论文部分内容阅读
提出了一种衰减全反射红外光谱法快速分类和识别多种食用油的方法——KL-BP模型。此模型利用KL算法对原始光谱数据分类特征进行提取并对原始数据降维,降维后的数据作为神经网络的输入建立分析模型。实验共收集了九种食用油包括芝麻油、玉米油、油菜籽油、调和油、葵花油、花生油、橄榄油、大豆油、茶籽油,共84个样品,并测定了其衰减全反射红外光谱。为了对比所提方法性能,分别建立PCA直接分类、KL直接分类、PLS-DA、PCA-BP和KL-BP模型的分类结果进行对比。研究结果表明,对所研究的9种食用油,PCA直接分类、KL直接分类、PLS-DA、PCA-BP和KL-BP方法的识别率分别为59.1%,68.2%,77.3%,77.3%和90.9%。在数据降维中,KL算法通过分别提取使类间距离和类内距离比值最大方向的特征向量提取和包含在类内离散度矩阵中的分类信息,能够比PCA方法提取了更多的分类信息;引入BP神经网络能有效地提高分类能力和分类准确率;KL-BP综合了KL对分类信息提取优势以及BP神经网络自学习、自适应、非线性的优点,在分类和识别成分相近的9种食用油中表现出了最优秀的能力。
A method of rapid classification and identification of a variety of edible oils by attenuated total reflection (FTIR) spectroscopy was proposed - KL-BP model. In this model, the KL algorithm was used to extract the classification features of the original spectral data and to reduce the dimensionality of the original data. The dimensionality reduction data was used as the input of the neural network to establish the analytical model. A total of 84 samples of nine edible oils including sesame oil, corn oil, rapeseed oil, blend oil, sunflower oil, peanut oil, olive oil, soybean oil and tea seed oil were collected and their total reflection infrared spectra . In order to compare the performance of the proposed method, the classification results of PCA direct classification, KL direct classification, PLS-DA, PCA-BP and KL-BP models were compared respectively. The results showed that the recognition rates of the nine kinds of edible oil, PCA direct classification, KL direct classification, PLS-DA, PCA-BP and KL-BP methods were 59.1%, 68.2%, 77.3% and 77.3% 90.9%. In data dimensionality reduction, KL algorithm can extract more classification information than the PCA method by extracting the feature vectors that maximize the distance between classes and the distance-to-class distance ratio and the classification information contained in the in-class dispersion matrix, respectively ; The introduction of BP neural network can effectively improve the classification ability and classification accuracy; KL KL combines the advantages of classification information extraction and BP neural network self-learning, self-adaptive, non-linear advantages in the classification and identification of similar components 9 The kind of cooking oil showed the best ability.