低场核磁共振结合化学模式识别方法判别休闲豆干品牌

Distinction of leisure dried tofu brands by using chemical pattern recognition combined with low field nuclear magnetic resonance

为了探求一种能快速准确地判别豆干品牌的方法,本研究采用低场核磁共振仪,对休闲豆干样品进行测量获取横向弛豫数据,结合主成分分析法(principal component analysis,PCA)、偏最小二乘判别分析(partial least squares-discriiminate analysis,PLS-DA)和贝叶斯正则化误差反向传播人工神经网络(bayesian regularization back-propagation artificial neural network,BR-BP-ANN)等化学模式识别方法对试验数据进行模式识别分析。选用4个常见的休闲豆干品牌,每个品牌分别收集5个批次的样品。每个批次随机选择16小包作为测试样品,共获得320个样品。使用低场核磁共振仪对这些样品进行测量,然后采用模式识别方法进行品牌判别。试验结果表明:对预测集豆干样品采用PCA进行判别分析时,从三维投影图中难以对各品牌进行人眼识别;运用PLS-DA方法对训练集样品的品牌识别率为86.3%,预测集样品的识别率为81.3%;然而使用BR-BP-ANN方法对预测集样品进行判别预测,预测值与实际期望值高度吻合,判别正确率均为100%,能够很好的实现对豆干品牌的判别。因此,采用BR-BP-ANN方法能够快速而准确地对豆干品牌进行识别,可为休闲豆干的品牌判别提供较好的技术支持。

By using low field nuclear magnetic resonance （LF-NMR） spectrometer, the leisure dried tofu samples were measured for obtaining transverse relaxation data. The experimental data were analyzed by pattern recognition methods including principal component analysis （PCA）, partial least squares-discriminant analysis （PLS-DA） and Bayesian regularization-back propagation-artificial neural network （BR-BP-ANN）. The research purpose is to search for a method which can quickly discriminate the brand of leisure dried tofu. The 4 common dried tofu brands were chosen and the samples of each brand were collected from 5 batches of leisure dried tofu. Sixteen small bags of dried tofu were randomly selected as study samples from each batch. In this way, 320 samples were obtained. These samples were measured with an LF-NMR instrument. Each sample was measured repeatedly 3 times. The average value of them was taken as the result of the measurement. A total of 320 sample spectra were obtained. Then, 60 samples were selected randomly as training set from 80 samples of each brand. A total of 240 training samples were randomly acquired in this manner, and the remaining 60 samples were used as prediction set. These samples of 4 brands were used for rapid distinction by pattern recognition methods. The experimental results showed that each brand was difficult to be picked out with eyes by three-dimensional PCA scoring plot. Further, measurement data of dried tofu were treated by PLS-DA. The results displayed that the recognition rate of Brand 4 was the highest, 93.3%, and the worst was the recognition rate of Brand 3, 76.7%, and the total recognition rate of all brands was 86.3% for the training set. For the prediction set, the recognition rate of Brand 4 was also the highest, and the recognition rate of Brand 2 was the worst, 75%, and the total recognition rate was 81.3% for all brands. However, the BR-BP-ANN method can discriminate 4 bands simultaneously. The predicted value is in good agreement with the actual expected value for training set. In other words, the prediction values were highly consistent with the expected values of 1-0-0-0 for No. 1-60 samples, which met the criteria. So these samples belonged to Brand 1. Similarly, the prediction values were in agreement with the expected values of 0-1-0-0 for No. 61-120 samples. Then these samples were classified into Brand 2. In the same way, No. 121-180 samples were taken into account, and the prediction values were close to the expected values of 0-0-1-0. Consequently, these samples belonged to Brand 3. Similarly, No. 181-240 samples belonged to Brand 4 since the prediction values were similar to the expected values of 0-0-0-1. In the same way, good results were also obtained for prediction set since the predicted values were in perfect agreement with the expected values. It＇s interesting to note that the correct rate of prediction was 100% for brand discrimination. So the combination of LF-NMR and BR-BP-ANN can provide a fast and accurate method and better technical support for the brand discrimination of the leisure dried tofu.