A Study On Pear Quality Testing And Grading System Based On Computer Vision Theory

The research of pear automatic grading system is significance. Currently, people usually use manual and mechanical method in grading fruit. Manual classification is inefficient and vulnerable to subjective factors; mechanical grading has few functions, and may break the fruit. For pear, it is urgent to design a fast and efficient classification system.The emergence of computer vision technology makes a vast space for the achievement of the high efficient, non-destructive classifying. This paper aims to computer vision method, make research on the automation classifying system by the aspect of shap, weight, color and defects, so that to facilitate the transport and storage of agricultural products in the management, and improving its commercial value. The system includes for parts ,which is image acquisition, preprocessing, feature extraction and classification. Pictures used in this research are photographed by camera in a self-made light box. We collect four images on a single pear inorder to get its full surface in a white background.Firstly we do some preprocessing procedure such as color space conversion, filtering, image segmentation and edge detection. Then we make feature extraction by the image. In the process of removing image background we use global thresholding method in a transformed color space, which make result better than the traditional gray threshold segmentation; in the process of eliminating spots pear surface, we make a comparison of two method, which is the HSV dynamic threshold segmentation and morphological, both methods can achieve acceptable results, while the latter one is more efficient.By the study of National Agricultural industry standard "pear appearance grade standard", we should focus on the defects and shape evaluation. This study attempts to quantify the national standard in color, shape description, texture and other features on the defect. With these acquisted characteristics, we use BP neural network classifier, grading multiple varieties of pears in appearance and achieved some delightful results.