Optimizing Near-infrared Spectroscopy Model

Building a stability and accurate model is the key to NIR.At present,the main difficulties for establishing model have been the following: spectral data quantity has been more and more,traditional modeling algorithm was difficult to interpret physical chemistry information of spectroscopic data,meanwhile,instability model was bad for model-sharing.In this paper,some important issues was studied.JADE-ELM algorithm was put forward for optimizing NIR model,which used to establish quantitative analysis model and pattern recognition model.Specific content as follows:(1)JADE was proposed to extract the elementary spectroscopic signal.JADE used fourth-order moment of signals to obtain the latent variables.ICs retained amplitude and phase information of initial data were used directly to analyze content.At the same time,JADE reduced the interference between each ICs.To determine the optimal number of independent component to be extracted,the Random JADE by blocks was used.ICs contained useful information without noise signal.The extracted ICs were interpreted by practical guide to interpretive NIR.(2)Model by ELM was established.In this method,as long as determined the number of neurons,then neurons parameters could be randomly generated.It was helpful for improving the robustness of the model in the model sharing work.(3)In the application of quantitative analysis and pattern recognition,JADE-ELM algorithm was used to establish the multi-component analysis model of wheat in this paper.JADE was utilized to get the hybrid matrix from basic spectrum.The hybrid matrix could reflect the content of constituents and as input of ELM to build the analysis model.Analysis results show that JADE-ELM model's correlation coefficients of water,crude fat and protein respectively reached 0.96,0.91 and 0.93 which are more closer to the true value than PCA?SVM.JADE-ELM algorithm was used to identify the maturity degree of Chinese pears.The influence of the different pre-processing methods were discussed.After comparing a lot of pretreatment,MSC and WT were chosen.Finally,got a model which prediction accuracy was 95.58%.(4)Developed NIR analysis software based on JADE-ELM algorithm.This software contained two parts,one of them used for quantitative analysis and another used for pattern recognition.Both were realized to modeling and prediction.It was reliable and suitable for online detecting.