Research On Surface Defects Detection Classification And Image Generation Method Of The Magnetic Tile

In recent years,with the continuous development of the intelligent manufacturing industry,automated product defect detection technology based on artificial intelligence and machine vision has been widely used in various industries.Magnetic tile as the main component of permanent magnet motors,the quality of the magnetic tiles directly affects the service life and working performance of permanent magnet motors.At present,in most of the production process of the magnetic tiles,the quality detection of the magnetic tiles is still mainly based on manual visual detection.This detection method has the problems of low detection efficiency,inconsistent detection standards,and high labor costs.In addition,due to the many types of surface defects on magnetic tiles,the surface texture is complex and the contrast is low,it is difficult for traditional visual detection and image processing techniques to accurately detect and classify the surface defects of the magnetic tiles.Therefore,researching an automatic defect detection and classification technology suitable for the magnetic tiles is of great significance for the generation of the magnetic tiles.Machine vision and image processing are important research contents in the field of Artificial Intelligence.The features of image data are mainly extracted and learned through convolutional neural networks to obtain the information contained in the image and further process the image.At present,the application of machine vision and image processing to the research of the magnetic tile defect detection has achieved certain results.In this article,research is contrary to the deficiencies in current magnetic tile defect detection.The specific research content and innovations are as follows:（1）In view of the low accuracy of the magnetic tile defect detection,this article proposes a grid search and particle swarm optimization based local receptive field enabled extreme learning machine（GSPSO-ELM-LRF）magnetic tile defect detection and classification algorithm.Based on the traditional extreme learning machine based on local receptive field（ELM-LRF）algorithm,a grid search method is used to optimize the algorithm’s balance parameter C and the number of feature maps K to obtain the optimal parameter combination（C,K）.The particle swarm algorithm is used to optimize the initial weightA^init of the input to improve the classification performance of the algorithm.It can be seen from the experimental results that the proposed algorithm not only improves theaccuracy of the magnetic tile defect detection,but also realizes the classification of the magnetic tile images with defects after detection,which is beneficial to the improvement of the magnetic tile production processes.（2）Aiming at the problems of the magnetic tile defect image samples are small,and the number of different types of defect samples is uneven during the training of the magnetic tile defect detection classification model.This article proposes the gaussian mixture model for deep convolution generative adversarial networks（GMM-DCGANs）magnetic tile defect image generation network.Based on the traditional deep convolution generative adversarial networks（DCGANs）,the potential space of the input noise of the network is complicated into a Gaussian mixture model.Thus it can improves the learning ability of the image generation network for a limited number of inter-and intra-class diversity training samples.The experimental results show that the GMM-DCGANs network can generate the magnetic tile defect images with better quality and richer defect types,and the generated images meet the requirements of defect detection and classification.