基于二维相关谱灰度统计特征判别掺杂尿素牛奶

Detection of Adulterated Milk Based on the Gray Statistical Characteristic of 2D Correlation Spectrum

如今,乳制品掺杂问题日益突出,为寻求一种快速高效的鉴别掺杂牛奶的方法,提出将二维相关谱和灰度统计特征分析技术应用于判别掺杂牛奶的研究。运用MATLAB软件绘制出掺杂牛奶和纯牛奶的二维相关谱灰度图像,利用图像纹理分析中的灰度统计特征的方法提取出平均值、方差、偏度、自相关、惯性矩等11个图像灰度统计特征参数。为避免"特征维数灾难"问题的产生从而影响分类效果,采用主成分分析的方法提取出10个主要特征,实现对图像灰度统计特征的降维。设计贝叶斯分类器和BP神经网络分类器对掺杂牛奶进行判别和分类,基于二维同步-异步近红外光谱贝叶斯分类器和BP神经网络对掺杂尿素牛奶判别,其正确率分别为100%和92.31%,判别纯牛奶的正确率分别为92.3%和100%。

Recently, the problem of milk adulteration was increasingly prominent. To develop a new kind of fast and efficient method of distinguishing adulterated milk, the gray statistical characteristic of two-dimensional correlation spectrum analysis technology was applied on judging adulterated milk. The gray image of the two-dimensional correlation spectrum of adulterated milk and pure milk was obtained with MATLAB software, and exploring the method of extracting gray statistical characteristic of image texture analysis, such as the mean value, variance, self-correlation, moment of inertia, and other imaging gray statistical characteristic parameter, was studied. To avoid the characteristics of ＂dimension disaster＂ problem which affects classification effect, the principal component analysis method was applied to extract the 10 key characteristics,and the method of PAC was used to reduce the dimensions of the image gray statistical characteristic. The Bayesian classifier and BP neural network classifier were used to classify the adulterated milk and pure mink. Based on two-dimensional synchronous-asynchronous spectrum of the near infrared, Bayesian classifier and BP neural network were used to classify the milk of mixed urea and pure milk, the results showed that the accuracy of classifying mixed urea milk was 100% and 92.31%,and the accuracy of pure milk was 92.3% and 100%.