Identification Of Rice Varieties Based On Hyperspectral Image

Rice is one of the most important food crops in the world and the main ration in China.The nutritional value and processing characteristics of different rice varieties are not the same.The identification of different rice varieties is a realistic need to improve the living standards of residents and develop high-quality grain projects.As a comprehensive high-tech,hyperspectral imaging can detect the spectral reflectance of each point in the ilage while acquiring the external image information of the target object.It can simultaneously acquire spatial image and spectral information of objects and has the advantage of union of imagery and spectrum.It has the advantages of lossless,fast and high recognition accuracy.In this paper,a total of 400 samples of 10 rice varieties were taken as research objects,and hyperspectral images at 400-1000 nm were collected to identify rice varieties.The specific research contents are as follows:(1)Selection of regions of interest and comparison of preprocessing methods:The range of 30*30 pixe12,20*20 pixel2 and 5*5 pixel2 were selected as regions of interest(ROI)in the lemma,palea and rachilla regions of rice.The reflectance difference was as follows:lemma>palea>rachilla.The original spectral data in ROI region are pre-processed by S-G filtering derivation,MSC derivation and SNV derivation respectively.The waveform fluctuation amplitude after processing was as follows:SNV derivation>MSC derivation>S-G filtering derivation.(2)Characteristic extraction of rice varieties:the thresholding method was used to separate the target from tlie background,and the morphological parameters of each grain were calculated as follows:area,circumference,length of long axis and length of short axis.Meanwhile,principal component analysis(PCA)was used to analyze the reflectance data.In 400-1000 nm band,it was found that the wavelength with the largest weight coefficient was located at 426 nm,512 nm,640 nm,707 nl,790 nm and 860 nm and used as the characteristic wavelength.Loading the texture image from a single wavelength image at 790 nm,the texture characteristic para1eters of each rice sample were calculated as follows:Mean,Variance,Homogeny,Contrast,Entropy and Correlation.The results of variance analysis showed that P<0.05.These characteristics could be used to identify rice varieties.(3)Establishment of rice variety identification model:Based on the on the spectral,image and fusion characteristics,the Fisher discriminant analysis model,partial least squares regression(PLSR)mode and artificial neural network model(ANN)were established respectively for rice variety identification.The accuracy of rice variety recognition was compared by using different characteristics of the model.The results show that the recognition model of the ANN model based on the fusion feature is as high as 99.80%,which is superior to the Fisher discriminant analysis model and the PLSR model in classification accuracy,and can be used for the good identification of rice varieties.The contribution of each characteristic parameter to the identification of rice varieties and the relationship among rice varieties were further analyzed.