Study On Detection Method Of Vegetable Quality Based On Hyperspectral Analysis

Head cabbage,potato and minituber were taken as the research objects and the non-destructive acquisition of quality information of those vegetable were conducted based on spectral technology.The research result provides theoretical and technical support for the specification,grade,external defect detection and variety identification of those agricultural products.The main contents of this paper are as follows:[1]Appearance quality detection of cabbage based on machine vision technologyCombining machine vision technology with BP neural network,a BP neural network recognition model for shape of head cabbage was established.The head cabbage has three types according to its external ball shape,tip shape type,flat shape type and round shape type.The traditional identification method of cabbage ball shape is artificial method.This paper proposes a new method for rapid identification of cabbage ball shape using machine vision technology combined with BP neural network.Firstly,extract four absolute cabbage shape parameters such as heightr width,long axis,and area based on image processing technology.Define five relative shape parameters based on the above absolute parameters,which were ratio of height to width,circular degree,rectangle degree,ellipse degree,and dome shape index.These nine parameters describe the cabbage shape.Since the parameter ranges overlapped,the individual parameter had not separating classification ability.Secondly,three recognition models of cabbage ball shape with BP neural network were established using three types of input datasets,four absolute parameters(long axis,height,width,area),five relative parameters(ratio of height to width,circular degree,rectangle degree,ellipse degree,dome shape index),and all above nine parameters.The test results showed that the prediction accuracy of BP model taking four absolute parameters as the input was 62.5%,and the prediction accuracy values of other two models were 100%.Combining machine vision technology with fuzzy cluster analysis,a grading model of head cabbage was established.In the People's Republic of China National standard NY/T 1586-2008 Brassica oleracea grade specification,head cabbage was divided into three grades.Using image processing technology,extract the shape,color and texture characteristic parameters of head cabbage.Based on above parameters,divide the calibration set samples into three class levels by fuzzy clustering analysis.Based on the results of cluster analysis,by analyzing the average values and variance values of shape,color and texture feature parameters of the three classes,determine the correspondence between classes and grades,the third classification corresponds to super grade head cabbage,the second classification corresponds to the first grade head cabbage,the first classification corresponds to the second grade head cabbage.By calculating the Euclidean distances between the tested head cabbage and every grade samples centers,the grade of the tested sample was determined by the minimum distance.[2]Intrinsic quality detection of cabbage based on near infrared spectroscopyNear infrared spectroscopy(NIRS)combined with multiple regression and partial least squares regression(PLS)were used to predict vitamin C content in head cabbage.The head cabbage as one of the main vegetable is rich in vitamin C.The first derivative spectral pretreatment method could well improve modeling accuracy;it is the best spectral data pretreatment method in this experiment.For the PLS regression model based on seven principal components,PLS regression model can be used to predict the content of vitamin C with high precision,which can replace the traditional detection methods and provide a new way for the quality evaluation of cabbage.The wavelengths were selected by stepwise regression method,the linear regression models corresponding to 8 6 and 5 selected wavelength variables,which meant that using less wavelength variables to predict Vc content in head cabbage was practical.Multiple linear regression models performed with 8,6 or 5 selected wavelength variables,with the wavelength is less,the R2 decreased,but could provide technical support for the development of portable testing instrument.Near infrared spectroscopy(NIRS)combined with competitive adaptive reweighted sampling(CARS)method and partial least squares regression(PLS)method,a prediction model of soluble sugar in cabbage was established.Select eighty-four wavenumbers by Competitive Adaptive Reweighted Sampling(CARS)method.The CARS algorithm reduced the modeling variables so that the complexity and the accuracy of CARS-PLS model were improved.The wavenumbers selected can introduce both the spectra related components information and the spectra related the background information to improve the adaptivity of the calibration model.The prediction model of soluble sugar content in head cabbage was practical.[3]Detection of potato diseases based on hyperspectxalThe midpoint between the extreme points and the extreme points on the spectral curve,these key points have "fingerprint" effect.Finding the key points of the average spectral curve of normal,dry rot and scab potatoes,then a standard pattern sequence is formed based on the reflectance corresponding to the key points on the average spectral curve.By calculating the mahalanobis distances between the corresponding pattern sequences and the three standard pattern sequences of the samples to be tested,the classification of the samples to be tested is determined by the minimum distance,the correct recognition rates of normal,dry rot and scab potato samples were 100%.The spectral curves of normal,dry rot and scab potato samples have six same extreme points,the trend of curves between extreme points is basically the same,but samples fluctuate value is not same at extreme points.The fluctuation of the waveform reflects the change of the internal material,which can be reflected by the slope of the line between the adjacent two points.Different combinations of the six same wavelengths are made,wavelength combination(911,1269,1455)the slope of every two adjacent points forms a standard pattern sequence,the minimum mahalanobis distance is used to determine the classification of the samples to be tested.The correct recognition rate of normal and scab samples was 100%,and the correct recognition rate of dry rot samples was 97.6%.[4]Classification and detection of potato micro-seed potato based on hyperspectralUse the first three principal components as categorical variables.Use linear discriminant analysis,nonlinear discriminant analysis,BP neural network and support vector machine to study the classification and recognition of Atlantic,Holland-14,Ho land-15041,Holland-15Q8,Jizhangshu-12,Jizhangshu-8,Xingjia-2 and Y-2.Because of the large number of varieties,establish a single classification model for eight varieties,and the correct recognition rate is poor.Adopt hierarchical classification modeling to improve the correct recognition rate.Divide the judgment process into three layers...