Research On Automatic Segmentation Of Grape Leaf Image Under Natural Condition

In recent years,with the rapid development of computer technology,machine vision technology is increasingly being used in the field of agriculture,and the video surveillance equipment is used to monitor crop growth,obtaining its growth information in order to analyze the crop growth status and disease information.As the leaves are the most significant and observable part of the grapes,but the grapevine images and videos collected under natural conditions will affect the segmentation of the leaves because of the complexity of the background,the disease of the lesion,the different illumination and other factors.Therefore,this paper will study the grape leaves automatic segmentation algorithm,which will provide a solid foundation for the follow-up study of the diseases.The main research work of this paper includes:(1)Through a large number of green leaves of color information statistics,the main distribution of G / R components of green leaves and the main distribution of the chromaticity factor a * are obtained,using this information to complete the automatic selection of seed points for the algorithm.The probability density distribution of the blade and the background is estimated by using the mixed Gaussian model.The energy function of the pixel feature is established based on the Markov random field.The blade is realized by solving the energy function minimization.The results show that: for leaf images in different time and weather,single G / R and a * have good effect,the segmentation accuracy is 86.74% and 92.38% respectively.In addition,segmentation effect will be further improved if these methods are combined as a double feature,the segmentation accuracy is up to 95.03%.(2)For the non-normal leaves(diseased leaves,etc.),as its leaf area and the normal leaves of the color difference between the larger,it can not be completed by the above method of automatic segmentation and introduce target detection technology.Target detection technology usually uses an image as input and output several boxes,called the detection box which contains the identified targets.The central region of the selected detection frame as foreground and to establish a seed point of prospects parametric model,and parametric model established outside the detection pixel of the frame,the foreground parametric model and parametric model area subtraction,i.e.,the remaining portion of the initial background parametric model,again Through iterations,gradually seek more accurate background and foreground parametric model.Finally,the blade region of the whole image is segmented by the method of energy minimization in Markov random field.The experiment shows that this method can solve the problem of automatic segmentation of complex background and multiple lesion leaves when grape leaves are used as the segmentation object.