Будь ласка, використовуйте цей ідентифікатор, щоб цитувати або посилатися на цей матеріал:
http://elartu.tntu.edu.ua/handle/lib/36932
| Назва: | Classification of rolled metal defects using residual neural networks |
| Автори: | Konovalenko, Ihor Maruschak, Pavlo Mosiy, Lyubomyr Duchon, Frantisek Kelemen, Michal |
| Приналежність: | Department of Industrial Automation, Ternopil National Ivan Puluj Technical University, Rus’ka str. 56, 46001 Ternopil, Ukraine Slovak University of Technology in Bratislava, Ilkovičova 3, SK-812 19, Bratislava; Slovak Republic Technical University of Kosice, Letna 9, 04200, Kosice, Slovak Republic |
| Бібліографічний опис: | Classification of rolled metal defects using residual neural networks / Ihor Konovalenko, Pavlo Maruschak, Lyubomyr Mosiy, Frantisek Duchon, Michal Kelemen // ICAAEIT 2021, 15-17 December 2021. — Tern. : TNTU, Zhytomyr «Publishing house „Book-Druk“» LLC, 2021. — P. 98–104. — (Electrical engineering and power electronics). |
| Bibliographic description: | Konovalenko I., Maruschak P., Mosiy L., Duchon F., Kelemen M. (2021) Classification of rolled metal defects using residual neural networks. ICAAEIT 2021 (Tern., 15-17 December 2021), pp. 98-104. |
| Є частиною видання: | Proceedings of the International Conference „Advanced applied energy and information technologies 2021”, 2021 |
| Дата публікації: | 15-гру-2021 |
| Дата внесення: | 28-гру-2021 |
| Видавництво: | TNTU, Zhytomyr «Publishing house „Book-Druk“» LLC |
| Місце видання, проведення: | Ternopil |
| Часове охоплення: | 15-17 December 2021 |
| Теми: | metallurgy steel sheet surface defects visual inspection technology classification neural network |
| Кількість сторінок: | 7 |
| Діапазон сторінок: | 98-104 |
| Початкова сторінка: | 98 |
| Кінцева сторінка: | 104 |
| Короткий огляд (реферат): | The authors investigated deep residual neural networks, which are used to detect and classify defects found on the rolled metal surface. Based on the neural network with ResNet152 architecture, a classifier for recognizing defects of three classes was built. The proposed technique allows recognizing and classifying surface damage with high accuracy in real-time based on its image. The average binary accuracy of the classification made based on the test data is 97.3%. Neuron activation fields were studied in the convolutional layers of the model. The results obtained show that areas, which correspond to those with damage in the image, are activated. False-positive and false-negative cases of classifier application are investigated. Errors were found to occur most frequently in ambiguous situations when surface artefacts of different types are similar. |
| URI (Уніфікований ідентифікатор ресурсу): | http://elartu.tntu.edu.ua/handle/lib/36932 |
| ISBN: | 978-617-8079-60-4 |
| Власник авторського права: | © Ternopil Ivan Puluj National Technical University, Ukraine, 2021 |
| URL-посилання пов’язаного матеріалу: | https://www.kaggle.com/c/severstal-steel-defect-detection |
| References: | 1.Chen, H., Hu, Q., Zhai, B. et al. (2020). A robust weakly supervised learning of deep Conv-Netsfor surface defect inspection. Neural Comput & Applic, 32, 11229–11244. doi:10.1007/s00521-020-04819-5 2.Dhua, S.K. (2019). Metallurgical analyses of surface defects in cold-rolled steel sheets. J Fail.Anal. and Preven, 19, 1023–1033. doi:10.1007/s11668-019-00690-2 3.Fang, X., Luo, Q., Zhou, B., Li, C., Tian, L. (2020). Research progress of automated visual surfacedefect detection for industrial metal planar materials. Sensors , 20, 5136, doi:10.3390/s20185136 4.François Chollet. (2017). Deep Learning with Python. Manning Publications. 5.GOST 21014-88. (1989). Rolled Products of Ferrous Metals. Surface Defects. Terms andDefinitions; Izd. Stand.: Moscow, USSR; p. 61. (In Russian) 6.He, K., Zhang, X., Ren, S., Sun, J. (2015). Deep Residual Learning for Image Recognition. arXiv,arXiv:1512.03385v1. 7.Kaggle Severstal: Steel Defect Detection. Can You Detect and Classify Defects in Steel? (2019).Kaggle. Retrieved from https://www.kaggle.com/c/severstal-steel-defect-detection. 8.Konovalenko, I., Maruschak, P., Brezinová, J., Viňáš, J., Brezina, J. (2020). Steel Surface DefectClassification Using Deep Residual Neural Network. Metals, 10, 846. 9.Kostenetskiy, P., Alkapov, R., Vetoshkin, N., Chulkevich, R., Napolskikh, I., Poponin, O. (2019).Real-time system for automatic cold strip surface defect detection. FME Trans. 47, 765–774. doi:10.5937/fmet1904765K. 10.Lee, S.Y., Tama, B.A., Moon, S.J., Lee, S. (2019). Steel Surface Defect Diagnostics Using DeepConvolutional Neural Network and Class Activation Map. Appl. Sci. 9, 5449, doi:10.3390/app9245449. 11.Lin, T.-Y., Goyal, P., Girshick, R., He, K., Dollár, P. (2017). Focal Loss for Dense ObjectDetection. arXiv, arXiv:1708.02002v2. 12.Luo, Q., He, Y. (2016). A cost-effective and automatic surface defect inspection system for hot-rolled flat steel, Robotics and Computer-Integrated Manufacturing, 38, 16-30. 13.Takahashi, R., Matsubara, T., Uehara, K. (2018). RICAP: Random Image Cropping and PatchingData Augmentation for Deep CNNs. Conference on Machine Learning, Proceedings of The 10th Asian Conference. 14.Tao, X., Zhang, D., Ma, W., Liu, X., Xu, D. (2018). Automatic metallic surface defect detectionand recognition with convolutional neural networks. Appl. Sci. 8, 1575, doi:10.3390/app8091575. 15.Zhou, S., Chen, Y., Zhang, D., Xie, J., Zhou, Y. (2017). Classification of surface defects on steelsheet using convolutional neural networks. Mater Technology, 51(1):123. |
| Тип вмісту: | Conference Abstract |
| Розташовується у зібраннях: | International conference „Advanced Applied Energy and Information Technologies 2021“, (ICAAEIT 2021) |
Файли цього матеріалу:
| Файл | Опис | Розмір | Формат | |
|---|---|---|---|---|
| ICAAEIT_2021_Konovalenko_I-Classification_of_rolled_98-104.pdf | 1,59 MB | Adobe PDF | Переглянути/відкрити | |
| ICAAEIT_2021_Konovalenko_I-Classification_of_rolled_98-104.djvu | 442,39 kB | DjVu | Переглянути/відкрити | |
| ICAAEIT_2021_Konovalenko_I-Classification_of_rolled_98-104__COVER.png | 447,65 kB | image/png | Переглянути/відкрити |
Усі матеріали в архіві електронних ресурсів захищені авторським правом, всі права збережені.