Research On Feature Extraction Method Of Strip Surface Defect Classification

The steel industry is one of the pillar industries in China,and hot and cold rolled strip is the main product form of steel.In 2017,China's annual steel output was 1.1 billion tons,of which 46% was hot and cold rolled strip.However,due to issues such as raw materials and manufacturing processes,the surface of the strip often has some defects.These defects not only affect the aesthetics of the product,but also the product's corrosion resistance,abrasion resistance,and fatigue strength.Due to the complexity and diversity of strip surface defect textures,it is difficult to classify strip surface defects.In order to improve the recognition of strip surface defects,various types of features need to be mixed;the use of a large number of features will increase redundancy,reduce computing efficiency,and it is difficult to meet the actual efficiency requirements of production.In addition,the complex appearance of defects leads to weak discriminability and weak robustness of partially extracted image features.Therefore,it is important to select features with high discriminative power from the original feature set.In this context,this paper studies the pre-processing techniques,feature extraction and feature selection techniques,and the final classification methods used to process images of strip surface quality defects.main tasks as follows:(1)This paper proposes an adaptively optimized Gabor filter based on a whale swarm algorithm.First,anisotropic diffusion filtering is used to suppress false edges in defect pictures,and then the maximum inter-class difference between different types of defect features is used as the objective function.The parameters of the Gabor filter are optimized variables.The whale swarm algorithm is used to find Gabor parameters Excellent,then the obtained Gabor features are fused,and finally imported into the classifier for classification.The experimental results show that the method is distinguishable and robust.For common strip surface defects: punching,stains,scraping edges,black oxide strips,and scarring,the final classification accuracy can reach 97.5%.(2)This paper proposes a multi-scale global local binary mode that introduces Gaussian difference space to classify strip surfaces.First,according to the human visual attention mechanism,Gaussian difference space is used to pretreat the surface defects of the strip.Then multi-scale improved fully local binary mode is used to extract features from the pre-processed picture.Since there are many redundant features in the extracted features,non-linear popular learning is used to reduce the dimensions of the features.Finally imported into the classifier for classification.The experimental results show that the method has good discrimination.For the common strip surface defects: punching,stains,scratches,black oxide strips,and scars,the final classification accuracy can reach 98.5%,which is better than the current traditional method..