Study On Apple Grading Method Based On Machine Vision

Fruit classification is a system of measures of according to the provision of the state standard, according to the quality grade, to determined the purchase price and sale price by the levels of fruit.In the apple grading system of China, classification are mainly based on apple surface defect and fruit shape size. This system combines the technology of machine vision and digital image processing technology, adopts a image-processing platform to ap-ple grading research that Microsoft Visual Studio2010 combine with OpenCV. Firstly, de-fect segmentation by SUSAN operator; secondly, according to the shape feature provides a method for minimum circumscribed circle diameter detection of Apple size; Finally, provide a reference basis for the realization of apple grading. The main research contents of this subject are as follows:(1) Building a suitable experimental platform and choosing best image background of apple collecting,then preprocessing with the apples of camera collection, including image binarization, background segmentation, smoothing, edge extraction and so on, to extract the edge contour image of apple. Provide a good experimental platform and previous data supporting for apple grading research.(2) Apple will have some collision and damaged in the process of transportatio n, we need to test the color feature of apple surface defect before apple size gradin g. According to the color feature extraction of RGB and HSI color feature parameters, we set feature extraction of image grayscale and defect segmentation by SUSAN op erator, to achieve the effect of defect detection of apple samples.(3) According to GB/T10651-2008 standard and the spherical apple shape, put for-ward a method of based on the minimum circumscribed circle diameter to detect the size of the apple according to apple diameter size.(4) This study select the same species of red Fuji apple buying from the market ran-domly, respectively choose 30 small fruit,30 medium fruit,30 large fruit as test sample, contrast test results with artificial results, test the accuracy of 92.22%.