Research On The Vision System Of Tomato Harvest Robot

Modern agriculture requires agricultural machinery automation andintellectualization，and the emergence of harvest robot lays the foundation for it. Inforeign countries, people do study in the respect earlier and a lot of researchresults have been achieved. In domestic, research in this field started later, alot of technology also needs to be improved and it’s a long way until practicalapplication. The paper is based on the mature tomatoes in natural environment.According to the characters of tomato, a tomato harvest vision system isstudied. In this paper, the research content and the method is as follows:(1) Firstly, the paper preprocesses the tomato images picking in the naturalenvironment including filtering, histogram equalization, sharpening and so on.The illumination in the environment has influence on image segmentation.According to the theory of color constancy, the paper extracts tomatoillumination irrelevant images to eliminate the influence of illumination.Segmenting tomato illumination irrelevant images based on the boundary andthreshold segmentation method and analyze the results. At last this paperadopts OTSU method.(2) Considering tomatoes are shaded or close to each other in the actualsituation, this paper proposes a watershed algorithm based on distancetransform for separation of multiple objective. Experiment results show thatthe method is successful to separate multiple targets. But there is excessivesegmentation phenomenon and the method needs further improvement.(3) Analyzing the features of tomato, extracting the color, shape and texturefeatures of the separated tomato images, getting training samples and testingsamples, the paper adopts SVM method to classify tomatoes and background.Through the experimental the paper analyzes recognition accuracy of differentkernel functions. This paper uses radial basis kernel function of support vectormachine to recognize tomatoes and the recognition accuracy reaches up to95.36%. Then the paper adopts cross validation and the grid search method tooptimize parameters and the recognition accuracy reaches up to96.89%.(4) Tomato spatial orientation is based on binocular stereo vision theory.The paper adopts Zhang Zhengyou camera calibration method to calibrate the internal and external parameters of binocular camera. This paper presents astereo matching algorithm based on the center of mass and uses the matchingconstraint conditions to match the binocular tomato images. Finally, the threedimensional positioning determines the depth of the target. According to thecamera calibration parameters, the paper obtains the target depth based on theparallax method. Experimental results show that when the working distance ofthe camera in300-400mm, the error of centroid depth is smaller to positionthe tomatoes optimally.