Cascading Convolutional Neural Network for Steel Surface Defect Detection
Steel is the most important material in the world of engineering and construction. Modern steelmaking relies on computer vision technologies, like optical cameras to monitor the production and manufacturing processes, which helps companies improve product quality. In this paper, we propose a deep learning method to automatically detect defects on the steel surface. The architecture of our proposed system is separated into two parts. The first part uses a revised version of single shot multibox detector (SSD) model to learn possible defects. Then, deep residual network (ResNet) is used to classify three types of defects: Rust, Scar, and Sponge. The combination of these two models is investigated and discussed thoroughly in this paper. This work additionally employs a real industry dataset to confirm the feasibility of the proposed method and make sure it is applicable to real-world scenarios. The experimental results show that the proposed method can achieve higher precision and recall scores in steel surface defect detection.
KeywordsFully convolutional networks Defect detection SSD ResNet
This work was supported by the Ministry of Science and Technology, Taiwan, under Grants MOST 106-2218-E-468-001, MOST 107-2221-E-155-048-MY3, and MOST 108-2634-F-008-001, and under Grants from China Steel Corporation RE106728 and RE107705.
- 1.Peacock, G.R.: Review of noncontact process temperature measurements in steel manufacturing. In: SPIE Conference on Thermosense XXI, pp. 171–189. SPIE, Florida (1999)Google Scholar
- 3.Duan, X., Duan, F., Han, F.: Study on surface defect vision detection system for steel plate based on virtual instrument technology. In: 2011 International Conference on Control, Automation and Systems Engineering (CASE), pp. 1–4. IEEE, Singapore (2011)Google Scholar
- 4.Balchin, N.C., Blunt, J.: Health and Safety in Welding and Allied Processes, 5th edn. Woodhead Publishing, Cambridge (2002)Google Scholar
- 5.Simonyan, K., Zisserman, A.: Very Deep Convolutional Networks for Large-Scale Image Recognition. CoRR abs/1409.1556 (2014)Google Scholar
- 6.Krizhevsky, A., Sutskever, I., Hinton, G.E.: ImageNet classification with deep convolutional neural networks. In: Advances in Neural Information Processing Systems 25 (NIPS 2012), pp. 1097–1105. Curran Associates, Inc., Nevada (2012)Google Scholar
- 7.Ren, S., He, K., Girshick, R., Sun, J.: Faster R-CNN: towards real-time object detection with region proposal networks. In: Advances in Neural Information Processing Systems 28 (NIPS 2015), pp. 91–99. Curran Associates, Inc., Canada (2015)Google Scholar
- 8.He, K., Gkioxari, G., Dollár, P., Girshick, R.B.: Mask R-CNN. In: 2017 IEEE International Conference on Computer Vision (ICCV), pp. 2980–2988. IEEE, Venice (2017)Google Scholar
- 10.Redmon, J., Kumar Divvala, B., Girshick, R., Farhadi, A.: You only look once: unified, real-time object detection. pp. 779–788. IEEE (2016)Google Scholar
- 11.Hu, J., Shen, L., Sun, G.: Squeeze-and-excitation networks. In: The IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 7132–7141. IEEE, Salt Lake City (2018)Google Scholar
- 12.He, K., Zhang, X., Ren, S., Sun, J.: Deep residual learning for image recognition. In: The IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 770–778. IEEE (2016)Google Scholar
- 13.Powers, D.M.W.: Evaluation: from Precision, Recall and F-measure to ROC, Informedness, Markedness and Correlation. 2017 (2007)Google Scholar