Study On Detection Method Of Milk Protein Content Based On Hyperspectral Imaging Technology

In the evaluation of nutrients in dairy products,milk is an essential source of protein absorption in people’s daily life,and the more important index is protein content.In recent years,the health of consumers and the development of dairy industry are closely related to the quality of milk.Therefore,the detection of milk protein content is a very important link.The traditional detection methods consume a long time and waste a lot of human resources,and will also lead to the deterioration of the environment.Therefore,to find a more rapid and accurate detection method of milk protein content is of great significance to dairy quality inspection.Therefore,this thesis uses machine learning combined with hyperspectral imaging technology to quantitatively evaluate the milk protein content,so as to provide a feasible scheme for the detection of milk protein content in the market.The specific research work and conclusions are as follows:(1)Genetic Fisher discriminant analysis(GA-Fisher)was used to select the characteristic wavelength of milk spectrum,scan the milk samples with visible / near infrared hyperspectral imaging system to obtain the hyperspectral images of various brands of milk,and then extract the region of interest(ROI)from the hyperspectral images of milk with envi software to obtain the spectral data.The milk spectral data were preprocessed with maximum and minimum normalization,MSC and S-G smoothing respectively,and PLSR modeling was performed on the preprocessed data.MSc was used to preprocess the spectral data,and the model score was the highest.Taking the milk spectral data pretreated by MSC as the research object,the GA Fisher algorithm is used to screen the characteristic bands of milk spectrum.The SVR model is established with the full spectral data and the characteristic band data selected based on GA Fisher algorithm as independent variable input respectively,and compared with spa,cars and other characteristic band selection methods.The results show that GA Fisher algorithm is a meaningful feature selection method.It eliminates the redundant information of spectral matrix,simplifies the model,and can also predict the protein content of milk with high precision.(2)A nondestructive quantitative prediction method of milk protein content based on improved sparrow search algorithm and optimized error back propagation algorithm(BP)was constructed.Under the same data environment,compared with the traditional BP neural network and the traditional sparrow search algorithm optimized BP neural network,the improved sparrow search algorithm optimized error back propagation algorithm(BP)is better for the prediction of milk protein content.(3)The software system of milk protein content prediction was designed.By analyzing the system requirements and designing the relevant functions of the system,the basic functions of system user login,data preprocessing,training,prediction and so on are realized.The results show that the milk protein content prediction system can effectively preprocess the hyperspectral image data and accurately predict the milk protein content.It can basically achieve the expected function and meet the design requirements.