteristic wavelengths. Finally, the LED light source with characteristic wavelength was applied to the multispectral detection system, and the PLSR model and MLR model were established to determine the content of TVB-N in pork. The experimental results show that the characteristic wavelengths selected by SWA-SPAGA can well reflect the information of the whole spectrum, the model is effective and the number of variables is greatly reduced. The LED light source with optimized characteristic wavelengths can be used to detect the TVB-N content in pork in a multispectral detection system. The results of MLR model are better than those of PLSR model. The correlation coefficient of correction set and the root mean square error of correction set are 0.9050, respectively. The correlation coefficient of prediction set and the root mean square error of prediction set are 0.9040 and 3.81 脳 10 ~ (-5), respectively.
【作者单位】：中国农业大学工学院国家农产品加工技术装备研发分中心;
【基金】：国家重点研发计划(2016YFD0401205)
【分类号】：O657.33;TS251.51

【相似文献】