Study On The Prediction Of Rice Noodle Raw Material Index Content By Deep Feature Fusion

In order to produce high quality,high safety and high nutrition rice noodles products,it is necessary to find the raw material and its ratio suitable for the production of rice noodles products,and the raw material ratio can only be obtained after repeated experiments and tests,which will not only cause the waste of manpower,but also cause the waste of raw material resources.In order to find the appropriate raw materials and their ratio,this paper conducted a prediction study on the index content of rice noodles raw materials.The core of this research problem is how to improve the prediction accuracy of the index content of raw materials.To solve the problem of improving the prediction accuracy of raw material index content as much as possible,this paper proposed a deep feature fusion prediction method of rice noodles raw material index content,which could provide reference and help for the food production industry in raw material selection,raw material ratio and product quality detection.In this paper,the prediction method of deep feature fusion is carried out in the following four parts:(1)Data preprocessing and prediction model selection.The raw material index data and product index data of rice noodles were preprocessed,and then the raw material data and product data of rice noodles were preliminarily extracted.Three machine learning models were selected to predict the index content of rice noodles.(2)Feature fusion and depth feature fusion of data.Correlation analysis,factor analysis,random forest,XGBoost and other methods were mainly used in feature fusion extraction of rice noodles product indexes.Subsequently,the feature sets extracted by various methods are fused.The method of feature fusion is weighted fusion,and the depth feature fusion uses two layers of feature fusion.(3)Validation experiment of method validity.The correlation verification experiments are carried out for both the prediction method of feature fusion and the prediction method of depth feature fusion,and the experimental results are compared and analyzed to verify the reliability of the method.After confirming the validity of the depth feature fusion prediction method,it is applied to the experiment of predicting the index content of rice noodles in this paper(4)Overall optimization of the prediction process.PSO algorithm was used to optimize the prediction process.The optimization of the prediction process is divided into two aspects: the optimization of the structure of the prediction model and the superparameter,and the overall optimization of the prediction model and the fusion weights.The final prediction accuracy is further improved through optimization.After the depth feature fusion was used to predict the content of rice noodles raw materials,the RRMSE of 10 raw materials indexes was lower than 5%,and that of 4 raw materials indexes was lower than 1%.Moreover,the accuracy of prediction results is improved by about 1% when depth feature fusion is used to verify other data sets.The above results show that the prediction method of deep feature fusion can effectively predict the index content of rice noodles raw materials,and is expected to be applied in the process of raw material selection and matching in related food enterprises.