| Seeds are an important basis for ensuring national food security and the effective supply of major agricultural products,and are also the starting point of all crop industry chains.According to the"14th Five-Year Plan for Modern Seed Industry Improvement Project Construction Plan",by 2025,China will form a modern seed industry development pattern of protection,breeding,testing,and propagation of the division of labor and cooperation,closely linked,to comprehensively improve the level of modernization of the seed industry,in order to achieve the self-sustainability of the seed industry,independent control of seed sources.New technology and new theoretical research on seed quality measurement is an important part of the evaluation of seed"testing",to help improve the facilities and equipment conditions and variety testing capabilities.Most of the traditional crop seed quality testing techniques use laboratory wet chemical analysis methods,which have problems such as cumbersome pretreatment,destructive sample,and environmental pollution,which are difficult to meet the new needs of rapid and non-destructive testing proposed by the development of modern agriculture.Terahertz technology,as an emerging cross-cutting frontier,has shown a vigorous application prospect in the field of quality detection of biological samples such as seeds due to its rich fingerprint spectrum,security,and perspective.With the support of the National Natural Science Foundation of China project“Research on the key technology of rapid nondestructive detection of maize seed vigor by terahertz time-domain spectral imaging（61807001）”,in this thesis,taking"corn",the largest crop in China,as the research object,using terahertz time-domain spectral imaging,chemometrics,machine learning and digital image processing technologies and methods,the focus is on exploring the feasibility of seed component content detection methods closely related to seed vigor changes and non-destructive visual analysis of seed components,the details are as follows:（1）Keeping the moisture content of seeds in an appropriate range is conducive to ensuring the stability of seed vigor.Real-time,rapid and accurate detection of seed moisture content is of great significance for seed selection and breeding,increasing yield and income.However,the interference of environmental moisture,system noise and non-target components will cause serious problems to the spectral characterization and prediction of seed water.Therefore,in this thesis,a terahertz attenuated total reflection technology is used combined with spectral preprocessing and machine learning methods to establish a quantitative prediction model of corn seed moisture content.On the basis of the linear characteristic spectral region screening,a nonlinear model was constructed for rapid quantitative prediction of seed moisture.The results showed that the model prediction set correlation coefficient was 0.9930,and the prediction root mean square error was 0.0697.The established model has good practicability（PRD>3）,using the characteristic band screening method,the moisture sensitive spectral region is about 51.79～57.31cm-1,which is similar to the absorption peak(near 60cm-1)of theoretical simulated moisture in the terahertz band.In this study,by constructing a nonlinear model for quantitative prediction of moisture content based on the characteristic spectral region,the nonlinear interference can be significantly reduced,and the performance of the prediction model can be effectively improved.（2）Starch is one of the main sources of supply of energy required during seed germination and is closely related to changes in seed vigor.Accurate detection of starch content is extremely important for maintaining viability stability.In this thesis,the terahertz transmission imaging technology combined with digital image processing,machine learning and other methods is used to non-destructively extract the characteristic information of corn seeds and build a quantitative prediction model of starch content.Using the automatic thresholding method to effectively segment the seed area and extract the average spectrum for the collected corn seed image data with different aging degrees,the linear interval partial least squares model was used to filter the starch terahertz characteristic spectrum area,and then the feature spectrum area based on support vector machine and characteristic spectrum area was constructed.A fast quantitative nonlinear analytical model for seed starch.The research results show that the nonlinear SVR prediction model can effectively eliminate the nonlinear noise interference in the spectrum,and the mean rmse of the model is reduced by95%.The characteristic band of starch is around 144～148cm-1.This study is expected to provide a potential supplementary technology for the field of rapid nondestructive measurement of seed quality in the future.（3）Non-destructive analysis of the internal relationship between the change process of seed vigor and the spatial distribution of seed components plays an important role in studying the mechanism of seed vigor and preventing the loss of seed vigor.This thesis proposes to use the terahertz reflection imaging technology combined with the moving window correlation coefficient image analysis method to perform pseudo-color imaging,non-destructively to visualize and construct the spatial distribution map of starch and protein in maize seeds with different aging degrees,and to explain the changes in the spatial distribution of components in the process of seed vigor decline.The results showed that the spatial distribution of starch and protein content showed an overall decreasing trend in the range of 29.83～67.36cm-1 during the decline of seed vigor.This study preliminarily explores the feasibility of visualizing the spatial distribution of maize seed components with different aging degrees,and is expected to provide a new perspective for the research of germplasm bank seed vigor detection and analysis of seed physiological and ecological changes.New methods and new theoretical studies on seed quality determination not only help enrich the seed quality determination technology,but also better meet the new demands put forward in the development of modern agriculture in my country.The research in this thesis is expected to promote the intersection and integration of multiple disciplines such as seed inspection,optical imaging,information science and mathematics,and can provide theoretical basis and method reference for rapid non-destructive testing technology of seed quality.Significance and application prospects.Therefore,the research in this thesis has practical economic benefits and long-term strategic significance both for our living natural environment and for the sustainable development of agricultural economy. |
PDF Full Text Request