| With the increasing improvement of China’s social and economic level and the continuous improvement of people’s quality of life,the people’s demand for milk and quality requirements are gradually improving.Milk quality testing will have a profound influence on milk quality.The traditional method of milk nutrient composition detection has the disadvantages of low efficiency,high economic cost and complicated operation.Hyperspectral imaging is fast,nondestructive,easy to operate and accurate.The research of milk quality detection based on hyperspectral imaging technology has made good progress,but the existing research usually only models the single index of milk,and the applicable conditions of various spectral pretreatment and band selection methods are not clear.Based on this,this thesis uses hyperspectral imaging technology to model multiple indexes of milk at the same time,and proposes a multi-task neural network model based on automatic machine learning.The model is applied to milk quality detection system to realize milk classification and prediction of protein,fat,carbohydrate,sodium and calcium contents in milk.The main research contents of this thesis are as follows:(1)Milk hyperspectral data acquisition and extractionMilk samples,using hyperspectral image spectral imager to collect milk,write a script reads the hyperspectral images,the use of hyperspectral images in three bands of image synthesis of RGB image,choose interested in RGB image area,calculating the average area of each band reflectance of interest,to get the milk samples of hyperspectral data.(2)Milk quality detection model constructionDetermine the dataset partition scheme and classification task and nutrient content prediction task model of evaluation index,use the manual tuning parameter and automatic machine learning two methods respectively to different task to construct k-neighbor,decision tree,Ada Boost,ridge regression,single neural network and multitasking neural network model,compare different evaluation index of the model,The model that performs best.(3)Construction of milk quality detection systemAnalyze the functional requirements,environmental requirements and data requirements of the detection system,carry out the overall design and detailed design,database conceptual structure design,logical structure design and physical design,system architecture design,interface design and front-end routing design,to achieve the user interface and application of the system. |
