CNN Based Metal Surface Defect Detection Algorithm And Platform Development

As the basic material of industrial manufacturing,the quality of metal products will directly affect the service life and reliability of industrial products in rail transportation,aerospace,machinery manufacturing and many other fields.At present,the demand and quality requirements of metal surface materials in all walks of life are constantly improving.However,metal surface defects caused by process,environment,human error and other factors are often common,which seriously affect the performance of metal surface products.The existing manual visual inspection and related traditional methods cannot meet the requirements of mass production due to their limitations.With the development of artificial intelligence technology,methods based on machine vision,such as using Convolutional Neural Network(CNN)to classify and detect defects,are widely used.The topic of this project is the scientific research cooperation project of enterprises and institutions.Aiming at the problem of metal surface defect detection,this thesis carried out the research of defect detection algorithm based on CNN and designed and implemented the metal surface defect detection platform,providing a feasible scheme-for the typical application of artificial intelligence-based smart factory.The specific work content is as follows:This thesis describes the application scenarios of metal surface defect detection,analyzes the main challenges in the detection,and then summarizes the relevant research methods of metal surface defect detection.Finally,this thesis gives the related technology of the development of metal surface defect detection platform.In view of the problems such as large scale differences of metal surface defects,large differences between classes,scarce relevant data sets,and algorithm deployment on edge computing equipment with limited computing capacity,this paper proposes a metal surface defect detection algorithm based on CNN under the scenario of mass production of metal surface products.The proposed algorithm was lightweight based on ShuffleNetV2 network,and the 1×1 convolution kernel clipping method was used to reduce the computation of the detection algorithm.Then,a deformable convolution and attention module is introduced in ShuffleNetV2 network to ensure the accuracy of the detection algorithm.Furthermore,considering that the defect scale is variable in practical application,multi-scale convolution is introduced to further improve the accuracy of detection algorithm.Experimental results show that the proposed algorithm can effectively improve the detection accuracy while reducing the amount of computation.Aiming at the convenience and interaction of users in metal surface defect detection,a metal surface defect detection platform is designed and implemented in this thesis.Based on the functional requirements of the platform,six functional modules including data acquisition module,data storage module,algorithm analysis module,platform service module,platform management module and visualization module are designed and implemented.Finally,the function test and performance analysis of each module of the platform are carried out to ensure the stability and reliability of the platform.