Steel Surface Defect Detection System Based On Machine Vision

China has always been the world's largest steel producer.In 2019,China's crude steel output reached 955 million tons,and steel output reached 1205 million tons.In2020,China's crude steel output was 1.053 billion tons,a year-on-year increase of52%.Steel output was 1,324,an increase of 7.7%.The steel industry,as a very important basic industry in China,plays an extremely important role in economic construction and social development.However,the imperfect production process and the backward production equipment may cause defects in the production process of steel surface.Surface defects affect the appearance of the product,but also reduce the corrosion resistance and fatigue strength of steel products to a certain extent.Therefore,detecting the surface defects of steel is of great significance to the steel production enterprises and even the whole society.With the improvement and development of image processing,machine vision and neural network theory,steel surface defect detection technology based on machine vision has become the focus of research.In this topic,the image processing algorithm and deep learning algorithm of steel surface defects are studied in depth.The main research contents are as follows:(1)The image processing algorithm and deep learning algorithm are studied,the existing results of image processing algorithm and deep learning algorithm are summarized,and the limitations of current research are analyzed.(2)To solve the problem of uneven illumination in the steel surface defect image,a multi-layer processing strategy for image processing with uneven illumination is presented.The algorithm first corrects the image gray level by morphological processing and target background difference.Then the Retinex algorithm is used to process the image.Finally,an improved adaptive threshold segmentation algorithm is used to generate a binary image.The experimental results show that the algorithm has a good performance on images with uneven illumination,can effectively eliminate the influence of uneven illumination,and can accurately segment the defect target and background.(3)A Faster R-CNN algorithm for feature fusion and cascade detection network is presented to solve the problem of low detection accuracy caused by reduced structure information when target detection is performed by deep learning algorithms.Based on the traditional Faster R-CNN algorithm,the algorithm improves the backbone network and detection network.In the main network part,VGG-16 is used as feature extraction network,and VGG-16 is feature fused.This method reduces the reduction of image structure information during feature extraction in the backbone network.In the detection network section,the detection accuracy of the algorithm is improved by cascading two detection networks with different Io U thresholds.The algorithm is trained on the NEU-DET dataset to get different detection models.The test results show that the best detection model can effectively improve the detection accuracy of steel surface defects without affecting the detection speed.(4)Set up a steel surface defect detection system,analyze the functional requirements of the software,and complete the development of the system software.Through the combination of hardware and software,the detection of steel surface defects is achieved.