Aiming at the problems of similarity between product shapes as well as its geometric distortion and edge blurring in production line

a multi-species product recognition algorithm based on improved invariant moments combined with probabilistic neural networks（ PNN） is proposed. The marginal feature information of the product is extracted through image pre-processing

the obtained marginal features is described using the improved invariant moments. It is proved that the improved invariant moments not only are invariant when the image is shifted

rotated and scaled

but also the influence of the distortion scale factor on the invariant moments in the discrete state is eliminated. The PNN classifier judges the type of products by training standard parts with different forms and proportions. The recognition accuracy of the product standard parts reaches 100 %

and the accuracy of the identification of the product parts reaches 98. 61 %. The proposed algorithm is robust and reliable. The experimental results show that the proposed algorithm is an effective identification method to deal with multivariety products in production line

which is valuable.