基于近红外光谱的冻融猪肉糜鉴别模型研究

Study on discrimination model of frozen-thawed minced pork based on near infrared spectroscopy

中文关键词:  猪肉糜  冻融次数  近红外光谱  子空间判别  混淆矩阵

英文关键词:minced pork  freeze-thaw times  near infrared spectroscopy  subspace discriminat  confusion matrix

中文摘要:

      目的 建立基于近红外光谱的定性分析模型, 实现对冻融猪肉糜的判别。方法 采用近红外光谱分析技术对新鲜猪肉糜和不同冻融次数猪肉糜分别进行无损鉴别, 建立了窄神经网络(narrow neural network, NNN)、线性判别(linear discriminant, LD)、支持向量机(support vector machine, SVM)和子空间判别(subspace discriminant, SD) 4种不同的判别模型, 并对所建立的模型性能采用正确判别率、混淆矩阵(confusion matrix, CM)、受试者工作特征曲线(receiver operating characteristic curve, ROC)和曲线下面积(area under the curve, AUC) 4个指标进行评价。结果 基于SD建立的判别模型较优, 其预测集正确判别率为96.2%, 高于基于LD (94.3%)、NNN (79.0%)和SVM (54.8%)所建的判别模型正确判别率, 并且其CM、ROC和AUC均显示基于SD所建判别模型对于冻融猪肉糜分类的优越性。结论 本研究建立的近红外光谱技术结合SD模型对冻融猪肉糜的鉴别能力较强, 可为工业化在线检测方法的开发提供技术支撑。

英文摘要:

      Objective To establish the qualitative analysis model by near infrared spectroscopy and realize the classification of frozen-thawed minced pork. Methods Near infrared spectroscopy was used to identify fresh minced pork and minced pork with different freeze-thaw times, and 4 kinds of discriminant models, narrow neural network (NNN), linear discriminant (LD), support vector machine (SVM) and subspace discriminant (SD) were established, separately. The performance of the models was evaluated by 4 indicators: Correct discriminant rate, confusion matrix (CM), receiver operating characteristic curve (ROC) and area under the curve (AUC). Results The model based on SD was superior, and the correct discriminant rate of its prediction set was 96.2%, which was higher than that of LD model (94.3%), NNN model (79.0%) and SVM model (54.8%), respectively. Moreover, the CM, ROC curve and AUC of the SD model exhibited the superiority in the classification of frozen-thawed minced pork. Conclusion In this study, the established method that applied NIRS technology combined with SD model illustrates a strong ability to identify the frozen-thawed minced pork, which can provide technical support for the development of industrial on-line detection.