Research On The Prediction Model Of Rice Seed Vigor Based On Near-infrared Spectroscopy Data

As the largest rice producer in the world,stable annual rice yield has become the basic guarantee of people's life.The selection of high vigor rice seeds for sowing is one of the important links to ensure stable and high yield of rice.The traditional methods of seed vigor determination are mainly germination test and chemical test.These methods have some shortcomings,such as long test cycle,waste of samples and cumbersome operation.Therefore,a fast and non-destructive seed vigor detection technology has great potential benefits for agricultural production.In this study,near infrared spectroscopy(NIRS)was used to measure the vigor of rice seeds quickly and nondestructively,so that the germination rate of rice seeds could be calculated quickly by measuring the spectral curve of rice seeds.The main contents of this paper are as follows:(1)Modeling data acquisition.Taking rice seed No.4 of Wuyou rice harvested in Wuchang City of Heilongjiang Province in 2017 as the research object,the rice seeds were put at 45~oC and90%humidity for artificial aging for 0 d,1 d,2 d,3 d,4 d,5 d and 6 din order to obtain rice seeds with different vigor levels and collect their corresponding near infrared spectra data.Sixty groups of sample data were collected for each aging day of rice seeds,and 420 groups of sample data were collected altogether.Then germination experiments were carried out to obtain the true germination rate of rice seeds in different aging periods after collecting spectral data.(2)Modeling data processing and modeling.In the process of data processing,first of all,the spectral data of rice seeds with different aging days were removed by Monte Carlo Cross Validation(MCCV)method.A total of 32 abnormal samples were removed.Then,the feature wavelength of the spectral data after eliminating the abnormal samples is selected by the method of non-information variable elimination(UVE),which reduces the spectral data from 1845 data points of the whole spectrum to 524 data points,and greatly reduces the dimension of the data.After dividing correction set and verification set by SPXY method,PLSR,BP,RBF neural network and improved PLS+BP and PLS+RBF models are established respectively.(3)Evaluation of model results.Comparing the evaluation parameters of the five models for predicting the germination rate of rice seeds,we can see that the evaluation parameters of the model established by PLS+BP method are better than those of the other four models.The model,R_c~2,RMSEC,R_p~2,RMSEP and RPD are 0.86,1.99,0.83,2.01 and 2.55,respectively.The experimental results show that the prediction model established by PLS+BP method shows good accuracy and stability.The results show that the spectral data processing algorithms such as MCCV and UVE can effectively improve the prediction performance of the model.The model of rice seed germination rate measurement based on near infrared spectroscopy was established by PLS+BP method,which can accurately determine the germination rate of rice seeds.At the same time,this study also provides a reference for the determination of seed germination rate of other crops.