Research On Corn Seed Testing Technology Based On Machine Vision

Nowadays,people’s living standard is improving day by day,and the demand for high-quality crops,especially high-quality protein maize,is increasing.This raises higher requirements for breeding new maize varieties and screening high-quality maize varieties.Traditional maize detection method has been unable to meet the need of high efficiency.With the development of machine vision and deep learning,the new technology proposed in this paper has been widely used in agriculture and other aspects,and has achieved relatively ideal results.In this paper,the methods of machine vision,digital image processing and deep learning are used to study the various shape parameters and classification of maize kernels.Because some shape parameters of maize kernel can be obtained by visual processing,this paper uses machine vision and other methods to calculate some shape parameters of corn kernel.For the classification of maize kernels,it is difficult to obtain ideal results only by traditional machine vision,therefore,this paper uses deep learning and other techniques to study the classification of maize kernels.The main research contents of this thesis are as follows:1.A hardware system based on the machine vision,including industrial cameras,lenses,light sources and frames is built to extract shape parameters of maize kernels.This system has the advantages of simple structure,easy construction,and low maintenance cost.2.Through the methods of gray processing,spatial filtering,image segmentation and morphological processing,the existing image preprocessing scheme is simplified,which greatly reduces the hardware dependence of the maize screen machine recognition and speeds up the processing.The experiments show that the pretreatment scheme can better meet the demands.3.The extraction scheme of maize kernel shape parameters(area,perimeter,centroid,long and short axis,minimum circumscribed rectangle,etc.)was proposed.The conversion of the area and the perimeter is achieved according to the pixel scale method,and the centroid is obtained by the pixel mean.The long and short axes are obtained by measuring the arbitrary line segments of the centroid and the minimum circumscribed rectangle is obtained by using the long and short axis translation method.At the same time,the statistical analysis was carried out on the characteristics of maize kernel area,perimeter and long and short axis.4.This paper uses deep learning and other techniques to classify and identify corn kernels.This paper uses a variety of data processing forms and optimization algorithms to optimize the experimental results,including data enhancement and format conversion,Dropout,Adam,Batch normalization and other algorithms.Through these methods,the training speed and the generalization ability of the model can be improved greatly.On this basis,different varieties of maize kernels are identified,and the accuracy of these methods proposed in this paper can be improved to over 97%by means of migration learning.