Research On Combinatorial Feature's Reduction, Cluster And Recognition Algorithm For Board Strips Surface Defects

The board strip is one of the important products in the steel industry used in automobiles, electrical home appliances, shipbuilding and ministry of aeronautics and astronautics industry.Its surface quality can has a direct effect on the product quality. The board strips surface defects inspection and classification can dedicate not only to the theory but also to the economical values.In this paper, we put forward the combinatorial feature's reduction,cluster and recognition algorithm for board strips surface defects, because present board strips surface defects inspection systems can't to recognize images of new defect classification and deal with error-recognized images effectively.At first,intent to accurate describe defect image,various features of board strips surface defect image are extracted. Then optimum combinatorial feature are selected by separable feature discriminant base on distance of classes. Secondly the optimum combinatorial feature which is imported into SOFM neural network is reduce dimension.We adopt an improved near neighbor's cluster method to cluster the optimum combinatorial feature for recognition. The experiment indicates,this method combine feature reduction of neural network with feature near neighbor's cluster and have realized the mutual supplement with each other's advantages of those. The method has raised the recognizable rate of defect image.The system still has the expanded function of defect recognition model.Through the model revise and expand,the system can recognize the error-recognition images and the unknown defect surface images which expands the ability of board strip surface defect recognition system..In this paper, we just use the above theory and method to classify six type defect surface:edge sawtooth, welding seam, mixed material, wrinkles, abrasion and slice. By experiments, we find that the theory can reach average recognition rate of 98.3% and the recognition rate of board strips surface defects images can be improved afeter adjusted the recognition model.All research works are based on an independently development software environment of classification and recognition of board strip defects. It has established solid foundation for the further on-line inspection and classification research of board strip defects.