Study on Defect Extraction of Pears with Rich Spots and Neural Network Grading Method

In order to study the feasibility of the computerized grading system of pears with rich spots, large amount of pears are put orderly in a light box, where images of different aspects of them are captured by cameras which are connected to the computer. Additionally, two methods are proposed to remove the spots on the surface of pears in the image to reduce their effect on defect detection. And advantages and disadvantages of the two methods are discussed. One method works by applying adaptive threshold and the other works by combining filtering with edge detection. In reference to the National Standard of China, grading based on shape and defect is studied respectively. ANN (Artificial Neural Network) model which colligates the information of shape, color and defect is established to grade pears comprehensively. The method of adaptive threshold has better effect when it is used in processing pears with rich spots, such as Laiyang pear, whereas its executive efficiency is slightly lower. Yet to the species the spots of which are not so conspicuous, such as Huangjin Pear and Fengshui Pear etc., the method which combines filtering with edge detection is supposed to be applied for better performance. The results of the grading by ANN Model based on shape, defect and comprehensive quality respectively reach the accuracy of 87.5%, 92.6% and 90.3%.The methods proposed to remove spots and the model constructed for grading and recognition have positive significance to grading pears with rich spots by appearance.