Steel Plate Surface Defect Detection And Application Based On Distributed Deep Learnin

Steel plate is an important raw material for industrial production and plays an important role in military defense,aerospace,automobile manufacturing,and construction industries.In the production environment,due to the interference of external factors and the limitation of production technology,the produced steel plate will have surface defects such as cracks,plaques,punches,and indentations,which seriously affect the appearance and performance of the steel plate.In the past,the detection of steel plate surface defects was mainly manual,which was inefficient and time-consuming.At present,the traditional machine vision technology,which is widely used,also has the disadvantage of low reusability and is difficult to meet the different production needs of steel enterprises.How realize automatic detection of steel plate surface defects has become an urgent problem for iron and steel enterprises.With the arrival of the era of big data and the improvement of hardware level,the research of steel plate surface defect detection based on deep learning has made much progress,gradually surpassing the traditional steel plate surface defect detection methods in terms of accuracy and speed.Therefore,this article creatively proposes two different deep learning defect detection algorithms to achieve the automatic detection of steel plate surface defects.The main contributions are as follows.This thesis comprehensively introduces the related work of distributed depth learning and steel plate surface defect detection.A distributed deep learning network is designed to improve the training speed of the model.Two different depth learning algorithms are designed for the detection of steel plate surface defects,and a prototype system of steel plate surface defect detection based on depth learning is developed with the detection algorithm as the core.In terms of distributed deep learning,aiming at the problem of slow training of deep learning models,this thesis designs distributed deep learning networks on big data clusters,which can significantly improve the training speed of models and reduce the time cost.In terms of algorithm feature extraction optimization,this thesis designed a variety of attention modules to enhance feature information extraction,which significantly improved the performance of the steel plate surface defect detection model.In view of the shortcoming of the large number of parameters in the deep learning model,this thesis designed a lightweight structure to replace part of the convolution layer of the model,and used convolution kernel with different expansion rates to increase the receptive field to improve the model performance.In addition to the network model,a CDF-Loss function was designed and Mish activation function was used to assist model training.Transfer learning and data enhancement techniques were used to improve the final training effect of the model.The experimental results show that the network model designed in this thesis achieves a better balance between accuracy and speed,which is superior to the current mainstream detection network,and can meet the requirements of automatic detection of steel plate surface defects.The designed distributed deep learning is suitable for multi-node model training and can make full use of the computing and storage capabilities of the distributed cluster.Finally,this thesis develops a steel plate surface defect detection prototype system based on two steel plate defect detection algorithms,which can label,train and detect steel plate images directly in the production environment.