The Design And Implementation Of The Beet Seed Germination Prediction System Based On Hyperspectral

Beet is one of the main sugar crops in China,but in the process of sugar beet planting,because of the low germination rate of its seeds,has been restricting its large-scale popularization of planting.In order to improve the germination rate of beet seeds,the method of artificial selection is usually used to remove the seeds with lower vigor.However,this method has the disadvantages of strong subjectivity,time-consuming and laborious.based on this,a hyperspectral germination prediction system for beet seeds was designed to provide a new detection technique for batch online nondestructive testing of beet seeds.The main contents of this paper are as follows:(1)The hyperspectral image of 3072 samples was obtained by using the near infrared hyperspectral imaging acquisition system;the image was cut,threshold segmentation,expansion,hole removal,contour extraction and the average spectrum of the final extraction seed region was used as its characteristic spectrum.the spectra were pretreated using six pretreatment methods:MSC,SNV,SG,DET,1D,2D.SVM(RBF),random forest and Light GBM classification models were established respectively.the experimental results show that the data after second-order difference preprocessing Light GBM the best modeling effect.In full wave The prediction accuracy of germination classification of test set was 93.7%.(2)the spectral data were reduced by feature importance and correlation coefficients.the spectral data were extracted at 455,465,495,505,719,866,908,930,946,983,989,994,1005,1021 and 1037 nm.After parameter optimization,the classification prediction accuracy of the model reached 90.8%.(3)Combined with the model selected in the above modeling process,an online germination prediction system for beet seeds was designed.And the system is divided into two parts: the first part is the server,the server is built on the Baidu cloud server,and the communication with the client is TCP/IP.The server mainly completes the prediction of the data uploaded by the client,and returns the prediction result.This part mainly includes communication module,spectral preprocessing module,prediction module under full band and prediction module under characteristic band.The second part is the client,the main functions of this part include the module of importing forecast sample data,communication module,display interaction module and function selection.The results show that the system can predict the germination of beet seeds accurately and effectively,and put forward an idea for on-line nondestructive testing of beet seed batch.