Optimization Of The Partial Least Square Method For The Factor And Its Application In Determination Of Fructose Content Based On Hyperspectral Technology

Partial least squares regression method is composed of multiple linear regression,principal component analysis and canonical correlation analysis,and it can eliminate multiple collinearity caused by the number of samples far less than the number of independent variables,so it has been widely applied.Reducing the number of modeling factors while improving the prediction accuracy has an important impact on the stability of the model.Based on this expansion,we optimize partial least squares method,and verify the improvement effect of the optimization method in hyperspectral experimental data.In recent years,hyperspectral technology has developed rapidly with its abundant information and multiple processing methods.However,the massive data brought by the rich spectral information also greatly increased the complexity of data processing.Therefore,the partial least squares method of optimizing the number of factors is applied to the modeling of hyperspectral data,which can be guaranteed.It is of great significance to further simplify data even under the premise of improving accuracy.Through the analysis of the experimental results,it is found that the optimization method can not only reduce the complexity of the model,but also improve the prediction ability of the model to a certain extent.This provides a new improvement method for the application of partial least square method in chemometrics.It is worth mentioning that the model established by the characteristic band of the joint spaced partial least square method has the best prediction performance,and the optimization method has the greatest improvement to the model.In addition,the difference of fructose content was characterized by visual angle,and good results were obtained,which provided a theoretical basis for further improving the accuracy ofthe internal quality of hyperspectral detection of fruit.The main work of this paper is as follows:1.The research status and application fields of partial least squares are introduced.The characteristics,advantages and basic principles of partial least squares(PLS)algorithm are expounded in detail.The importance of partial least squares algorithm in predicting the internal quality of fruits and vegetables is highlighted,and the significance of the optimization study of partial least squares is shown.2.introducing the related basic knowledge involved in this paper,including partial least squares principle and improvement ideas,preprocessing algorithms,sample division method,band selection algorithm,while band selection algorithm includes genetic algorithm,successive projection algorithm and combined interval partial least squares algorithm.3.introducing the process and method of the experiment in detail,and the method of obtaining the spectral data and the value of the apple sugar degree is explained.We use standard normal transformation to preprocess spectral data and compare the influence of two commonly used sample diversity methods on modeling effect.Finally,the SPXY method with better performance for sample division is select.After comparing the optimized partial least squares method and the full band spectral modeling,we predict the sugar content performance of the model,and verify that the optimized partial least squares method can reduce the amount of modeling data and improve the prediction performance.4.hyperspectral data contains rich information,and there is also a high redundancy of data.Before modeling,it usually needs band selection.This paper focuses on the simplified model,reducing the amount of data and improving the modeling accuracy.So,the principle of continuous projection algorithm,genetic algorithm and siPLS are expounds to select useful spectral band,and compared with the partial least squares algorithm.It is further verified that the optimized partial least squares algorithm can reduce the amount of data on the basis of the selection of the characteristic band,and improve the prediction performance of the model.5.summarizing the main work of this paper,and the application of the optimized partial least square method is envisaged and prospected.