Research On Detecting Method Of Surface Defect Of Cold Rolled Sheet Based On Convolutional Neural Network

Cold rolled sheet has a very important role in production.The surface defects of the cold rolled sheet not only affect the appearance of the product,but also reduce the product quality.The detection and classification of surface defects image can eliminate defective products in time and adjust the production process according to the types of defects,thereby improving the quality and production efficiency of cold rolled sheet.Aiming at the problems of the detection speed and accuracy existing in the detection of surface defects of cold rolled sheet,this paper proposes a fast image screening and high-precision classification method,which is verified by experiments.At the same time,according to the method proposed in this paper,write a prototype system software for surface defect detection of cold rolled.A method for screening the surface of cold rolled sheet based on image gradient mean is proposed.Gaussian filtering is used to denoise the cold rolled sheet surface to reduce the effect of noise on image gradient calculation.According to the difference between the gradient image of the defect image and no defect image,the average value of the gradient threshold value of the image is counted through experiments,and selecting a reasonable threshold value.Through this threshold,a large number of defective images in a large number of pictures are screened,thereby reducing the post-processing data and improving the detection speed.A surface image classification method for cold rolled sheet based on convolutional neural network is proposed.Using the established steel surface image data set,the optimal optimization method and activation function are selected through experiments.By setting reasonable hyperparameters,selecting proper parameter initialization methods,and using reduced overfitting methods,a convolutional neural network model for image classification of cold rolled sheet is established.Using the test set to test the method of screening the surface of the cold rolled sheet based on image gradient mean,the experimental results show that this method can quickly and accurately screen the surface image of the cold rolled sheet.In addition,experiments show that the convolutional neural network established by the text can classify the surface image of the cold rolled sheet with high accuracy and solving the currently existing problems of detection speed and accuracy.Finally,writing prototype system software with MFC and implement the above screening and classification methods in this software.Through the implementation of software,and once again proved the effectiveness of the screening method and classification method in this paper.