The invention discloses an image identification method used for a wheel hub sorting system. The method comprises the following steps of A, performing grey level transformation on an acquired image; B, performing filtration processing on a gray level image to improve the resolution of the gray level image; C, acquiring a feature region on the gray level image, and eliminating a false feature region in a determined feature point; and D, comparing the acquired feature region with a preset feature region database to obtain an image category. According to the method, the deficiencies in the prior art can be overcome and the identification accuracy of machine vision detection is improved.