Research On The Non-destructive Testing And Classification Technology Of Blood Oranges Based On Computer Vision

With the population aging in the 21 st Century, the labor costs increase continuously. The traditional artificial sorting efficiency could be improved significantly so that the classifying costs decreased and the profits of fruit industry increased by using computer vision non-destructive test technology, Meanwhile the damage to fruits during the artificial sorting process could be reduced,and the competitiveness of the fruit industry could be enhanced. Based on this background, this paper makes a study of the methods of non-destructive inspection and classification for blood oranges based on computer vision by using digital image processing technology.The image which acquired in the small studio was performed the smoothing process and the noisy Image was performed the completion process by using mean filter and median filter technology. We use the Laplace operator method and gradient method to sharpen the image, so that edge information and the outline feature of image was enhanced. This paper introduced three classical image restoration method including Constrained least squares filter restoration, Wiener filter and Lucy_Richardson filter. and the blurred image due to the objects which performed uniform linear motion was repaired by constrained least square filter, Lucy_Richardson filter and Wiener filter. At last, constrained least square filter was chosen to applied in image restoration process. The Image preprocessing technology of blood orange image above-mentioned lay a foundation for the further segmentation and analysis.The whole pixel points of blood oranges could be calculated by the Application of Binaryzation Algorithms of the blood orange images after preprocessing, and the size of blood orange could be gained according to the unit pixel area of the standard reference. The Roberts operator, Prewitt operator and Sobel operator was used for the binary image edge detection of blood oranges, finally the edge information which gained by Sobel operator was applied to calculate the perimeter of blood orange by means of the same standard reference above-mentioned. The image segmentation was performed by the Otsu method, minimum-error threshold and Maximum Entropy Method, at last we choose the Otsu method to get the segmentations of colored parts and not colored parts of blood orange surface, and the maturity of blood oranges could be obtained by calculating the colored area proportion according to counting pixel points.This paper introduced a concept of fuzzy clustering to efficiently classify the quality of blood oranges. The fuzzy statistical method, dualistic contrast compositor method, the trichotomy method and the expert experience method were compared. Finally the expert experience method was adopted to determine the membership function, and these three characteristic parameter values of the membership function were corrected so that the membership function was updated.According to the membership function, the blood oranges were classified to four levels: very well(A), well(B), common(C), bad(D).The size, the perimeter and the maturity of blood oranges which are the three important classification characteristic parameters could be gained by image acquisition, image preprocessing and image segmentation technology, and these three characteristic parameters could be used to efficiently classify the quality of blood oranges. The non-destructive test and classification of blood oranges based on computer vision was investigated in this paper, and the detection and classification result is satisfactory. This research lay the foundation for further automated non-destructive testing of blood oranges.