Application Of Hyperspectral Technology In Nondestructive Detection Of Potato Starch Content

The potato is the fourth largest crop in the world after wheat,corn and rice.In 2015,China has began to carry out potato staple food work,and potato-related industries began to develop rapidly.During the “Twelfth Five-Year Plan” period,special potato-related industries and the like have come into effect,whose productivity of processing and utilization of by-products constantly increase.In 2016,China issued the “13th Five-Year Plan” for potato processing industry.The plan points out that with the promotion and application of a series of emerging technologies,new technology upgrades will be brought about for potato breeding,planting and processing,which provides guarantees for the continuous development of its processing industry.Potato starch content will have a certain impact on the quality of its processed products.At present,the determination of potato starch is also largely relying on past chemical methods,which are complicated and inefficient,destructive and costly.They cannot meet the actual needs of the potato processing industry.As a new type of detection method,Hyperspectral imaging technology has developed gradually,because of its non-destructive,multi-band and high resolution features and so on.It has achieved remarkable results in non-destructive testing of potato and other agricultural products.However,the data collected by this method is a type of three-dimensional cube,which has excessive useless data in a large amount,making the information processing becomes time-consuming and labor-consuming,reducing the speed of model construction,and having a certain impact on the accuracy of the prediction.Therefore,improving the measurement speed and forecasting accuracy of potato starch content and exploring the treatment methods of hyperspectral information analysis can lay a better foundation for the continuous progress of the technology and promote the development of the potato processing industry.This article takes the potato in Keshan County of Tsitsihar,Heilongjiang Province as the research object.Using hyperspectral analysis,data acquisition,mathematical statistics and other means to analysis systematically and deeply in the spectral pretreatment model construction,pattern in the detection of potato starch content,recognition and a series of processing methods.Combining appropriate pre-processing methods,model construction and pattern recognition methods to improve the prediction accuracy of starch content,improving the detection speed and accuracy of detection by determining the appropriate characteristic wavelength selection algorithm.As follows here are the main contents and conclusions:(1)The effect on the model of potato starch sample between spectral pretreatment and modeling methods was compared.Potato samples were established by threes models such as the PCR,PLSR and SVR.Meanwhile,the method of smoothing 13 points,first and second derivatives,SNV,and MSC were used to calculate the results,through which to make the contrasts with the modeling results of the original spectrum.The results of the present study indicated that PLSR was the optimum detection model of potato starch content.The optimal model was obtained by the model construction effect of smooth 13 points.As the number of principal components was 12,the R ?RMSEC?R and RMSEP of optimal starch content detection model were obtained respectively.They were 0.8324,0.4487%,0.8279 and 0.3977% in sequence.(2)The research studied potato by the characteristic wavelength selection method for the model of starch content.The performance of wavelength selection of hyperspectral features of six samples including GA,UVE,CARS,SPA,Random-frog,and iRF leaping were compared.It is proved by experiments that the Random-frog algorithm works best.