Research And Software Implementation Of Milk Powder Analysis Model Based On Mid Infrared Spectroscopy

It is of great significance to conduct the quick quantitative determination for the main nutrients in milk powder. Compared to the traditional detection method, infrared spectroscopy technology is characterized by the simple operation, low cost and fast speed. Conducting the qualitative and quantitative analysis through the back calculation of infrared spectrum is the effective means to detect the milk powder rapidly. This paper builds the common IR spectrum library of milk powder, establishes the inversion model using the data in spectrum library and verifies the model, analyzes the validity of unknown milk powder. On this basis, it develops the web-based analysis software for the qualitative and quantitative analysis. The main research results in this paper is as follows:1. Design the structure of IR spectrum library of milk powder, build the MIR spectrum library of milk powder, and input the records of 303 kinds of milk powders, including the information of MIR spectrum and milk powder properties.2. In order to judge the matching degree of the milk powder to be detected and that in the library, this paper builds the similarity detection model, compares the extracting principal component, differential, standard normal variable, multiplicative scatter correction and wavelet transform and many other pretreatment methods, compares the modeling effect of correlation coefficient method, included angle cosine method and Euclidean distance method. The results show that the matching degree of spectra feature extraction of milk powder based on wavelet transform and the correlation coefficient method is the best, and the accuracy rate is up to 90%, better than other methods.3. This paper establishes the analysis model for the essential nutrients, including quantification of protein, fat and carbohydrates, and tests several pretreatment methods and the regression methods. The results show that the effect is the best that the standard normal variable is used to preprocess feature extraction of wavelet transform and extract the first 17 principal components, and the regression modeling of support vector machine combined with a genetic algorithm is adopted for the quantitative calculation of protein content. The correlation coefficient of prediction set is 0.9741, and RMSECV is 1.1846. The results show that the effect is the best that the standard normal variable is used to preprocess feature extraction of wavelet transform and extract the first 20 principal components and the partial least squares is adopted for the quantitative analysis of fat. The correlation coefficient of prediction set is 0.9773, and RMSECV is 1.0526. The results show that the effect is the best that the standard normal variable is used to preprocess feature extraction of wavelet transform and extract the first 18 principal components and the partial least squares is adopted for the quantitative analysis of fat. The correlation coefficient is 0.9028, and RMSECV is 2.4934.4. Write the intermediate infrared spectroscopic analysis software of milk powder, software integration similarity detection model, and quantitative analysis model and spectra library of milk powder, and publish by means of web application.This paper builds the matching model with the known milk powder,and the calculation model of essential nutrients in milk powder based on the intermediate infrared spectrum, and develops the web-based analysis software that can be used in the on-site rapid qualitative and quantitative analysis of milk powder.