Grape Grain Size Non-Destruction Detection Based On Machine Vision

According to the United Nations Food and Agriculture Organization (FAO) statistical database data, it is showed that as of2011, the production of the grapes in China ranks first in the world. At present, the classification of grapes in our country is mainly artificial. There are many disadvantages such as subjectivity, low working efficiency and the fatigue caused by the long time working, and so on. With the increase production of grape, and improvement of the quality requirement, the grape quality classification is getting more and more important in the consumer market, the non-detective automatic classification of grape has become an urgent problem, which can improve the quality and classification standard of the grape as a whole.In this paper, the non-destructive automatic size classification of grape is proposed based on the machine vision technology. The method is consisted of two parts of machine vision system and the grape image processing. The key point of machine vision system is the appropriate choice of the light source and camera. In this paper, white annular light source, yellow Led surface light source and near-infrared annular light source is experimented to found that the near-infrared annular light source is the best choice, because of high transmittance, combined with the black and white CCD camera, the grape image with clear outer boundary and the inner edge of single grain can be obtained, also with the interfering information concealed.For the grape processing part, with the comparison of the brightness equalization methods and different edge detection operator, the color gradient edge detection operator is found to be the best choice of the pre-processing, by the color gradient edge detection operation, not only the image with clear edge of single grain can be obtained, but also the interfering information caused by the inner stems or vines of grape surface can be concealed. Therefore, the process of the grape image processing is:first, the single grain edge of the grape bunch is obtained by the color gradient edge detection, then, three appropriate points is get along the edge obtain, the rotation direction and the rotation angle of the three points along the edge can be calculated with the relative location between the first point and the end point of the initial arc which is calculated by the rotation matrix; then, the single grape edge sampling point matrix can be calculated according to the principle that grey value of the points on the edge is greater than the grey value of the points in other area; finally, the grape edge can be fitted with the method of least squares ellipse fitting, with the best ellipse equation obtained, which contains the characteristic parameter value of the grape, in order to realize the automatic size classification of grape.