Research On Computer Vision Based Recognition Methods Of Longing Tea Sprouts

China is the most important tea producing and selling country with theincreasement of tea production and sales annually in the world. As a big teaproduction province in China, the tea production of Zhejiang Province accounts formore than two-thirds of our country. Facing a huge requirement of tea, there has beena supply bottleneck. Recently, picking tea manually is the main approach which isinefficient, time-consumed and costly. There will be a massive labor gap during thepeak period of tea-picking annually because of the difficulty of hiring workers.Though mechanical method may improve the picking rate and ease shortage of labor,it may not be suitable for Longjing Tea because of lacking selection which willpicking both the old leaves and new leaves. Also, that method may reduce the qualityof tea which results from destructing the leaves and sprouts. Based on thesesituations, there has been a demand for an efficient, selective and less destructivemethod to realize the goal of picking Longjing tea automatically. One of the the keytechnology is researching on the detection and identification of Longjing tea sproutsautomatically.The purpose of the research in this article is combine the computer vision withtea picking which will make an efficient and accurate detection and identificationautomatically of tea picking by applying the classical algorithms in vision technologyon the tea images through the processing of computers. The method proposed by thisarticle is that it firstly digitized the appearances of Longjing tea, then the imagesegmentations technology is applied to get the regions of tea sprouts which will beused to get the features of sprouts to train the classifier after which we may make anidentification of tea sprouts. What is more, the num of types of Longjing tea sproutsis four which stands for one sprout, one sprout and one leaf, one sprout and twoleaves, one sprout and three leaves according to the differences of the tea sprouts’form. Experimental results show that the proposed method in this article could meetthe requirement of efficient and accurate Longjing tea sprouts’ classification andidentification which will make a theoretical groundwork and technical support for automatic massive Longjing tea sprouts picking and has a broad application prospect.The main researches are:Firstly, the application of computer vision technology on the research fields oftea and the image segmentation technology are briefly summarized. Several classicalimage segmentation algorithms have been applied on the Longjing tea images tomake a comparison on the experimental results. In addition, the classification modelnamed support vector machines is introduced.Secondly, considering the characteristics of tea images achieved on the spot, ahybrid initial segmentation method on Longjing tea sprout is designed. This methodfirstly utilizes the feature ExG-ExR (excess green, excess red) to tick the backgroundincluding part of old leaves, branches, soil and other non-sprouts regions throughthreshold technique and morphological transform. Then a pre-processed image hasbeen achieved. The gradient operators are applied on that image to get the gradientimage. After that, a threshold technique is proposed to make a adaptive binarizationon the gradient image to get the initial markers. In addition, in order to avoid over-segmenting the tea image, an area threshold algorithm is applied on those markers toeliminate some markers. Finally, the Meyer watershed transform is used to segmentthe pre-processed Longjing tea images with those markers to achieve initial fastsegmentation on those images.Thirdly, a fast statistical region merging algorithm using the color and texture ofthe initial segmented Longjing tea images is proposed for the reason that the initialsegmented tea sprouts images is always over-segmented.This method at first sorts theregions using a linear time algorithm to get the merging orders. Then merge the givencouples of regions based on statistics if they meet the proposed criteria. Theexperimental results show that the Longjing tea sprouts could be achieved from theimages completely by applying the proposed region merging algorithm on theover-segmentedFinally, owing to the differences between the features of sorts of extractedLongjing tea sprouts extracted, a discriminative model named support vectormachines is built to make a recognition on tea sprouts according to the currentgrading standards of Longjing tea sprouts. Class A stands for one sprout and onesprout and one initial exhibited leaf, class B stands for one sprout and one leaf and one sprout and initial exhibited two leaves, class C stands for one sprout and twoleaves, class D stands for one sprout and three leaves. The radial basis function (RBF)is used as the kernel function of SVMs through analysis and comparison fourcommon kernels which are linear kernel, polynomial kernel, radial basis functionkernel and sigmoid kernel. Moreover, grid-search algorithm is used to determine theparameters of RBF kernel. Finally, SVMs is applied on the classification of teasprouts. What’s more, detailed analysis and discussion of the experimental results isincluded.The proposed hybrid image segmentation algorithm in this paper hassuccessfully extracted the Longjing tea sprouts from tea images. Then a classifierSVMs trained by the features extracted from those sprouts is applied to make arecognition on some unknown samples. The experiment shows that a significantresult has been obtained which has provided a theoretical basis of picking Longjingtea sprouts automatically based on compuer vision.