Research And Implementation On Cultivar Discrimination System For Maize Seeds Based On Biomimetic Pattern Recognition

Fast, accurate, and stable discrimination method of seed varieties is one of theimportant issues that need to be solved in the development of modern agriculture, is also anew direction for research scholars in recent years. Near infrared spectroscopy has becomethe preferred technology for new discrimination method of seed varieties with itsnondestructive, fast, low-cost characteristics.The subject overcoming the limitations of existing methods of cultivar discrimination;whose experimental materials are corn varieties passing the variety certification; whose corealgorithm is biomimetic pattern recognition; which is for the differences in performance andfeatures between the PC and embedded systems; whose ultimate goals are low-cost, low datavolume, small volume, easy to operate, the short period and high accuracy, is aimed toestablish a convenient and efficient maize varieties authenticity identification system,and isapplicable to embedded systems.On the basis of in-depth study of the data processing and pattern classificationtechniques, this thesis propose two metheods to solve the parameter drift problem of spectruminstruments from the "correction" and "inclusive" two point of view.One method is transformthe spectral acquired at different times with the DS algorithm which is belonging tocalibration transfer to obtain the spectral data can be used in the model has beenestablished.Another method is inclusive the uncertain information in modeling set by thePLS+LDA supervised feature-based inter-day combination modeling approach to obtainadaptable model.The test result shows that both of them can effectively solve the stabilityproblem of the model,and the second is better.Cutting down the near-infrared spectral scanning range can reduce the cost of opticaldevices, and can greatly reduce the amount of original data.And this can greatly reduce the requirements on the storage capacity and computing power of the processing system.In thisthesis discrimination performance using short wave band of near-infrared spectra was studiedand further researched on the ability of discrimination using the different bands.Finally, adiscrimination of corn varieties approach based on biomimetic pattern recognition which issuitable for embedded implementation is proposed, at first, use the1100-1752nm wave bandto replace full-band, and then PLS+LDA supervised feature-based inter-day combinationmodeling will be applied, the identification rate attains96.08%.