Recognition Algorithm Of Plug Seedling Based On Machine Vision

The cultivation of tobacco seedlings in plug is the key stage in the early phase of tobacco planting,and the most important foundation for obtaining high quality tobacco is to cultivate enough healthy and robust seedlings.A strong tobacco seedling needs to be cultivated in each hole of the plug,yet there may be some hole without seedlings or more than one seedling,for empty cell,a seedling should be replanted,as for plug holes with more than one seedling,weak seedlings needed to be pull out to retain the strong one,known as thinning seedlings.Artificial reseeding and thinning are very inefficient.Therefore,it is very necessary to develop automatic seeding and thinning machine.In order to solve this problem,this paper studies the algorithm of hole and strong seedling identification with machine vision technology,develops the corresponding software system,which laid a theoretical foundation and technical support for the further development of automatic seeding and thinning machine.The main contents of this paper include the following parts:(1)Build up the closed image acquisition system for plug seedlings.The system includes power supplies,computer,digital camera,fluorescent lamp,loading stage,hand-held illuminometer,and closed box.To ensure the even intensity of illumination of seedlings in each plug,the appropriate shooting height and resolution are chose.(2)Research of AdaBoost based plug tobacco seedling lattice image segmentation method.Firstly,the author preprocesses the collected image of plug in size of 10*20,containing histogram equalization,image normalization,median filtering,and plug image correction.Secondly,the Haar features of these images are extracted with AdaBoost algorithm,then,the training samples of plug tobacco seedling can be established according to the Integral Graph and Cascade Classifier,finally,training classifier can be obtained and the plug tobacco seedling lattice can be detected,tabbed and segmented.Experiment shows that,the accuracy rate of this method are 100%.(3)Research of multilayer perceptron(MLP)and convolutional neural network(CNN)based classification recognition algorithms of tobacco seedlings in plug.MLP and CNN model algorithm are used to train single,multiple and hole seedlings in plug lattice,by extracting the data of CNN training model of platform,the relationshipdiagram between iteration,accuracy and Loss function are drawn.Then these two algorithms are compared in accuracy by predictive set simulation experiments.Experiments results shows that the accuracy rate of these two algorithms are 96.75%and 97.58% respectively.Thus,the convolution neural network model algorithm are chose to study the number of tobacco seedlings in plug.(4)Research on machine vision based strong seedling identification algorithm in multiple plant plug tray tobacco seedlings.Firstly,coordinate locations of multiple plant plug tobacco seedlings were extracted by coordinate locations,then K-means clustering color segmentation,binary,median filtering are processed in the figure,finally,the number of multiple plant plug tobacco seedlings was determined by calculating the number of pixels in the single connected area.By comparing the five geometric characteristic parameters area,circumference,roundness,aspect ratio,and rectangularity,we are able to recognition and marking of strong seedlings.When tobacco seedlings are overlapped,the expansion and corrosion calculation of seedlings would be computed;When tobacco seedlings cross the border,the binary area of the image will be calculated,and calculated area less than 1/4 max area will be abandon.Experiments shows that,The accuracy of this method can reach 99.05%.Software system development of recognition tobacco seedlings in plug.The author develops a software system for preprocessing,classification and strong seedling recognition of tobacco seedlings in plug based on Qt platform and C++language.Results shows that the method based on machine vision is effective in detecting,segmenting,classifying and identifying strong seedlings.