Research On Nondestructive Prediction Method Of Grain Quality Based On Visible Near Infrared Spectroscopy

Grain is the foundation of human survival and development,and grain security is related to the overall situation of national security.How to ensure the quality of grain and store it safely has always been a problem that the country attaches great importance to.Near-infrared spectroscopy analysis technology is a non-destructive detection technology for grain quality that has developed rapidly in recent years.It has the advantages of not damaging the grain itself,convenient operation,high stability and high efficiency.It analyzes the spectral data of the grain and establishes the relationship model between the spectral data and the grain composition,so as to realize the prediction of the quality of the unknown grain sample.The core of near infrared spectroscopy analysis technology is to establish a stable and accurate prediction model.Because near-infrared spectral data has the characteristics of high dimensionality,collinearity and information redundancy,the analysis of near-infrared spectral data starts from spectral dimensionality reduction,and then establishes a suitable regression prediction model.Therefore,in order to better solve the analysis and research problems of nearinfrared grain spectral data,this thesis focuses on the two key technologies of spectral dimensionality reduction algorithm and establishment of regression prediction model,and makes the following work:(1)Firstly,collect and store grain visible near-infrared spectral data through spectrometers and other equipment.Through the analysis of the spectral data,it is found that the spectral data has the characteristics of high dimensionality,collinearity and information redundancy.Through literature research,understand and organize the currently used spectral dimensionality reduction algorithm and quantitative regression model in the field of near infrared spectroscopy to prepare for the subsequent specific research;(2)Secondly,in view of the difficulty of dimensionality reduction of grain nearinfrared spectroscopy data,this thesis proposes a dimensionality reduction method based on optimized spa and a dimensionality reduction method based on optimized sparse autoencoder SAE?GA from two aspects of feature selection and feature extraction.The idea of dimensionality reduction method based on SPA optimization is to reduce the randomness of spa algorithm by calculating the purity value of each spectral feature in advance,optimizing the initial band,and then screening the spectrum;The idea of SAE?GA dimensionality reduction method is to use sparse autoencoder to learn the spectral features,mining the non-linear relationship of spectral features,and then combine the prediction output with genetic algorithm to further select features.Through experiments on two sets of near-infrared data of grain,comparing the accuracy of traditional successive projections algorithm,principal component analysis algorithm and single sparse autoencoder algorithm on regression model,the effectiveness of the method proposed in this thesis is proved;(3)Finally,in view of the traditional methods of near-infrared spectroscopy modeling,with the increase of the number of samples,the adjustment of model parameters is tedious,which can not meet the needs of mass sample detection,and the generalization performance is poor,this thesis proposes a regression prediction model based on the optimized convolutional neural network.The model is applied to the nearinfrared spectral data of grain with the proposed dimensionality reduction algorithm,and a grain quality prediction model based on the regression CNN is established.Compared with the traditional modeling method,it has the advantages of high accuracy and strong generalization.