Research On The Recognition And Classification Method Of Tea Sprouts Based On Computer Vision

The export demand for tea is increasing,among which high-end quality tea has been widely sought after by the public.At present,the famous and high-end quality tea,which are mainly green tea and white tea,are mainly picked manually.However,this kind of picking method is inefficient,and the labor cost is high,especially the seasonal picking is short,the number of short-term workers is huge,and the manpower requirements are high.Therefore,there is an urgent need to study automatic picking devices for high-end quality tea,and the most important thing is the classification and identification of tea sprouts and the identification of position orientation information.This is the basis of automatic tea picking and has become a research hotspot in related fields.This paper uses computer vision technology to classify and identify tea sprouts and determine the specific orientation of the tea sprouts.The specific research content is as follows:(1)Build its own dataset.The data collection work in this paper is to collect the data of the tea in the actual tea garden base scene during the tea picking season.The actual data collected will be divided into three types according to the shape difference of the tea sprouts.They are: types of fully-open tea sprouts,types of half-open tea sprouts,and types of one-bud two-leaf tea sprouts,from which to construct their own datasets.The positive samples in this article are three types of tea sprouts data,and the negative samples are the background area of the non-tea sprouts;after constructing the tea sprouts dataset,we will do a good job of labeling our own dataset.(2)An improved bilinear branch residual network Res Net18 model is proposed to realize the classification of tea sprouts.In order to enhance the model’s ability to perceive tea sprouts,a bilinear branch network structure is applied on the basis of the Res Net18 network.The upper branch focuses on global features,the lower branch does fine-grained detection,and focuses on local features;In order to reduce the amount of parameters and computation of the network model,this paper adopts the method of global pooling instead of the fully connected layer to solve the problem of over-fitting of the network calculation results,and has achieved good results in the verification of the data set.(3)An improved tea sprouts segmentation method of GMM model is proposed,which realizes the identification of the position and orientation of the tea sprouts.First,obtain the local area location of the tea sprouts in the actual tea sprouts data through the improved network model;Then,based on the Gaussian mixture model,optimize the parameters among them to obtain the binarized segmentation image of the local area of the tea sprouts;Finally,use the image invariant moment algorithm to calculate the centroid and orientation of the tea sprouts to obtain the accurate position information and orientation angle of the tea sprouts;It is verified on the actual image data to achieve better results,and obtain the position and orientation angle information of the tea sprouts,which provides a data basis for the subsequent grasping actions of the manipulator.