Real-time Inspection Technology Of Fruit Quality Using Machine Vision

Method for fruit quality inspection based on machine vision was developed to improve the quality of fruit in China, which was a sample for quality inspection of agricultural product using machine vision system. The method for surface area inspection of spherical fruits and the line-segment based digital image description method were new methods for other application of machine vision system.The results and conclusions were listed as follows:①A method for surface area inspection of spherical fruits with a real-time machine vision system based on stripe was introduced. A stripe was a narrow region of intresting (ROI) on the surface of a fruit, and its image was a horizontal line segment which width was just one single pixel. A stripe was considered as a part of a cylinder during area calculating. The height h and radius r of the cylinder was calculated with its vertical coordinate, and the central angleαof the stripe was calculated with the start horizontal coordinate and the end of it. The area of a stripe was acquired by calculating h×r×α. By summing all the area of stripes of a fruit, the area of ROI on a fruit was get.②A line-segment based digital image description method was developed, where a digital image was scanned horizontally and a group of continuous pixei with similar information was described as a line segment by recording the horizontal coordinate of the start pixel and the end to a node, and all the nodes on the same row were linked as a linked list called row list, all the row list of the image were stored in an array called image table by their vertical coordinate. The adjacent relationship between two vertical neighbor line segments was judged by the start and end coordinate of one line segment and that of the other. If the two line segments were adjacent, they were grouped to a ROI which was used to symbol an object. The operating of image filtering, object detecting, contour tracing was finished in one times. By comparing to laplace method, and the processing speed was improved 3 times.③Software vernier caliper method and Ellipse regress method for size inspecting of ellipse fruit were developed. The operating time and precision of these two methods and MER were compared, and it was showed that the operating time of Ellipse regress method was less than the others andthe precision of software vemier caliper method was best. The reason for the result was that the test point with vernier caliper agreed with china standard.④HIS color model, principal component analysis and Mahalanobisdistance method was applyed to sort a fruit by the color on its surface. 800 fruit images was use to test, and it was shown that the total error was only 1.75％. A new method was used to transform a pixel from RGB color space to HIS color space, where the speed was improved 20％。⑤A photometry model of fruit image was setup, and the gray value of a pixel was calculated ??using it. The gray difference between the real value and the calculated was used to get a gray difference image of a fruit, which was used to get the ROI, from which defect area, calyx area and stem area were segmented from normal area. The ROI was regressed to segmented defect area from calyx area and stem area. 1778 normal fruit images and 390 defect fruit images were tested, and the rate of correctness was 94％, the rate of error was 1.5％.⑥The test result of a fruit on different inspection channels was tested using F-test and t-test, and it was shown there was no significant difference on 0.01 level.The work presented above made its contribution to the first product line for fruit quality inspection and sorting using machine vision system in China.