Research On Surface Quality Detection Method Of Galvanized Steel Plate Based On Deep Learning

As China's national strength continues to increase,the overall level of manufacturing industry is also rapidly increasing.The automatic production rate of a series of products mainly represented by steel has already reached the world's leading level.But with the continuous enrichment of manufacturing technology and people's higher attention to the quality of the products,how to evaluate the surface quality of the product efficiently and accurately has become the bottleneck of the upgrading of the production level of the iron and steel industry.The traditional method of artificial detection is overly dependent on the subjective consciousness of the operator,and the environment of steel production is not suitable for people to work for long time.In addition,the steel production line is generally in a state of high speed,the ability of the human eyes to identify the feature will be weakened.Therefore,the method of artificial detection can not meet the needs of the development of steel manufacturing industry.With the development of computer technology,the surface detection technology based on machine vision has gradually become an alternative way of artificial detection.Until today,the machine vision technology has developed into various forms,but its core detection algorithm usually relies on machine learning with shallow models.The value of this kind model relies on the features selected by human,and when it face to the complex classification problems,the classification ability of the limited features is very difficult to meet the requirements.As a branch of machine learning,deep learning is characterized by the ability to enhance the feature extraction capability of the models through a deep network structure.When it face to the complex and diverse classification problems,the deep learning model has a greater advantage than the shallow learning model.In this thesis,after considering various factors,the convolution neural network model is used as the core model for improving the surface defect detection algorithm of galvanized steel sheet.The advantage of the convolution neural network is that it simulates the working patterns of the visual nervous system through the convolution layers and the pooling layers alternately connected,and reduces the complexity of the model through the non-fully connected structure.Therefore,the rich original image information can be efficiently used.The convolution neural networks have great advantages in the field of image recognition.Although the deep learning model can achieve more powerful feature extraction and classification ability than the traditional shallow learning model,there are still many obstacles in feasibility when it combined with practical applications.In this thesis,the deep learning model is used to replace the traditional shallow learning model to combine with machine vision systems.A deep learning algorithm model is proposed to detect and classify the surface quality defects of galvanized steel.According to the actual background,this paper optimizes many aspects such as the use of image data,the hardware and software design of model,and the selection of model parameters.It not only gives full use to the advantages of deep learning algorithm,but also ensures the high efficiency of the model.After the final test,the classification accuracy and classification speed of the model meet the real-time requirements of the defect detection system.