Study On High-speed Determination Of Corn Seeds Quality And Variety Using Near Infrared Spectroscopy

Corn is one of important crop in our country.It is very important for keeping continuous growth of the corn agricultural to develop the assessment methods of corn seeds quality(such as the purity and viability of corn seeds)and the technique of breeding the excellent variety(such as haploid breeding techniques).However,traditional assessment methods of seeds purity and viability are time consuming and seeds damaging,and sorting of haploid maize seeds still relies on labor with low efficiency and high error rate.Those problems have been restricting the development of corn breeding industry.Therefore,studying the rapid evaluation method of corn seeds quality and the high-speed sorting technology of haploid maize seeds have great practical significance for large scale corn breeding production.Near-infrared spectroscopy(NIRS),which contains abundant molecular information and has the characteristics of rapidness and non-destructiveness,has potential to develop the rapid evaluation method of corn seeds and high-speed sorting technology of haploid seeds.To date,there has been few reports on those high speed techniques,because the spectral noise induced by variation of complex shape and component inhomogeneity of corn seed as well as very short exposure time for collecting dynamic spectra of moving seed under high speed is too heavy to effectively eliminate for establishing practically feasible methods of quality assessment and haploid recognition using NIR.These problems have led to the failure of NIR spectroscopy to achieve breakthroughs in the field of rapid quality evaluation and sorting of maize seeds.This thesis aims to study the problems mentioned above and to develop the high-speed techniques.In this thesis,first,feasibility of seed quality assessment and haploid seeds recognition using static NIR spectra in combination of pattern recognition was studied by artificially putting corn seed on the sample holder of a benchtop FT-NIR spectrometer which has a high spectral resolution and a better signal-to-noise rate.Then,a home-made spectra collection device of corn seed moving under high speed was used to collect the dynamic spectra of the samples.Common spectral pretreating methods were tried to eliminate the heavy noise of the dynamic spectra.Comparison of recognition results using static spectra and dynamic spectra was made.The common preprocessing methods are not suitable for establishing high-speed methods.Several effective de-noising methods were investigated respectively to develop the high-speed methods for seed quality assessment and haploid sorting.The research results are as follow.1.Rapid assessment of maize hybrids purity using NIRS.Y1,Y2 and Y3 corn hybrids(total of 390)were collected from Denghai Seeds Co.,Ltd,Shandong province.The benchtop FT-NIR spectrometer were used to collect their diffuse reflection NIR spectra.Before modeling,spectral data were dealt with using MSC,SNV,1st and 2nd derivative,respectively,and SIMCA method were applied to build their discriminant models,accordingly.The correct recognition rates of SIMCA models using MSC,SNV,1st and 2nd derivative were 93.33%,93.33%,90%and 90%for the validation set,and 90%,86.67%,90%and 83.33%for test set,respectively.Then the dynamic spectra of the three kinds of maize kernels(total of 600)at the speed of 0.4 m/s were collected.Due to the best performance(93.33%)shown in static spectrum study,MSC was used to process the dynamic spectra.In comparison with the static spectra models,the correct recognition rates of the dynamic spectra SIMCA models fell to 83%and 80%for the validation and test sets,respectively.To eliminate the redundancy variables before modeling,CARS was used to select 35 characteristic variables at last.They were applied to build SIMCA model,and excellent accuracy(90%)was achieved.2.High-speed recognition of haploid maize seeds using NIRS.The static and dynamic spectra of single-kernel maize(210 haploid seeds and 222 hybrid seeds)were acquired by FT-NIR spectrometer and dynamic spectra capture system,respectively.First,the static spectra were dealt with using MSC,SNV,1st and 2nd derivative,respectively,and SIMCA method were applied to build discriminant models.The correct recognition rate of SIMCA models using MSC,SNV,1st and 2nd derivative were 90.83%,90%,95.27%and 93.70%for the validation set,and 89.76%,87.50%,94.17%and 90.55%for test set,respectively.Second,in consideration of 1st being the best of the pretreating methods for static spectra study in this section and CARS as the best for variable selection of dynamic spectra in seeds purity assessment,1st,CARS and the combination of 1st and CARS were used to process the dynamic spectra,and the correct recognition rates of their SIMCA models were 85%,87.5%and 85.83%,and 81.10%,86.61%and 83.33%for the validation and test sets,respectively.Compared with the static spectra SIMCA models,the identification ability of dynamic spectra SIMCA models declined dramatically and can't meet the practical needs.It is not difficult to see that the dynamic spectra contain much more noise than the static ones and can't be effectively eliminated by the common pretreatment methods.In consideration of the useful spectral signals for recognition being different from noise in frequencies,a de-noising method of dynamic spectra of seeds moving under high-speed was proposed using discrete wavelet transform.Mother wavelet Daubechies 4 with decomposition level number 7 was used to decompose the dynamic spectra.It was found that performance of the SIMCA model established using the reconstructed spectra that did not include the approximate signal(to cA7)and high frequency signal(to cD1)is the best and excellent accuracies for haploid(93.33%)and hybrid(95.27%)seeds under 20 kernels/s were achieved.High-speed sorting of haploid and hybrid corn seeds was successfully realized.This research achievement has an exemplary significance for large scale sorting of haploid seeds.3.Assessment of corn seeds viability using NIRS.The benchtop FT-NIR spectrometer was used to collect static spectra of dry corn seeds(total of 1300)and corn seeds which imbibe water(total of 170),respectively.The germination test of sand bed and TTC staining method were applied to detect the viability of dry and imbibing corn seeds,respectively,and then,high viability and non-viability seeds were defined according to the result of test.Spectral data were dealt with using MSC,1st derivative,CARS and DWT,and SIMCA models were built for each pretreating method.The correct recognition rates of SIMCA models using MSC,1st derivative,CARS and DWT were 55.88%,58.82%,61.76%and 61.76%for dry seeds,and 75.51%,81.63%,83.67%and 76.92%for imbibing seeds,respectively.The results showed that imbibing can increase obviously the correct recognition rate of viability model using NIRS and it still needs to improve the correct recognition rate further.