基于支持向量机的新鲜与解冻许氏平鲉(Sebastes schlegeli)近红外光谱鉴别技术

Distinguishing the Fresh from the Frozen-Thawed Sebastes schlegeli Using the NIR Spectroscopy Identification Technology Based on Support Vector Machine

中文关键词: 近红外光谱  许氏平鲉  解冻  主成分分析  支持向量机  鉴别

英文关键词: NIR  Sebastes schlegeli  Frozen-thawed  PCA  SVM  Identification

中文摘要:

      本研究建立了一种近红外光谱技术，用于鉴别鱼类是否经过解冻处理。首先测定了120个样品的近红外光谱，通过主成分分析对原始光谱数据进行降维处理，再结合支持向量机建模进行分类鉴别。对所有建模样品的主成分1和2按得分值绘制得分图，进行分析聚类，并将前10个主成分的得分值作为支持向量机的输入，优化惩罚参数c和核函数参数g，对90个样本训练；用未知的30个样本进行判别验证，建立鉴别鲜、冻许氏平鲉的支持向量机分类模型，预测准确率达100%。研究表明，近红外光谱技术结合主成分分析和支持向量机可以作为一种简便、快速、准确的方法用于判断鱼类是否经过解冻处理。

英文摘要:

      Frozen fish are usually less desirable than the fresh counterparts on the market. As a result, it has become a common issue that the frozen-thawed fish are disguised as fresh for a higher price. In this study the near infrared (NIR) spectroscopy was employed to separate the frozen-thawed fish from the fresh. One hundred and twenty prepared Sebastes schlegeli samples including 60 fresh and 60 frozen-thawed were scanned using a near-infrared spectroscopy system between 10000–4000 cm-1 wavenumbers. The working model was based on the fact that the average NIR spectra of fresh fish were distinctive from the frozen-thawed and possessed certain characteristics and fingerprint resistance. Principal component analysis (PCA) was used for the dimension reduction of the spectra data. The first two principal components (PCs) explained over 98% of variances in all the spectral bands. Clustering was performed and analyzed based on the first two PCs of all samples. The principal component score plot demonstrated that the fresh (above the X axis) and the frozen-thawed samples (below the X axis) were well separated, and that the distribution of fresh samples was dispersal. In order to improve the accuracy of prediction, the support vector machine (SVM) classification model was developed to differentiate the fresh fish from the frozen-thawed, based on principal component analysis scores. The score values of the first ten PCAs were used as the input variables of the SVM, and the penalty parameter c and kernel function parameter g were optimized. Ninety samples were used for building the SVM model. This model was then applied to predict the rest 30 unknown samples, and the prediction rate was 100%. These results suggested that the near infrared spectroscopy combined with principal component analysis and support vector machine could be used as a rapid, simple and reliable method to identify the fresh and frozen-thawed fish.