Objective: In order to solve the problems of low efficiency and high false detection rate of current fruit classification detection methods, a fruit classification system based on machine vision was designed with apples as the sorting object. Methods: The apple image collected in real time was preprocessed, the improved Canny edge detection algorithm was used to extract the edge, and the radius of the apple cross-section was obtained by fitting the edge coordinates with the minimum peripheral circle method. The acquired RGB image was converted to HSI image, and the proportion of red region is calculated according to the range of H component to determine the color of apple. Count the number of pixels in the area, and calculate the area and circumference of the apple respectively to calculate the roundness of the apple. Combined with three characteristic values of diameter length, color and roundness, apple was graded comprehensively. Results: Through the test of 50 apple samples, the error range of fruit diameter measured by the fruit grading system and manual sorting was within ±1.5 mm, the color characteristics of the samples were consistent with the actual appearance of the apple, and the result size of the roundness value was consistent with the actual shape. Conclusion: The system can meet the demand of apple classification in actual production and help to realize the accurate identification of apple grade.