几种模式识别方法在豆酱近红外光谱鉴别中的应用

The application of pattern recognition in soybean paste identification by near infrared spectrum technique

为建立快速鉴别豆酱品质的适用方法,利用近红外光谱技术,针对市场上常见的四类豆酱样本,分别进行异常样本剔除和预处理,采用判别偏最小二乘法(DPLS)、相似分析法(SIMCA)和误差反向传播神经网络(BPANN)定性模式识别方法,进行豆酱类别的识别研究。研究结果表明:3种识别方法的校正集正确识别率分别为99.10%、98.20%、100%,预测集识别率为94.55%、89.09%、90.91%。综合比较3种不同的识别算法,采用判别偏最小二乘法(DPLS)时校正集和验证集的正确判别率效果都较好。研究结果表明采用近红外光谱分析技术实现豆酱的快速准确分类和鉴别是可行的。

Four soybean paste samples were selected to establish a rapid identification method by infrared spectrum technique. Abnormal samples were discarded and all samples were through pre-treatment. Discriminant Partial Least Squares （DPLS）. Soft Independent Modeling of Class Analogy （SIMCA） and Back Propagation Neural Net （BP-ANN） were used to identify four kinds of Soybean Paste. The results showed that the classification rate of calibration was 99.10% .98.20% and 100% ; the validation rate was 94.55% .89.09% .90.91% respectively. By comparing three recognition algorithms, DPLS was superior to the other two on calibration and validation rate. The study showed that the near infrared spectroscopy technology is a feasible way for the soybean paste classification.