三文鱼肉质鲜美、 营养丰富, 尽管价格昂贵, 却深受广大消费者喜爱, 2017年我国三文鱼进口额达3.5亿美元。 近年来不法商贩为追求高额利润导致三文鱼消费市场“以假乱真”、 “以次充好”的问题日益突出, 主要表现为： (1)以价格低廉、 外观相似的淡水虹鳟、 大马哈鱼、 太平洋鲑鱼等冒充价格高、 消费者认可度高的挪威三文鱼; (2)将低成本、 低品质的冰冻三文鱼(-18 ℃储运、 保质期长、 组织结构被冰晶破坏、 口感风味破坏严重)化冻后冒充高成本、 高品质的冰鲜三文鱼(0～4 ℃储运、 保质期短、 无冰晶产生、 口感风味最大限度保持); (3)将次新鲜的三文鱼冒充新鲜三文鱼。 针对三文鱼品质感官检测误差大、 理化检测耗时费力的不足, 拟研究一种基于近红外光谱(NIRs)特征的真品/伪品三文鱼、 冰鲜/冻融三文鱼、 新鲜/次新鲜三文鱼快速鉴别方法。 首先采集真品(挪威三文鱼)/伪品(淡水虹鳟、 大马哈鱼和太平洋鲑鱼)三文鱼、 冰鲜(冰鲜1, 3和5 d)/冻融(冰冻15, 30和45 d并化冻)三文鱼和新鲜/次新鲜(冰鲜保藏0, 2, 4, 6和8 d)三文鱼样品的NIRs信息, 并将不同储藏天数的冰鲜三文鱼以国标法测定其TVB-N含量。 原始光谱经标准正态变量变换(SNV)等方法预处理后, 分别使用主成分分析(PCA)和遗传算法(GA)进行光谱数据降维及特征波长筛选。 最后, 结合K-最近邻法(KNN)和最小二乘支持向量机(LS-SVM)对真品/伪品三文鱼和冰鲜/冻融三文鱼构建识别模型; 结合联合区间偏最小二乘法(Si-PLS)对新鲜/次新鲜三文鱼构建TVB-N预测模型。 建模结果表明： 真品/伪品三文鱼LS-SVM定性识别模型对应的测试集识别率达97.50%, 冰鲜/冻融三文鱼LS-SVM定性识别模型对应的测试集识别率达98.89%; TVB-N对应的Si-PLS定量检测模型的预测集相关系数为0.864 1, 基于TVB-N预测值建立的三文鱼新鲜度定性鉴别模型对应的测试集准确率为90.00%。 研究结果表明, 利用近红外光谱特征结合化学计量学方法能够快速、 无损检测真品/伪品三文鱼、 冰鲜/冻融三文鱼和新鲜/次新鲜三文鱼, 实现三文鱼品质多指标快速检测。

Salmon is expensivebut popular among consumers because of its good-taste, sweet flavor and high nutritional values. The import volume of salmon in 2017 reaches 350 million dollars. The problems of selling shoddy salmon for quality salmon by unscrupulous businessmen, who are pursuing high profit only, become more and more serious. The problems can be mainly manifested by the following steps: (1) Using fresh water rainbow trout with low price and similar appearance like Amur salmon, Pacific salmon to masquerade Norwegian salmon that of high price and high consumer acceptance; (2) Replacing high cost and high quality fresh salmon (stored in 0～4 ℃, with short shelf life, on ice crystal produced and longest maintaining flavor and taste) with low cost and low quality frozen-thawed substitute (stored in -18 ℃, with long shelf life, destroyed organizational structure by ice crystal and destroyed flavor); (3) Selling stale salmon as the fresh ones. Therefore, considering the disadvantages of big error in sensory detection of salmon quality as well as the time consumption in physical and chemical testing, the article intends to research a fast identification method for genuine and counterfeit salmon, fresh and frozen-thawed salmon as well as fresh and sub-fresh salmon based on near infrared spectral characteristics. Firstly, genuine and counterfeit salmon samples were taken from Norwegian salmon and fresh water rainbow trout, Amur salmon, Pacific salmon; fresh and frozen-thawed salmon samples were taken from fresh salmon with chilling for 1, 3 and 5 d and frozen-thawed salmon with frozen for 15, 30 and 45 d; fresh and sub-fresh salmon samples were taken from fresh salmon with 0, 2, 4, 6 and 8 d storage. Secondly, NIRs information was collected, meanwhile, the salmon with different storage days were analyzed by national standard method for determination of the TVB-N. Thirdly, the different pre-processing methods (Standard normal variate transformation, Vector normalization, Multiplicative scatter correction, Savitzky-Golay, First derivative, Second derivative) were employed, then Principal component analysis (PCA) and Genetic algorithms (GA) were used to reduce the spectral and the excess spectral bands. Finally, K-nearest neighbors (KNN) and Least-squares support vector machine (LS-SVM) models were used for the construction of identification model of genuine and counterfeit salmon as well as fresh and frozen-thawed salmon; the prediction spectra were constructed associated with their corresponding TVB-N using Synergy Interval Partial Least Square Method (Si-PLS). Modeling results show that for genuine and counterfeit salmon, the spectral information were treated with SNV and PCA, the LS-SVM model recognition rate of the testing set is 97.50%; for fresh and frozen-thawed salmon, the spectral information were treated with SNV and PCA, the LS-SVM model recognition rate of the testing set is 98.89%; for fresh and sub-fresh salmon, the spectral information were treated with SNV, the feature spectra were associated with their corresponding TVB-N using Si-PLS, the Si-PLS model correlation coefficient of the validation set is 0.864 1, the Si-PLS model recognition rate of the testing set is 90.00%. According to research results, using combination of NIR spectroscopy and chemometrics, genuine and counterfeit salmon, fresh and frozen-thawed salmon, as well as fresh and sub-fresh salmon, can be detected quickly and non-destructively, thus realizing the rapid and multi-index detection of salmon quality.