Abstract

As the detection of the major and minor axes is an important indicator of potatoes grading, a fast, accurate and non-destructive method for sorting potatoes is in great demand. In this study, the technology of machine vision was employed to detect the major and minor axes of potatoes. A method to detect the size of potatoes is presented based on the centroidal principal axis of potatoes. Firstly, the color space of the image was converted from RGB to HSI, by applying the Otsu, a method of single threshold segmentation, to the H component values of the HSI image, binary image was gained through extracting potatoes from the gray images of H component values, which were preprocessed with the denoising method of a mean filter. Through filling and morphological erosion, the integrated potatoes image was obtained. Secondly, the potato image was rotated to a certain angle, which was calculated by the character of the principal moments of inertia of potato. The length and width of the external rectangle for potato in pixel-level are considered as the feature values of its major and minor axes. Thirdly, through regression analysis, the correlation coefficients of the major and minor axes are 0.9656 and 0.9166, respectively. For major axes,  the maximal absolute error and relative error between the practical value and predicted value are 4.20 mm and 6.08%. For minor axes, the maximal absolute error and relative error are 3.18 mm and 7.42%. It requires only 1.38 s forprocessing a image by the centroidal principal axis, however, the time is 15.72 s for MER. Experimental results show that, with accurate and fast properties, the proposed potato’s size detection method can provide a basis for on-line detection. Since the fact that the MER takes longer time to get the major and minor axes, in this study, the detection method based on the centroidal principal axis is introduced, which is accurate and fast. Besides, the method can be widely used in the dimension detection for other agricultural products.

Key words: Potatoes, machine vision, the major and minor axes, centroidal principal axis.