Research On Tea Visual Recognition Technology Of Tea-Picking Robot

With the improvement of people's living standard,more and more high to the quality of life requirement,growing demand for tea,but now tea mainly by hand picking tea plucking,weakness of labor market and labor cost is high,so the use of robots to realize automatic tea picking is the inevitable development trend of tea production,therefore to develop intelligent robot tea plucking related technology research has the vital significance.Tea identification technology is the precondition and key technology to realize automatic picking of robot.Research purpose of this article combined machine vision technology and tea,tea in a natural environment as the research object,the collected tea image after image pre-processing,image segmentation,color and shape feature extraction,PCA dimension reduction and K-means clustering algorithm to identify steps,such as tea recognition model,to implement the tea image recognition based on machine vision.At the same time,the reliability and feasibility of the tea identification model are verified through test experiments,which provide technical support for the intelligent picking of tea robot.The main research contents are as follows:(1)Analyze the influence of different color Spaces and filtering methods on tea image segmentation,and find the color space and filtering method which is best suited for tea image segmentation.First analysis the tea image in RGB,HSI,common Lab form of 3kinds of color space,separate each color channel color space,image contrast tea in gray image of each color channel by HSI color space S channel image segmentation processing of tea.Then,the image of tea is filtered in frequency domain and airspace to make noise analysis,and compared with the segmentation effect of the filtering method,the adaptive smoothing filter is used as the tea image filtering algorithm.(2)Based on the characteristics of the tea color respectively by Otsu algorithm,iterative optimal threshold method,genetic algorithm and watershed algorithm to the tea image in RBG,HSI,Lab color space segmentation comparison.In this paper,an improvedwatershed segmentation algorithm is proposed,which is based on the watershed algorithm.The first segmentation is to use the improved Otsu algorithm,which is the ratio of the inter-class variance and the intra-class variance as the threshold criterion.The ratio is the largest and the threshold is the best.Then,the segmentation algorithm is used to divide the image after the first segmentation,to remove the misjudgment and to divide the result accurately,so as to obtain the image of the connecting region(3)By extracting the tea color and shape characteristic as the tea recognition characteristics,color characteristics for the G,B,H,S,2 r-G-B,G/B,B/G,and B eight kinds of color features,shape features for 7 Hu matrix,the same amount of shape features.To 15 d color and shape characteristics of the normalized processing,then by using the method of PCA dimension reduction processing,get vector,principal component analysis to calculate the corresponding eigenvalue vector characteristics description.Finally,K-means clustering algorithm was used to train the model to verify the accuracy of the model.The correct recognition rate of the new leaf was 89.24%,and the recognition time of each image was 0.156 second.