Research On Key Algorithms Of Tea Bud Recognition Based On Machine Vision

With the aging of the population and urbanization in China,the problem of labor shortage has become more and more obvious.At present,most of the tea harvesting machines on the market are non-selective and cannot guarantee the integrity of the tea leaves picked.The raw materials for the production of famous tea still rely on manual picking.If we can develop a picking robot that can automatically recognize young leaves,then the picking efficiency will be greatly improved.Recognizing and positioning tea buds with machine vision technology is the key prerequisite for developing an autonomous tea tree picking robot.This paper focuses on the pre-processing of tea images under natural light,the extraction of cutting feature points,and the method of obtaining the bud depth information based on binocular vision.The main work and research results are as follows:(1)Research on image pre-processing methods that use bilateral filtering for image denoising and point operators to improve image contrast.Filtering and denoising the collected tea images.By comparing the experimental results of different filtering and denoising,it is concluded that bilateral filtering can remove noise and retain useful feature information in the image as much as possible.Then use the dot operator to adjust the brightness and contrast of the tea image,the processing speed is fast and the contrast of the young buds and old leaves of the tea can be improved well.(2)The image segmentation algorithm based on color threshold is used to carry out theoretical and experimental research on the tea image bud segmentation.K-means ++ clustering-based image segmentation method,Otsu(Otsu method)threshold segmentation method,and color threshold-based segmentation method are used to segment the tea tree buds.Finally,by comparing the segmentation effects of these three segmentation algorithms,the segmentation method based on the color threshold can segment the tea bud areas better.(3)Aiming at the problem of small positioning of the shoots and stems of tea buds,the determination of the shear point and the method of coordinate compensation were studied.Using the geometric moments of the image to obtain the centroid of the bud area and the center of the smallest envelope rectangle of the bud area to obtain the feature points of the tea bud area.The time complexity and experimental results of the program are compared by comparing the two methods.The geometric moment of the image is selected to find the centroid algorithm of the bud area.Finally,according to the tea fresh leaf grade standard,when the robot arm picks tea buds,the axial position of the end effector is compensated by 25 mm.(4)Based on the SGBM semi-global matching algorithm,the method of obtaining tea depth information using binocular vision was studied.Based on the classic pinhole camera model and binocular camera model,Matlab software is used to calibrate the binocular camera.The depth map of the tea tree was obtained using the SGBM semi-global matching algorithm.The three-dimensional coordinates of the tea bud were finally obtained through the centroid coordinates and depth map of the tea tree bud area.It is verified through experiments that the visual measurement error is within 5 mm,which basically meets the precision requirements of tea tree buds.The three-dimensional coordinates of the tea tree buds are passed to the autonomous picking robot to realize the tea tree buds picking.This subject takes tea images collected under natural light as the research object,and uses image processing and machine vision technology to study the key algorithms of tea bud recognition,which basically realizes the autonomous recognition and accurate positioning functions of the tea autonomous picking robot.The research of this subject has certain significance to the development of automatic tea bud picking robot.