Surface Defect Detection Of Strip Steel Based On Deep Learning

With the proposal of "making 2025 in China",The intelligentization and automation of manufacturing and testing have become the inevitable trend of the development of metallurgical manufacturing enterprises And strip steel is one of the main products of the traditional steel manufacturing enterprises,the detection of its surface defects is an important part of quality monitoring in the process of strip steel production However,it is often influenced by the factors such as the environment and the variety of the strip defect.The traditional detection algorithm is difficult to extract the essential feature of the strip defect image,which leads to the defect detection effect can not reach the ideal requirement.With the development of computer technology,internet technology and artificial intelligence,deep learning based on feature extraction of large data has become a key technology to solve the detection and recognition of strip defects.Deep learning is developed from traditional machine learning,absorb nutrients from large data,learning the intrinsic characteristics of data can be widely used and high accuracy.In this paper,the related theories of deep learning convolution neural network for image processing are studied in depth.On this basis,the target detection network for deep learning is analyzed.A surface defect detection network for strip steel based on Faster R-CNN was proposed.First,the training data set is extended and the data set is analyzed.Using Resnet101 as the feature extraction network for images.Then use the transfer learning method to train the network and optimize the model parameters.Avoid the local minimum of the optimized parameters in the process of avoiding network training.The stability and convergence of the network training process are guaranteed.The validity of the detection algorithm is verified by the test platform.Finally,by quantitative analysis of the test results,in this experiment the strip surface defect detection based on deep learning of the accuracy and recall rate,respectively 92.2%,98.8% and 100%.And the detection speed is very fast,it fully meets the requirements of the automatic testing of the strip in the industrial production.At the end of the article,the main work,innovation and shortcomings of the subject are summarized.The future development direction of automatic testing technology for strip defects is prospected.