Surface Defect Detection Method for Strip Steel Based on a Few Defect Samples
During the production process,strip steel is affected by factors such as process or equipment,resulting in surface defects such as crush and drops tar.The generative adversarial network model is proposed that only requires a few defective samples to training.The model learns the distribution characteristics of a large amount of defect-free image data,and then adds a few defect images for comparison,the model can learn the distribution of defect-free images and defective images,which is more conducive to defect detection.In order to improve the stability of the model,the pull-away term(PT)and diversity ratio factors are added to the loss function.And add the norm of optimized features to sparse the model and reduce the interference of useless features.The method proposed achieves an accuracy of 95%for surface defect detection of strip steel.
万方数据
维普
