Nondestructive Viability Detection Of Wheat Seeds Based On Hyperspectral Technology

Seed vigor is an important component of seed quality,and as seed is one of the most basic production materials in crop production,high vigor seed is a prerequisite for high and stable crop yield,therefore,the detection of seed vigor is particularly important.The existing seed vigor testing technology is time-consuming,laborious and expensive,and the development of non-destructive and rapid vigor testing technology is extremely important for breeding,field production and germplasm resource conservation.In this study,18 wheat varieties were selected,and the hyperspectral reflectance data of wheat seeds were collected using a geophysical spectrometer after artificial accelerated aging treatment.After the high-spectral reflectance curves were censored and removed,the original spectra were preprocessed using six preprocessing methods,such as the standard normal transform(SNV),multiple scattering correction(MSC),maximum-minimum normalization(Max-Min NOR),first-order differentiation(FD),logarithm of inverse(LOG(1/X)),etc.The full-band spectra were modeled by combining three modeling methods,such as partial least squares regression,principal component regression,and multiple stepwise regression,and excluding The model parameters were optimized after excluding some sample data,and the optimal model was established by using two characteristic band screening methods to obtain characteristic bands respectively.The results are as follows.1.Different varieties were selected for the artificial accelerated aging test can effectively affect the seed vigor,and the marsupial distance method can effectively reject the abnormal index data;the pre-treatment of the spectra can reduce many systematic errors,noise and abnormal spectral curves;the spectral pre-treatment can effectively change the trend of the spectral curves and can limitedly affect the correlation between seed vigor and spectra.2.The model with the best predictive ability for wheat seed vigor indicators was the partial least squares regression model,and the optimal pretreatment for germination potential and vigor index indicators was the standard normal transformation;the optimal pretreatment for germination rate and germination index indicators was the 5-point convolutional smoothing pretreatment and no pretreatment;after adjusting the samples,the modeling set coefficient of determination(R_C~2)of the full wave model for the four vigor indicators was greater than 0.9,the validation set The coefficients of determination(R_C~2)and validation set(R_P~2)of all four vitality indexes were greater than 0.81,and the relative analysis errors(RPD)were greater than 2.0.The predictive ability and stability of the models were good.3.The model with optimal predictive ability for wheat seed vigor index is the PLSR model,and the optimal models for GE and VI indexes are SG+SNV+PLSR;the optimal models for GR and GI indexes are SG+PLSR and NONE+PLSR,respectively;after adjusting some samples of the dataset and using the new dataset for model optimization,the results are:in the GE-SG+SNV+PLSR model R2P is not less than 0.87 and RPD is not less than 2.61 in the GE-SG+SNV+PLSR model;R2P is not less than 0.859 and RPD is not less than 2.48 in the GR-SG+PLSR model;R2P is not less than 0.86 and RPD is not less than 2.56 in the GI-NONE+PLSR model;R2P is not less than 0.82 and RPD is not less than 2.29 in the VI-SG+SNV+PLSR model.2.29.4.The full-band spectral data were screened by using successive projections algorithm(SPA)and principal component analysis(PCA)for the feature bands,and the screened feature bands were used for model building.The results showed that the feature band models were weaker than the full-band models across the board,among which VI indicators were not suitable for building PLSR models using SPA and PCA screening of feature bands,and the best results were obtained for the models built by the feature bands extracted by the SPA method among other seed vigor indicators models;when the maximum number of feature bands was set to 15 in SPA screening,the extracted feature bands were used for GP and GE model building The R2P is not less than 0.81 and 0.840,and the RPD is not less than 2.17and 2.31,respectively;when the maximum number of feature bands is set to 30 in SPA screening,the extracted feature bands are used for the establishment of GI models optimally,with R2P greater than 0.81 and RPD greater than 2.2.5.In summary,the results show that either using full-band hyperspectral data or eigenband data can be used to detect the germination potential(GE),germination rate(GR),and germination index(GI)indexes of wheat seeds,while the vigor index(VI)index is only applicable to detection using full-band hyperspectral data.