Research On Sorting Technique For Haploid Maize Based On Computer Vision

Haploid breeding can largely shorten the breeding cycle,which is one of the most efficient methods in breeding.Identifying haploid from the hybrid kernels crossed with the inducer has been an important technique in modern agriculture breeding research.The method based on machine vision is becoming the first choice in identification,classification and grading of agricultural products,due to its advantages of high intelligence,more information,good stability,high efficiency and no human intervention.In this thesis,an automatic maize sorting technique based on machine vision has been proposed to identify haploid and hybrid seeds with genetic marks.A sorting system which can realize automatic separation has been designed and implemented.The main research work is as follows:Firstly,an image acquisition module has been designed to get images with embryo surface automatically,including three parts: the rugged slope,the high-speed camera and the light source.In order to get various gestures of maize kernels during their falling process,an uneven surface texture was made on the rugged slope.Experimental results showed that the success rate to get images with embryo surface is 90%.Secondly,the original images have been preprocessed with related arithmetic.An image segment method which combine threshold value segmentation with edge detection has been proposed to segment the corn kernel area from the whole image.And the segmented image has been normalized in size to improve the recognition efficiency.Thirdly,some picture features which have robustness to posture changes has been analyzed and researched,such as color histogram,local binary pattern,scale invariant feature transform,speeded-up robust transform.Experimental results showed that the selected SURF feature has better performance.Two classification methods based on template matching and SVM(support vector machine)separately have been analyzed.Experimental results showed that the classification method based on SVM has better performance,the recognition rate of haploid can be 95%.Finally,a high-throughput maize haploid sorting system has been designed based on computer vision,including four parts: the transmission module,the rugged slope module,the image acquisition & processing module and the kernel separating module.