Research On Fast Tracking And Recognition Of Fruit For Apple Harvesting Robot

Fruit and vegetable industry is a pillar industry of agriculture in China.Fruit and vegetable are mainly harvested by hand.With the development of economy,the number of people engaged in the agricultural industry continues to decrease,and agricultural labor is insufficient.Picking period is short and fruits need to be quickly picked.Therefor,it is urgent to carry out the research of the harvesting robot.Instead of hand picking,machines should be invented to realize the mechanization and automation of fruit and vegetable picking.Under the support of National Natural Science Foundation of China and the project of‘Research on an efficient harvesting method of apples based on fast visual servo control in multi-illumination environment '(project number:31571571),the visual system of the apple harvester was studied.Focusing on the mature apple as the research object.The research studied the fast tracking recognition of the apple during the dynamic picking process of the harvesting robot.The main works are as follow:1.Design of the visual software.According to the main principles and basic framework of machine vision,build the experimental platform used in this study,and determine the hardware structure and the visual software of the experimental platform.2.Research of apple image segmentation method.To track and recognize Apple,the apple harvesting robot must Segment the target apple out the image first,and this is the basis part of this study.This paper introduces three different color spaces and commonly used segmentation methods.By comparing,finally chose k-means clustering segmentation in Lab color space which is irrelevant to hardware device and of good uniformity.There will be noises and holes in the segmented apple image,thus mathematical morphology method,threshold area preserving method and diffuse water filling method are used to process the image after segmentation,and finally an effective segmentation image of the apple is obtained.3.Research of apple tracking and recognition algorithm.The compression tracking algorithm is used in this paper,which is a method of using compressionfeature to describe and model the target apple and the background.During the process of image processing,the size of the tracking window can not be adaptively changed,and the speed of the classification and update of naive Bayes classifier cannot meet real-time requirements,so the algorithm has been improved in this paper:(1)The SURF algorithm is used to match the apple template in the previous frame with the image of the next frame.According to the matching feature point pairs,Parameters of the scale transformation can be calculated and the size of tracking window can be adjusted.(2)Using SVM classifier to classify the compressed features and update data on-line.The real-time performance of the algorithm can be improved by reducing initial training time and enhancing classifying performance of classifier.4.Experimental analysis of tracking and recognition algorithm.To verify the feasibility and effectiveness of the algorithm,the experiment mainly tries to get the accuracy and real-time performance of.In the SURF feature matching experiment,the matching time of feature points in each image group is 19.9ms on average.In the experiment of apple tracking and recognition,The central location error of apple fruit is within 15 pixels for the image sequence pixel of 480640?.In terms of the time used in the algorithm,The classification and updating speed of the improved algorithm is increased by 49.4%.The average processing time of each frame is0.0253 s,and the speed of tracking and recognition is increased by 24.2%.The effect of accelerated optimization is obvious,which can meet the requirements of apple harvesting robot.