Nondestructive Testing Of Water Flow And Distribution Characteristics During Seed Germination Under Multi-Factor Stress

Maize seed germination is easily affected by drought and saline-alkali ring and other environmental stress.Low field nuclear magnetic resonance（LF-NMR）technology has been widely used in food science and agriculture.The LF-NMR studies of corn seed germination under duress environment condition,not only can fast nondestructive monitoring of maize seed germination in the process of moisture transfer inside,can also be combined with intelligent algorithm,by a small amount of samples nuclear magnetic data to predict large seed development situation,for corn drought and salinity environment and production to provide scientific guidance.Firstly,this study simulated the single-factor salt stress environment,and analyzed the effects of Na Cl solution with different concentrations on the water content in the three stages of maize seed imbibition,germinating and germination according to the signals and images of low-field NMR relaxation spectrum.Secondly,the stress gradient of saline-alkali and drought composite factors was designed,and a more complete and refined NMR data set was collected by low-field NMR test.The germination index data of corn seeds were obtained by solution culture and seed standard germination test.Finally,the correlation between low field NMR relaxation spectrum parameters and germination index of maize seeds was analyzed,and the prediction model of maize seed stress level was established.The main research contents are as follows:（1）Single factor salt stress test:nuclear magnetic resonance imaging and nuclear magnetic resonance spectroscopy tests were performed on samples tested under different Na Cl concentration gradients.The results showed that there were three main water components:binding water(T₂₁),semi-binding water(T₂₂)and free water(T₂₃).In addition,in the T₂relaxation spectrum,secondary peak water(T₂₄)appeared to the left of the main peak of the binding water,with weak signal and relaxation time of 0.1～1 ms,which is a kind of binding water marking the germination of seeds.In the process of maize germination,the seed germination rate decreased from 90%to 0 with the increase of salt stress.When Na Cl concentration is 50 mmol/L and below,the amplitude of water signal in different phases is less affected by salt stress.When Na Cl concentration reaches 100 mmol/L,the amplitude of free water signal is suppressed at a low level,and when Na Cl concentration reaches 150 mmol/L,the amplitude of bound water signal and semi-bound water signal are significantly delayed.（2）Cross test of compound factors:transverse relaxation time T₂（spin-spin）relaxation time of the ¹H NMR signal and seed germination index data were collected under 13 stress gradients.Firstly,a variety of algorithms were used to extract T₂ relaxation information feature subset,and the nearest neighbor node algorithm was used to evaluate the advantages and disadvantages of the feature subset.Finally,Pearson coefficient method was selected to conduct correlation analysis on relaxation information and germination index,in order to obtain positive and negative indicators for the entropy weight method of data standardization processing.Secondly,the stress grade evaluation system was established by entropy weight method,and three stress grades were divided.LF-NMR relaxation information parameters were taken as input values,stress level as output values,and support vector machine model,Gaussian process regression model and BPNN（back propagation artificial neural networks）model were established respectively.In addition,with the aggravation of stress,the biological tissues of seed membrane and endosperm were destroyed and the ability of water absorption and whitening was lost.The nine parameters T_2a1,T_2a2,T_2b1,T_2c2,A,A₂₄,A₂₂,P₂₁,P₂₂obtained by parameter extraction optimization can effectively predict the stress level of seed growth environment,and the detection effect is better than the unoptimized relaxation information.The optimized parameter set has a good prediction effect in BPNN model,and its R²,RMSE,MAE are 0.920,0.725 and 0.532,respectively.