Classification Of Rice Blast Resistant Varieties Based On Near Infrared Spectroscopy

Rice is one of the most important food crops in the world.Identifying rice varieties with different resistances can increase the annual output of rice efficiently,quickly,environmentally,and non-destructively.Because of the near infrared spectroscopy(Near Infrared Spectroscopy,NIRS),it has many advantages such as high efficiency,fast speed,non-destructive,environmental protection,etc.It is widely used in the qualitative and quantitative detection and analysis of crops.NIRS can obtain information on rice varieties with different resistances more accurately and quickly without damaging the resistant rice samples.It can overcome conventional rice seed resistance identification.There are many work procedures and long time,and relevant professionals are required.As well as many problems such as environmental pollution,the combination of near-infrared spectroscopy analysis technology and qualitative and quantitative analysis and detection of crops has become the latest development direction of current food crop-related quality identification.Based on excellent analytical characteristics of near-infrared spectroscopy technology,this paper uses near-infrared spectroscopy as a means of detecting rice seed resistance.In response to the urgent need for rice seed resistance classification detection in seed selection applications,near infrared spectroscopy technology combined with chemometric methods And data mining technology established a discriminant model of rice seed resistance grading,and realized the non-destructive grading detection of rice seeds with different resistance levels.The work content of this paper mainly includes the following 3 parts:1.The principle of near-infrared spectroscopy analysis and the theoretical basis of using near-infrared spectroscopy to distinguish rice seed resistance are described.And based on the absorption characteristics of near-infrared spectroscopy,the detectable amino acid and resistance-related substance information in rice seeds were analyzed,which verified the feasibility of the experimental method.2.The near infrared spectrum information of rice is a large amount of data in high-dimensional space,because a single recognition model is difficult to play a role.Therefore,in this study,continuous projection algorithm and principal component analysis are used to reduce the dimension,combined with BP neural network,support vector machine,probabilistic neural network and partial least squares algorithm to recognize the pattern of rice varieties with different resistance.According to the actual breeding requirements,the resistance sensibility identification model was established,and the identification accuracy of rice model was as high as100%.The results showed that the multiple combination algorithm proposed in this study provided a powerful tool for the establishment of rice resistance prediction model based on the nonlinear near infrared spectroscopy of complex system.3.Establish BP,SVM,PNN,and PLS models based on the full-band spectral data.The accuracy of Raw-BP recognition is best to reach 100%,and the iteration time is 869 seconds.BP,SVM,PLS,and PNN models were established after SPA feature extraction from the spectral data.The accuracy of the four models did not improve,but the iteration speed of the BP model increased by 791 seconds.After the preprocessing of the multivariate scattering correction,the SPA feature extraction and the establishment of BP,SVM,PLS,PNN models,the accuracy of the BP model is still 100%,the iteration speed is accelerated by 840 seconds,which is 1/30 of the original data modeling speed;the accuracy of the PNN model Increased from 60% to 90%;the accuracy of SVM recognition increased from 60% to 85%.Based on the analysis of the comparison results of the above several models,the neural network identification model with513 inputs,8 hidden layers and 4 outputs of the optimal model MSC-SPA-BP model is finally established.The classification accuracy rate is 100%,and the iteration time is 29 s,indicating that the MSC-SPA-BP model can fully realize the rapid,accurate and non-destructive identification of four different resistant rice.The research results show that the optimal discriminant model MSC-SPA-BP can accurately classify rice with different resistances.The identification method of rice blast resistant varieties based on near-infrared spectroscopy is feasible and has accurate discrimination accuracy.