| As we all know,corn is one of the three major food crops in the world and plays an important role in production and life.It can be used as a raw material for food processing,as well as animal feed and industrial raw materials.Corn has the characteristics of large output and wide distribution of planting area,and has always occupied an important position in China's agricultural production and national economy.Therefore,how to accurately detect and classify the quality of corn has become a meaningful question.Near-infrared spectroscopy technology has developed rapidly in recent years,and it is becoming more and more mature.This technology not only involves the content of the field of spectrometry,but also usually combines the professional knowledge of multiple disciplines such as chemometrics and computer science.Near-infrared spectroscopy technology is widely used because it is an indirect measurement method.When it is used to detect the content of sample components,it does not need to destroy the sample or consume other chemical reagents to measure the sample.The environment causes damage,and the accuracy of the measurement is still very high.If the latest methods and technologies in the computer field are applied to near-infrared technology,usually good analysis results can be obtained.Since the near-infrared spectroscopy analysis technology is an indirect analysis technology,to achieve quantitative analysis of unknown samples,it is necessary to establish a proper calibration model.The thesis uses widely used partial least squares.At the same time,there are many general variables in spectral data,and there is a lot of redundancy.If full-spectrum modeling is used,it will lead to the problem of increased prediction error and poor stability.Therefore,how to reduce the number of variables and select the optimal combination of variables to build a model has become a key issue.Based on the Genetic Algorithm Synergy Interval Partial Least Squares(GA-siPLS),the Multigroup Parallel Genetic Algorithm Synergy Interval Partial Least Squares(MPGA-siPLS).The idea is to divide the full spectrum into several intervals continuously and uniformly.Calculate the root-mean-square error of the PLS model for each interval separately,select several intervals with a smaller root-mean-square error to form a new spectral database,then use multiple population genetic algorithms for wavelength selection,and finally use the selected wavelength combination to establish the model.The paper uses a public database to verify the proposed algorithm.This database is composed of 80 corn samples.In order to collect data,three different NIR spectrometers are used to measure the samples.The entire spectral range is 1100-2498 nm,and the spectral resolution is 2nm,that is,each sample has 700 spectral data.The characteristic values of each sample include moisture,oil content,protein content and starch value.The thesis quantitatively analyzed several main components in corn with the newly proposed method,and GA-siPLS,Genetic Algorithm Partial Least Squares(GA-PLS),full spectrum partial least squares(PLS)And Multigroup Parallel Genetic Algorithm Partial Least Squares(Multigroup Parallel Genetic Algorithm Partial Least Squares,MPGA-PLS)and other methods were compared,the experimental results show that:(1)Select the interval first,not only can reduce the number of wavelength selection,save calculation time,but also the prediction accuracy will be better;(2)The use of multi-group genetic algorithm for wavelength selection not only converges faster,but also results in a smaller root mean square error,that is,multi-group genetic algorithm for frequency selection has obvious advantages;(3)The multi-group genetic algorithm combined with the interval partial least squares method combines the advantages of the two,exerts the advantages of the two,and obtains better prediction results. |
PDF Full Text Request