Defect Detection Of Steel Surface Using Deep Learning Based On Faster R-CNN

Due to increasing demand of better surface quality in steel manufacturing,defect detection has been become challenging subject in the field of computer vision research.Early and accurate detection of defect plays an important role in order to control and improve the quality of the final product.Compare with manual inspection method and machine learning method,the deep learning is more robust,which does not need human contact inspection and manually extracted features respectively.With the advent of deep learning,the machine learning systems are able to recognize and classify objects of interest in an image.They are capable of extracting feature detection process automatically from large amounts of tagged training data.The work special interest in applying bounding box regression around the defected area of steel surface and classify each class of those defect.In order to achieve better performance,image enhancement is done by adopting laplacian operator in preprocessing step before performing deep learning.Furthermore,center loss is added to the original loss function to improve the network's ability to distinguish different types of defects.In the proposed method,Faster R-CNN adopted AlexNet architecture as backbone for feature extractor.The detection and recognition results can be used to adjust the production line and improved surface quality in steel manufacturing.The experiment results show that the proposed method can be implemented to detect the defected area of the steel surface by making the prediction of the bounding box and classify the defected surface more accurately.In addition,the experiment results show that the proposed algorithm could precisely predict the bounding box on the defect surface such as scratches,patches and inclusion defects.