Sophisticated Phenotyping And Equipment Development Of Individual Wheat Seed

The phenotypic characteristics of grain seeds are the basis of germplasm innovation and biological research,in this paper,based on near-infrared spectroscopy detection technology,three-dimensional machine vision technology and intelligent information processing technology,the quantitative and qualitative modeling methods of wheat kernel quality phenotype and morphological phenotype measuring methods are studied.On the basis of the phenotyping devices,a method of wheat kernel viability prediction based on multi-source phenotypic information fusion is carried out.Based on advanced sensor technology,mechatronics technology,computer technology,a wheat kernel sophisticated phenotyping integrative measurement platform is developed.The purpose of this work is to open a new idea for wheat seed phenotyping research,and the main contents and conclusions of the thesis are as follows:(1)Based on the analysis of the mechanism and method of spectral detection for individual wheat seed,aiming at the characteristics of small grain,irregular surface and sensitive to light source,a near infrared spectrum detection device based on a novel stereoscopic light source unit is designed and developed.The research on the optical path analysis and structure design of the stereoscopic light source structure is carried out,the scheme design of quality phenotyping device based on the stereoscopic light source is formulated,and the design of parts selection,fabrication,and,corresponding detection software are completed.The experimental results show that the NIR spectrum detection system with stereoscopic light structure can achieve real-time and Online spectrum acquisition of individual wheat kernel.In terms of performance,the maximum wavelength standard deviation is 0.04nm,the maximum reflected light intensity variation coefficient is1.9%,and the maximum absorbance variation coefficient is 0.4%,which show good performance in both wavelength and absorbance repeatability,and prove that can be used for sophisticated quality phenotyping of individual wheat kernel.(2)Based on pattern recognition algorithms and stereoscopic light source structure NIRs system,the quality phenotyping method of individual wheat kernel is studied.The classification ability of wheat black-tip disease is selected as the object of study,the stereoscopic absorbance near-infrared spectra of individual wheat kernels are collected.SPA and PCA are used to select the characteristic wavelengths and reduce he dimension of the spectrum data.Combined with SVM,ELM,RF and Ada Boost machine learning methods,a2-category,3-category and 4-category discrimination models of wheat black-tip disease are constructed respectively.The results show that the recognition effect,with the spectra device and by the algorithms,is between 93.3%and 98.6%,and the recognition accuracy decreases with the category number increases,so it is necessary to further study on improving pattern recognition method or optimizing parameters.Meanwhile,wheat kernel protein content prediction ability is also selected as the object of study,the stereoscopic absorbance near-infrared spectra of individual wheat kernels are also collected,according the real measured protein content,two models based on full spectra and characteristic wavelengths optimized by SPA are established.The~2 of calibration set is 0.96 and 0.85,the RMSE is0.56 and 1.04,respectively.While,the~2 of prediction set is 0.80 and 0.82,and the RMSE is 1.08 and 1.04,respectively.It shows that the model based on full spectra has higher prediction accuracy,and the model based on characteristic wavelengths has lower information redundancy,which is more suitable for embedded development and online quality phenotyping.(3)To solve the problem that it is difficult to accurately measure the three-dimensional morphology of wheat seed due to their small size and complex morphology,a method for obtaining the morphological phenotype based on omni-directional microscopic image sequences and 3D reconstruction is proposed.Based on the design of the three-dimensional machine vision sophisticated measurement scheme of individual seed,a set of machine vision platform for obtaining the omni-directional microscopic image sequence of individual wheat kernel is constructed.On the basis of establishing the rotation geometry model of wheat kernel micro imaging,a method of solving the measurement viewpoint attitude based on rotation axis calculation is proposed.The camera’s internal and external parameters are calibrated under the omni-directional perspective,which lays the foundation for the acquisition of 3D shape of wheat kernel.After the background separation and binarization of the obtained image sequence of wheat kernel,the contour sequence of kernel silhouette is obtained.Based on the marching cube algorithm,the visual hull of wheat kernel is established,and the fine 3D reconstruction model of wheat kernel is obtained.The reconstruction results show that the proposed method can not only complete the morphological reconstruction of micro seeds,but also reproduce the unsmooth surface and ventral groove.With the increase of the number of reconstructed images and the accuracy of voxels,the effect of 3D reconstruction is improved.(4)Based on the information of 3D reconstruction of kernels,a series of morphological phenotyping models are developed.According to the OBB bounding box,the length,width and thickness of kernels are determined.Compare the three phenotypic data obtained by reconstruction measurement with the manual measurement,results show the determination coefficients are 0.97,0.97 and 0.88,respectively.The surface area data were obtained by triangulating the grain surface;Point cloud data slicing method is used to calculate the kernel’s volume.Through the method of maximum section projection,the area and perimeter of projection are extracted.The reconstructed surface area,volume,projection perimeter and projection area are compared with the measured results of the calibration sphere,results show the relative errors are between 3.5%and 5.8%respectively,which proves that the proposed approach based on omni-directional microscopic image sequence 3D reconstruction is feasible,and it can be used for the wheat kernel’s sophisticated morphological phenotyping.(5)In view of the difficulty in finding the difference of viability between individual seeds by the existing spectral detection methods,and the neglect of the influence of seed morphological characteristics on the spectral prediction and classification model,an early discriminant method of viability of individual wheat kernel based on the fusion of spectral information and image information is proposed.The experimental scheme based on the fusion of feature layer and decision layer is designed in detail.On the basis of morphological phenotype extraction method with three-dimensional reconstruction of wheat kernel,six morphological parameters of wheat grain length,grain width,grain thickness,projection area,surface area and volume are extracted respectively;On the basis of quality phenotype NIRs detection method with stereoscopic light source,the spectral data of individual wheat kernel are obtained delicately.Through artificial aging and germination test,the kernels are divided into three viability levels.The morphological features and near-infrared spectral principal components/features are fused in feature layer.Combing with four pattern recognition methods of SVM,RF,ELM and Ada Boost,the viability discrimination model of wheat kernel is established.The results show that the performance of the eight discrimination models is quite different in the calibration set,and the accuracy is between 76.45%and 94.77%.The morphological+PCA-RF model and morphological+SPA-RF model have the best performance of 75.97%discriminant accuracy in the prediction set.Because the morphological+PCA-RF model has a higher comprehensive harmonic index of 10.60,it is considered to be the optimal model in feature layer fusion.After the spectral principal component information and the morphological feature information are respectively used to establish the viability discriminant model by RF algorithm,according to the set fusion decision rules,the fusion model of individual wheat kernel viability classification based on D-S evidence theory is established.The results show that the discrimination ability of individual wheat kernel’s viability based on decision-making level fusion is better than that of single method discrimination model,which has accuracy of calibration and prediction set of82.13%and 74.14,respectively.But compared with the results based on feature-based fusion,it is concluded that the discrimination performance of individual wheat kernel’s viability based on feature-based fusion is better,and it is a more suitable method for individual wheat kernel’s viability prediction.(6)According to the characteristics of modern agricultural equipment,integrating the spectral quality phenotyping device and morphological phenotyping device developed above,a set of solutions for automatic seed adsorbing,seed collecting,transportation and docking,accurate data acquisition and integrated implementation of individual wheat kernel is designed.The control and structure design of the platform is completed.The hardware design Including seed adsorption,transportation,collection module and image acquisition module,spectrum acquisition module,etc.,is constructed.The software design and development of the main computer and the slave computer are completed by adopting the distributed control architecture and the programming idea of multithreading and multitasking.Joint debugging,functional verification and performance analysis of the whole platform are carried out.The results show that the platform can realize the functions well,including seed population identification and location,seed adsorption,rotation,spectral data acquisition,image sequences acquisition,seed collection and online phenotype measurement.It is estimated that the platform can work continuously for 12 hours per day,and the measurement throughput can reach 3500 kernels/day without considering the online 3D reconstruction.In addition,for the repeated measurement of seed length,width and thickness on the platform,the coefficient of variation is less than 4%,the maximum coefficient of variation of volume is 7%,which shows a good repeatability on morphological phenotyping.For the repeated measurement of protein prediction of wheat kernel,the standard deviation is less than 0.3%,which shows a good repeatability on quality phenotyping.