为实现鱼新鲜度的快速定性、定量分析。采用电子舌技术对4 ℃下不同冷藏天数的鲳鱼进行检测。同时测量鲳鱼体内挥发性盐基氮（Total volatile basic nitrogen，TVB-N）含量及细菌总数(Total viable count，TVC)。对电子舌数据进行分析处理，构建了K最近邻（K-nearest neighbor，KNN）判别模型和BP人工神经网络（Back-propagation artificial neural network，BP-ANN）模型定性评价鲳鱼新鲜度。结果显示，KNN模型的训练集、测试集识别率分别为：99.11%和98.21%；BP-ANN模型的训练集、测试集识别率分别为：92.86%和91.07%。构建了电子舌数据和TVB-N及TVC之间的支持向量机回归模型对鲳鱼新鲜度进行定量评价，独立样本检验结果显示，对TVB-N及TVC的预测，支持向量机回归模型的预测值和实测值的相关系数分别为：0.9727和0.9457，预测均方根误差分别为2.8×10-4 mg/g和0.052 log(CFU/g)。可见三种模型均能达到较好的效果。研究表明：电子舌技术在鱼新鲜度的快速定性、定量评价中具有很大的潜力。

In order to quantitatively and qualitatively evaluate the fish freshness , an electronic tongue was employed to detect the pomfret stored at 4 ℃ for different days. The total volatile basic nitrogen (TVB-N) and total viable count (TVC) of the fish samples were detected concurrently. K-nearest neighbor（KNN）model and back-propagation artificial neural network ( BP-ANN ) model were built to assess the freshness of the fish. Results showed that identification rate of training set and prediction set of KNN model were 99.11% and 98.21% respectively. While, the identification rate of training set and prediction set of BP-ANN model were 92.81% and 91.07% respectively. Support vector machine regression (SVR) model was established between the electronic tongue data and TVB-N as well as TVC for quantitative determination. The correlation coefficients between SVR predicted and measured TVB-N and TVC values were respectively 0.9727 and 0.9457, and root mean square error of prediction were 2.8×10-4 mg/g and 0.052 log(CFU/g), respectively. The overall results sufficiently demonstrate that the electronic tongue technique combined with appropriate pattern recognition method has a great potential to quantitative and qualitative evaluation of fish freshness rapidly.

国家自然科学基金资助项目（31071549）；公益性行业（农业）科技专项（201003008-04）；江苏省高校优势学科建设工程资助项目；江苏省普通高校研究生科研创新计划项目（CXZZ13_0698）