Stripe Surface Defect Image Recognition by Supervised Tree ISOMAP and Incremental GRNN

摘要

The new isometric mapping dimensionality reduction algorithm with Incremental Generalized Regression Network has been primarily recognized for stripe surface defects images with the typical characteristics of complex texture, non-uniform image size, asymmetrical number of sample classes, variation illumination environment This method is suitable to resolve the problem of "short circuit", stored internal structure in lower dimension space. In addition, the algorithm parameters influence on the stripe surface defect images is greatly reduced. The finally experiment results show that it is effective and efficient for stripe surface defects with the highest recognition rate of stripe surface defect can reach to 97%, and the highest recognition rate of complex stripe surface defect can reach to 74%.