Recognition And Location Of High-quality Tea Buds Based On Computer Vision

Tea is a necessity for people's daily life,and also occupies a place in export commodities.While top-bracket tea is increasingly popular thanks to its good tea quality and fine processing quality.However,the original material of top-bracket tea,high-quality tea buds,is still in the state of manual picking.The low efficiency of picking and the high cost of labor have kept the price of top-bracket tea so high and prevented it from expanding production.Therefore,it is urgent to develop a set of fully automatic tea picking equipment.The most important and difficult problem in the process of developing this equipment is the recognition and location of tea buds.Considering the working environment of tea picking and the related progress of target recognition in recent years,the following aspects are studied in this paper based on the theory of computer vision:First of all,this paper proposes an improved PSO-SVM algorithm for segmenting high-quality tea buds.In the research process,firstly,the feature information of tea image is extracted,and these features are analyzed and compared to select the features with high difference between buds and background,so that the dimension of feature vector is reduced,which is conducive to shortening the image segmentation time.Then,an improved PSO-SVM algorithm is developed aiming at some drawbacks in SVM algorithm.Experiments prove that the improved PSO-SVM algorithm is superior to the traditional SVM algorithm in terms of processing speed and segmentation accuracy.In the experiments,the average processing time of the algorithm for the tea buds is about1 s,and the segmentation accuracy is above 94%,which can well complete thesegmentation of the tea buds.In order to better identify the picking points of the buds,the image needs to be post-processed after segmented,the residual impurities in the background are removed,and only the buds area is reserved.Secondly,this paper uses deep neural network to determine the picking point of buds.Due to the high efficiency goal of tea picking and the rapidity of picking point recognition,this paper finally chooses to build a bud picking point recognition model based on yolo algorithm.Before the model training,the bud samples are made and labeled.All the prepared images of the bud samples and the corresponding labeled information are fed into the bud picking point recognition convolutional neural network.And a good picking point recognition model is obtained after training.In the experiment,the picking point recognition model is applied to test the samples which not fed into the model.The accuracy rate of the experiment is above 84%,and basically fit the needs of picking high-quality tea.In the result images,the accurate positioning frame of the bud can be obtained,the skeleton of the bud is extracted in the positioning frame,and finally the picking point coordinates are determined.Finally,this paper designs the framework of full-automatic tea picking system,mainly including the following three key issues.First,the mechanical structure of the system is designed according to the environment of tea trees.The mechanical structure includes the selection of walking mechanism and picking manipulator,the determination of camera installation position and the selection of distance measurement method.The second is designed the work flow of the system,which is mainly about the cooperation between the mechanical structure part and the visual recognition part of the system.The third is designed the software interface required by the full-automatic tea picking system and the detailed software operation flow.These designs are of great significance for the final realization of the full-automatic tea picking system.