Research On Detection Method Of The Wheat Seed Quality Based On Spectral Imaging Technology

Quality evaluation is an important technical support in the process of wheat production,acquisition,storage and processing.Therefore,how to make use of fast and efficient detection method has great significance on directing planting structure,purchasing price by quality and reasonable utilize and so on.There are two traditional quality detection methods: sensory test and chemical analysis method.But because of the subjective factor effects,the high work intensity and the complex operation,the two are all difficult to meet the demand of the current seed market.The machine vision technology and near infrared spectroscopy technology(NIRS),which are widely researched at present,exist the limitations of low model reliability,unable to realize the single grain detection and the chemical component distribution analysis.The spectroscopy imaging technology is the perfect combination of the spectroscopy technology and imaging technology.It could obtain both the image information at different wavelength points and the spectroscopy information at different spatial pixels at the same time.Through the analysis of the image and spectral data,we could realize the sample components' content,spatial distribution and dynamic change,which provides the basis for spectral imaging technology to detect the seed quality comprehensively.In this paper,the wheat seed is selected as research object,the research work of wheat seed quality detection is carried out by space scale change from “multiple wheat grains” to “single wheat grain” to “wheat grain slice” based on the spectral imaging technology combined with chemo-metrics analysis methods.The main research contents are as follows:1.Multiple wheat grains quality analysis based on near-infrared hyper-spectral imaging technology.Moisture?protein and wet gluten content are the key indicators reflecting the quality of wheat seed.In this paper,the partial least squares(PLS)hyper-spectral quantitative models of the three indicators above are built separately based on the best spectral preprocess methods and feature spectrum intervals.And by comparing the hyper-spectral models with the NIR models,we find that the performance index of the hyper-spectral model is better than that of NIR model.2.Single wheat grain protein content detection based on hyper-spectral imaging technology.The single wheat quality is analyzed in two sides: the protein content prediction and the unsound grain recognition.Synergy interval partial least squares is applied to optimize the average model of wheat seed protein content.Then,the average model is applied to predict the protein content of the single wheat grain protein content.In order to realize quick and accurate identification of the unsound kernel in wheat,a novel method is studied based on Visible-near infrared hyper-spectral imaging technology and multi-class support vector machine.3.Fine analysis of the wheat grain slice based on the Raman imaging technology.In this paper,the wheat grain waist slice is selected as the research object,and the DXR confocal micro Raman spectrometer is employed to collect the samples' Raman image.A new wheat seed coat thickness measurement method is developed by using the spectrum difference of seed coat combined with the image processing technology.In addition,the distribution diagram of three main components(starch,cellulose and protein)is achieved by using the spectrum stripping method based on the full band or characteristic band.In a word,the fine analysis of the wheat grain slice is realized.