Research On Tea Recognition Method Based On Machine Vision Features For Intelligent Tea Picking Robot

Tea is an economic crop originated in China.With the development of social productivity and the increasing number of drinking population,its market value is constantly improving,and the emergence of famous and excellent tea has improved the brand benefits of tea and increased the value of tea products;famous and excellent tea has strict requirements on the part of tea picking in the tea production cycle,and only tea buds reaching a certain grade can be picked and made Famous tea.Because mechanized tea picking will destroy the quality of tea,today’s famous tea picking is mainly manual work of tea farmers,which has low picking efficiency,high labor cost and requires tea farmers to have rich experience in identifying buds.Therefore,it is necessary to study a scheme that can identify tea buds intelligently and carry out three-dimensional location picking of tea buds independently without manpower.The main purpose of this paper is to develop an intelligent tea picking robot based on machine vision.The self-developed tea picking robot can cut and collect the tea shoots in the field environment.The working platform of tea picking robot is equipped with an image acquisition system,which can recognize the tea buds by digital image processing algorithm,and realize the spatial positioning of tea buds by "one eye multi position" image matching.The ultimate goal is to realize the intelligent tea picking robot to pick the tea shoots efficiently.The main contents of this paper are as follows:The hardware design and construction of the intelligent tea picking robot platform developed independently.Each part of it is introduced.The overall structure of the tea picking robot can be divided into: the customized bipedal crawler robot realizes the mobile ability of the robot and provides remote sensing control function,while the external ceiling and cover tarpaulin ensure that the robot’s internal environment is separated from the external environment,so as to ensure the feasibility of tea picking under outdoor conditions;SCARA robot is the executor of the tea picking technology of the robot,cooperating with the installation The image sensor,tea cutter and related program code at the end of the actuator can realize the intelligent picking of the target tea.The research of tea related theory in the early stage of tea recognition.The requirement of famous and excellent tea grade(only the super and first grade tea have the picking demand)determines the inevitability of image recognition in the intelligent tea picking robot.For the color analysis of tea bud experimental image,it shows that there is a large difference in R,G,h and s color channels in RGB and HSV color models,which provides a theoretical basis for the threshold segmentation of color difference image.The algorithm of tea bud recognition based on machine vision image recognition.A new method of tea bud recognition based on improved color difference method is proposed: after the image cutting,the background of each part of the image is filtered by OTSU method,and then the tea contour is extracted by color difference method based on the difference of brightness in HSV space;the binary tea leaf contour image is denoised and filtered by first-order Gaussian filter;the opening and closing operation in image morphology is used to eliminate the noise Remove impurities.Based on the color difference between the young leaf and the old leaf,the accurate position of tea bud image can be obtained after image segmentation.In the actual experiment,the rate of top bud detection is 90%,and the single algorithm completion time is within 1s,which meets the recognition requirements of only picking robot.Study on the scheme of spatial location of tea shoots.It is necessary to use SCARA manipulator and single CMOS image acquisition equipment installed at its end to control the robot to collect tea images for many times to form a "one eye multi position" three-dimensional vision system.After getting the camera’s internal parameters by Zhang Juzheng’s method,the error of the actual focal length entry parameters is less than 0.5%,and then the tea shoot image is matched after feature extraction;it is found that choosing different image matching models will produce different results,and the module matching based on gray image will produce 15%error due to its poor robustness;and using SIFT algorithm,the parameter k = 0.05 is set,r = 10 to match the feature vector on the image,we can get a good matching effect,the matching similarity of the example image can be more than 90%.