Abstract: The image of the information about the shape of defects in the shuttlecock process ball was obtained after pretreatment of the images of the shuttlecock process ball. Then, the five polarradius moment invariants, circularity, and the difference between the cylindrical part and the center of the ball in processed images were extracted. The seven normalized eigenvalues were input into the BP neural network, based on LevenbergMarquardt algorithm, and the trained network was used to be a detector to recognize the image pattern. Experimental results show that a relatively good recognition performance can be achieved to detect defects in shuttlecock process balls.

Key words: shuttlecock process ball; defect detection; polarradius moment invariant; BP neural network