Research And Implementation Of Automatic Sorting Technology For Haploid Kernels In Maize

It is significant for maize haploid efficient breeding and improving maize production to identify the haploid kernels rapidly and accurately.However,the efficiency of the existing detection technology for maize haploid is very low,so studying a newly efficient method for identification of haploid maize is the hotspot at home and abroad.Near Infrared Spectroscopy(NIRS)qualitative analysis techniques should be used to develop the high throughput automation equipment for sorting haploid maize kernels because of the features of rapid,nondestructive,low cost detection,easy to operate,etc.The lab materials are haploid maize kernels and hybrid kernels which have been marked on gene.According to the slightly difference between maize haploid and heterozygote and based on studying in detail the algorithm commonly used on NIRS qualitative discrimination process,the subject is aiming to develop an automatic identification and sorting system for maize haploid seeds which can get high quality spectra and high accuracy recognition.Based on Near Infrared Reflectance Spectroscopy,this paper separately used Biomimetic Pattern Recognition(BPR)and Support Vector Machine(SVM)as the core algorithm to build NIRS qualitative analysis models,and conducted a comparative study about the models.The experiment results show that the results are better when using PLS+OLDA+SVM to extract features of the data and build models.But it's still unable to get the ideal effect.Further experiments show that the position effect existing in the acquisition of haploid maize kernels is the primary cause that haploid kernels and hybrid kernels can not be classified efficaciously.The quality of spectra has a direct impact on model identification rate,so we can set up an accurate and stable analysis model and resolve the position effect of Near Infrared Reflectance Spectroscopy Fundamentally by using Near Infrared Transmittance Spectroscopy.This paper researched the parameters of NIRS,and examined how the quantity of light through the spectrometer and the light intensity affect the quality of spectra and model identification rate.Then we proposed two methods to improve the quality of spectra.If we reduce the working voltage of light source and use visible light cut-off filter,we will acquire high intense and high efficient light source suitable for Near Infrared Transmittance Spectroscopy analysis.In order to identify and select haploid maize kernels rapidly,nondestructively and automatically,this paper designed and developed an automation equipment which can achieve making kernels in line,information acquisition and processing and selecting kernels.