Research On Surface Defect Detection Algorithm Of Industrial Parts Based On Deep Learning

In the process of industrial production,due to the influence of raw materials,manufacturing technology,production process and other factors,product quality defects may be caused,of which surface defects are the most common and intuitive form.In order to ensure the qualified rate and reliable quality of products,it is necessary to carry out surface defect testing to avoid unqualified products entering the market.The traditional manual visual detection method is inefficient,time-consuming and laborious,and the detection results are easily influenced by human subjective factors.It cannot satisfy the real-time detection requirements of industrial production,and cannot adapt to the development trend of industrial information.It has been gradually replaced by other more advanced methods.This paper firstly introduces the research status of surface defect detection methods at home and abroad,and then the definition of defect detection and the key technology of using deep learning to complete the defect detection task are introduced.(1)In view of the problems existing in the current deep learning-based defect detection methods,such as excessively complex network model and poor anti-interference ability,this paper proposes a surface defect detection model based on Convolutional Block Attention Module(CBAM)and atrous convolution.In this model,the surface defect segmentation and classification tasks of products are combined,and the context information of multi-scale images is obtained by Atrous Spatial Pyramid Pooling(ASPP).Then,the CBAM is used to redistribute the weight of the network to enhance the attention to the defect area,to improve the discrimination of extracted features.In addition,atrous convolution is introduced into the segmentation network,which simplifies the complexity of the defect segmentation network and improves the real-time performance of the model defect detection.The experimental results show that the model is better than the current mainstream defect detection methods for Kolektor SDD and Magnetic Tile datasets,and has a wide applicability in the surface defect detection of industrial products.(2)To solve the problem of insufficient defect data and annotation data of industrial parts,this paper proposes an unsupervised defect detection method based on generative adversarial network and memory module.This method uses the improved U-Net as the backbone network of the generator,and a memory module is embedded to record the data distribution characteristics of normal samples,and to limit the encoded latent feature vectors,thereby amplifying the reconstruction error of defective samples.The generative adversarial network is used to train the model and supplemented by the discriminator to guide the optimization of the generator,to enhance the generation ability of the generator.Experimentally evaluated on the Severstal steel defect dataset,the results show that the proposed unsupervised defect detection model achieves an average precision of 92.56%,which has a good reference for the research and application of unsupervised detection of surface defects in industrial parts.Finally,the surface defect detection method proposed in this paper is summarized,and the future research work is prospected.