Automatic Grading Method Of Kiwifruit Based On Mechine Vision Technology

In order to realize the automatic grading of kiwifruit, in this paper, we took “Hayward”and “Qin Mei” two kiwifruit varieties as the research objects, and carried on the fruit surfacedefects detection and fruit Sensory index grading from two aspects.This paper’s research contents and conclusions are mainly as follows:（1） A method of surface defects detection in kiwifruit grading system. After investigationand statistics, we found that Kiwifruit surface defects mainly include bruising, scratch andburning. We set up a kiwifruit machine vision test system with the near-infrared light source,and removed all kinds of noise of original images with medium filtering; then we get theoptimal threshold by image analysis; at last the surface defect area was get by imagesegmentation. The experiments indicate that we can effectively extract kiwifruit defects suchas bruising, scratch and burning using near-infrared light sources, which could help to avoidthe light spot area under traditional light source.（2） Automatic detection of kiwifruit defects based on near-infrared light source. Amathematical model that expresses the relationship between Near-infrared light intensity andautomatic threshold for automatic kiwifruit surface defect detection was established. Byapplying different levels of Near-infrared light intensity to machine vision system,268imageswere collected. Then the images were processed with MATLAB using the method to detectkiwifruit defects based on Near-infrared light source.The obtained268sets of data onAutomatic Threshold T₀and Manual Threshold T₁were divided into20groups according todifferent aperture and light intensity. After processing data, a series of linear equations aboutthe relationship between Near-infrared light intensity and Automatic Threshold T₀, withfunction fitting coefficient of R2>95%was obtained. Finally, relationship between T₀and T₁was analyzed according to the effectiveness of image processing results and constant P wasintroduced to revise Automatic Threshold T₀. Thus, a mathematical model needed to gainkiwifruit defects detection threshold, namely Model Threshold T, was established.（3） Kiwifruit grading method based on the statistics of external dimension. In this partwe had carried on the investigation of kiwifruit grading indexes and statistic analysis of fruit data. After analyzing the relationship between the shape size, image area and single fruitweight, we found that the best fitting results was the relationship between single fruit weightand fruit long axis. The long axis and short axis of the fruit were get by using the minimumcircumscribed rectangle method, and the image pixel values is associated with the real longand short axes of kiwifruit. By data analysis, We found that the fitting effect of long and shortaxes and the image pixel values of Hayward kiwifruit are better, the fitting effect of the imagepixel values of the long and short axis and single fruit weight of Qin Mei kiwifruit is better.So the ratio of short and long axis can be used to identify the fruit shape, which was proposedto distinguish normal fruit and deformity fruit.（4） Based on principal component analysis （PAC） kiwifruit appearance featureclassification method research. The grading indexes of548Hayward kiwi fruits are weight,long axis, short axis, image long axis, image short axis, the difference of long axis and shortaxis, the sum of long axis and short axis, the ratio of the long axis and short axis, the ratio ofimage of the long axis and short axis, using PAC to them. We extracted3principalcomponents. With different variance contribution ratio of the characteristic value as theweighting coefficient, using the function Y=m1F1+m2F2+m3F3to calculate the score of eachsample. The last, all of the samples were ordered.