基于近中红外光谱技术鉴别4种大黄鱼产地

Identification of 4 kinds of main Larimichthys crocea production areas based on Fourier transform infrared spectroscopy

中文关键词:  红外光谱  大黄鱼  产地区分  主成分分析  聚类分析

英文关键词:infrared spectroscopy  Larimichthys crocea  geographical identification  principal component analysis  cluster analysis

中文摘要:

      目的 基于近中红外光谱技术准确快速区分 4 种产地来源的大黄鱼(福建养殖大黄鱼、温州养殖大黄 鱼、舟山养殖大黄鱼和舟山野生大黄鱼)。方法 本研究应用傅里叶变换红外光谱仪对 4 种大黄鱼样本在 4000~650 cm?1 的红外吸收指纹图谱进行测定, 基于特征波段下的光谱吸收差异并结合主成分分析(principal component analysis, PCA)、聚类分析(cluster analysis, CA)和线性判别分析(linear discriminant analysis, LDA)、支 持向量机(support vector machine, SVM)模型对大黄鱼样品进行产地区分。结果 采用测定波段 4000~650 cm?1的 全光谱采集信息经过 Savitzky-Golay 算法平滑预处理后建立的 SVM 模型效果最优, 对 4 种大黄鱼样本的测 试集准确率为 83.3%。进一步对福建养殖大黄鱼和野生大黄鱼的产地区分方法优化后, 选取特征波段 3690~2800 cm?1+1800~650 cm?1 的光谱信息经过一阶导数(first derivative, 1st Der)、二阶导数(second derivative, 2nd Der)和标准正态变换(standard normal variate transformation, SNV) 3种方式预处理后建立 LDA判别模型, 光 谱训练集与预测集的准确率均达到 100%; 3690~2800 cm?1+1800~650 cm?1 的光谱信息经 SNV、多元散射校正 (multiplicative scatter correction, MSC)预处理后的 PCA 效果最佳, 2 种大黄鱼样本间彼此间距远、无重叠, 且前 两个主成分累计贡献率均在 90%以上; 经 SNV 预处理后的 CA 分析中, 除 21 号野生大黄鱼外, 其余产地相同 的大黄鱼样本均各自聚为一类。结论 基于近中红外光谱测定并结合化学计量学处理的方法能够对大黄鱼产 地进行较准确地快速区分, 从而为大黄鱼溯源鉴别提供技术支撑。

英文摘要:

      Objective To accurately and quickly distinguish 4 sources of Larimichthys crocea (Lar. crocea) (Fujian farmed Larimichthys crocea, Wenzhou farmed Larimichthys crocea, Zhoushan farmed Larimichthys crocea, and Zhoushan wild Larimichthys crocea) based on Fourier transform infrared spectroscopy. Methods The infrared absorption fingerprints at 4000?650 cm?1 of Lar. crocea in 4 kinds of producing areas were determined by Fourier transform infrared spectroscopy. Based on the spectral absorption difference under the characteristic band and combined with models of principal component analysis (PCA), cluster analysis (CA), linear discriminant analysis (LDA) and support vector machine (SVM), the geographical identification of Lar. crocea was distinguished. Results The SVM model established using the full-spectrum acquisition information at 4000?650 cm?1 after Savitzky-Golay algorithm smoothing pretreatment had the best effect, and the test accuracy was 83.3%. After the further optimization of the method for geographical identification of Fujian and wild Lar. crocea, the spectral information at 3690?2800 cm?1+ 1800?650 cm?1 was selected and preprocessed by first derivative (1st Der), second derivative (2nd Der) and standard normal variate transformation (SNV) to establish the LDA discriminant model. The accuracy of the training set and the prediction set all reached 100%. PCA model of spectral information at 3690?2800 cm?1 + 1800?650 cm?1 after SNV or multiplicative scatter correction (MSC) pretreatment was the best. Samples of Lar. crocea from the 2 producing areas were far away from each other, and the cumulative contribution rate of the first 2 principal components was more than 90%. CA analysis after SNV pretreatment showed that samples with the same origin were respectively clustered into one category except one wild Lar. crocea (No.21). Conclusion The results show that the method based on near-mid infrared spectroscopy combined with chemometrics can accurately and quickly distinguish the geographical origin of Lar. crocea, so as to provide technical support for the traceability identification of Lar. crocea.