Researching On Maize Seed Quality Detection Based On Hyperspectral Imaging Technology

Seeds are the core of agriculture and it is also an important carrier of agriculture scientific and technological progress.Seed quality is the key to determine the harvest of crops.It is also the core competitiveness of an enterprise and a country.Compared with the quality of general commodities,its impact is longer,more profound and broader.Corn is mainly distributed in the northeast,north and southwest of China.Compared with traditional grain crops such as wheat and rice,corn is more adaptable to the environment.It is not only a critical high-yield food crop,but also an important feed source and chemical raw material.Improving the quality of single maize seed is not only conducive to increasing farmers’incomes and improving the development of the rural economy,but also conducive to ensuring the safety of seed storage and transportation and reducing economic losses.The promotion of single grain sowing technology also puts forward higher and ore refined requirements for the quality of seeds.However,the current detection method in detecting the quality of maize seed will cause damage to the seed and the efficiency is low,while the common detection equipment in the market cannot carry out single maize seed detection and the accuracy is low.Therefore,the non-destructive testing technology for single maize seed quality is beneficial to development of agriculture,commerce and planting industry.This paper takes commercial maize seeds as the research object,and studies the detection methods for analysis of moisture content and maturity of single maize seed by using hyperspectral imaging technology combined with pattern recognition technology.The main research contents are as follows:The rapid non-destructive detection method of single maize seed moisture content by hyperspectral imaging technology with a spectral range of 930-2548 nm was studied.In this section,the full surface spectra of embryo(S1)and endosperm(S2)side were collected,respectively.Then,the fusion spectra(S3)were calculated by using S1 and S2.The singular samples were eliminated by Monte Carlo cross validation method.The three types of spectra were analyzed with different pretreatment methods,the results indicated that SG-1Der,SG-SNV and SG-MSC methods were suitable for S1,S2 and S3 spectra,respectively.In order to reduce the complexity of the model and improve accuracy,the UVE combined with SPA was used to select feature wavelengths,and then PLSR and LS-SVM model were built based on the feature wavelengths.The results showed that the performance of the model based on S3 spectra was better than that of other spectra model.The best prediction models were UVE-SPA-PLSR and UVE-SPA-LS-SVM,the R_p of prediction set was 0.92and 0.94,and RMSEP was 1.22%and 1.20%,,RPD was 2.61和2.65 respectively.The UVE-SPA-LS-SVM model based on the S2 spectral obtained the strongest applicability.The model can be used to predict the moisture content based on embryo and endosperm spectral,respectively.the R_p of prediction set were 0.91and 0.76,the RMSEP were 1.38%and 2.98%,respectively.The rapid non-destructive detection method of single maize seed maturity by hyperspectral imaging with a spectral range of 1000-2300 nm was studied.In order to explore the influence of spectra at different on maturity detection,hyperspectral images on the endosperm side(T1)and embryo side(T2)were collected,and the band image at 1098 nm was used to mask the hyperspectral image,then the average spectra of full surface was extracted.Next,the fusion spectra(T3)were calculated by using T1 and T2.Three types of spectra were pretreatment by different preprocessing methods,then PCA algorithm was used to extract the feature wavelengths,and PLS-DA and DT were used to establish the classification models based on feature wavelengths.In order to reduce the impact of the division of calibration set and prediction set on the classification accuracy,50 independent experiments were carried out,and the average classification accuracy was used to evaluate the model performance.The results indicated that the PLS-DA model based on the 12feature wavelengths extracted from T2 spectra preprocessed by SG-1Der obtained the best classification effect and the strongest robustness and applicability.This model can be used for T1 and T2 spectra,the average classification accuracy were 98.7%and 100%,and the errors were 1.40%and 0%,respectively.In summary,this article used hyperspectral imaging technology to conduct non-destructive detection of single maize seed quality.It laid the foundation for developing the fast,non-destructive equipment for seed quality grading.