基于近红外高光谱成像技术的宁夏羊肉产地鉴别

使用900~1700 nm高光谱成像系统采集宁夏银川、固原、盐池三个不同产地的绵羊后腿样本的近红外高光谱数据,对光谱采用面积归一化方法预处理,利用SPA、CARS、UVE算法对预处理后的光谱数据提取特征波长分别为17、40、121个;结合PLS-DA及KNN建立特征波段下的判别模型。结果表明KNN判别模型效果较差,3种特征波长中利用CARS提取的特征波长建模效果最佳,代替全光谱建立PLS-DA判别模型是可行的;综合对比模型效果,CARS-PLS-DA为最优模型,校正集正确率90.48%,预测集正确率84.21%。证明利用近红外高光谱成像技术对羊肉产地鉴别是可行的。

Identification of geographical origins of mutton in Ningxia based on the near infrared hyperspectral imaging technique

Near-infrared hyperspectral imaging system that was ranging from 900 nm to 1700 nm was used to collect near-infrared hyperspectral data of the sheep hind leg samples from three different habitats in Yinchuan,Guyuan and Yanchi of Ningxia province. The spectral normalization method was used to pre-treat the spectrum. The spectral data of the characteristic wavelengths extracted from the pretreatment were 17,40,121 respectively. Using SPA,CARS and UVE,and the discriminant model under the characteristic band was established by combining PLS-DA and KNN. The results showed that the KNN discriminant models were less effective. It was feasible to construct the PLS-DA discriminant model instead of the whole spectrum using the best characteristic of the characteristic wavelengths extracted by CARS. Compared with the model effect,CARS-PLS-DA was the optimal model,the correctness of the calibration set was 90.48%,and the correctness of the forecast set was 84.21%. It was proved that it was feasible to identify the mutton place of origin by near infrared spectroscopy.