| Fruit quality testing and grading is one of the important link in the fruits circulation and processing processes. How fast and accurate test the quality of fruit appearances for solving the problem of detection method lagging behind standard has become an increasing imminent and needs to be resolved. Because of computer vision method has advantages of high classification precision, automation and untouched inspection method, so it has better applicable value and application prospects applied to the exterior quality detection of fruits. The domestic and foreign scholars did lots of study work of computer vision used in appearance quality detection of fruits, and got a lot of researching results. These researchers made great contributions to the nondestructive detection method of computer vision. As a whole, however, some problems that need urgent solution still exist. First, most researches on exterior quality detection of fruits are observed in nearly globular objects such as apples, pears and citrus, but fruits of spheroidicity shape like papaya are almost unconcerned. Secondly, a lot of research works have been carried out for visual testing of static fruit, which has high identification accuracy but not strong in practicality. Correspondingly, there are lack of studys on the dynamic fruits. Based on a lot of technical literature both abroad and domestic, it was proposed exterior quality detection method of papaya which combined the computer vision, and computer visual detecting system of papaya was designed in this paper. And the MATLAB software was applied to the realization of all algorithms.The research contents and results were as the follows:(1) Image collecting system was built and carried through calibration works using a simple method. In order to acquire comprehensive information on papaya surfaces, an inclined plane with small angle was employed as the work platform for image acquisition of papayas. The inclined angle could be fixed by calculation making papayas roll purely along the slope. Papaya images were acquired dynamically by the inclined platform, and were cut for decreasing data.(2) The algorithms of the pretreatment methods, such as image grayness, image filtering, image enhancement, image segmentation, and edge detection were compared. And the experiment result showed that proper low image processing methods for papaya classification are selected.(3) Based on feature extraction approaches of fruits from domestic and foreign scholars, the advantages and disadvantages of the algorithms were analyzed and compared, and the proper methods of feature extraction of color, size, shape and surface defects of papaya were obtained. For feature data extracted of image color, CIE-L*a*b model was extracted describing papaya colors. Using the minimum out-connected rectangle and perimeter described size of papaya. Exterior characteristic of papaya was recognized by rotundity degree and Invariant moments. And the defective area of papaya was described by area ratio of injure region.(4) According to comparison of artificial neural network approaches, external quality classification of papayas was realized by BP network, RBF network and SVM network. And these neural network classifiers achieved between 75% and 80% classification accuracy. The major reason that the classification accuracy couldn't up to 100% was insufficient training and test samples.According to the results of classification, the design method based on this research has reasonable structure, practicable software and hardware. Accurate classification is obtained, also system performance can meet the actual need for exterior quality detection of papayas. |
PDF Full Text Request