在鋼鐵業中，鋼鐵表面瑕疵辨識是實現產品品質檢驗的重要技術，透過檢測來分析造成瑕疵的原因，藉此減少產品製程中的瑕疵，以確實降低瑕疵率。然而，隨著深度學習的崛起，大多的研究皆使用相關技術來提高準確率與模型的性能，但在模型訓練下，至今還存在著需要大規模精準標記的數據集和大量的運算資源之問題。因此本研究希望解決上述之問題，並同時改進瑕疵檢測的效能，從而提高生產效率和維持產品品質。
本研究提出一種運用類神經網路，並結合注意力機制，來探討不同尺度特徵之間所隱藏的訊息與關係，藉此有效地辨識鋼鐵表面上多種類型的瑕疵，以提升模型預測的準確度。本研究主要分為兩個部分：第一部分將會使用SqueezeNet 來提取不同尺度的特徵圖；第二部分則透過多尺度注意力機制來處理不同尺度的特徵圖，以強化影像中特定區域或特徵，從而準確地分類多種瑕疵。
根據實驗結果顯示，在 NEU Surface Defect Dataset 中，本研究提出之模型由於考慮到多種不同尺度特徵圖的重要特徵，在識別鋼鐵表面瑕疵方面表現優異，相較於過去的研究，模型不僅能夠達到 99.72%的分類準確率，同時，減少模型訓練過程中計算資源的消耗。在少量訓練資料的情況下，即在 NEU 訓練資料集中，每個類別隨機挑選出 100 張、50 張、10 張影像，本研究所提出之方法在分類多種類型瑕疵時的準確率也能保持在 92%以上。除此之外，在不同領域的應用中，本研究之模型也展現出良好的泛化能力。

In the steel industry, identifying surface defects on steel is a crucial aspect of quality inspection. Despite advancements in deep learning, which have led to improved accuracy and model performance in related research, challenges such as the need for precisely labeled large-scale datasets and significant computational resources for model training persist. To address these challenges, we propose a multi-scale attention network (MSA) to explore hidden information and relationships between features of varying scales. In this method, SqueezeNet is used to extract featuresof different scales, and a multi-scale attention mechanism processes these features, enhancing specific regions or features in the image for accurate classification of multiple defects. Experimental results show that our MSA achieves a classification accuracy of 99.72% while reducing computational resource consumption during training on the NEU surface defect dataset, maintaining an accuracy of over 92% when classifying multiple types of defects with less data. Additionally, the model demonstrates good generalization ability across various application domain.

Aghdam, S. R., Amid, E., and Imani, M. F. (2012). A fast method of steel surface defect detection using decision trees applied to lbp based features. In 2012 7th IEEE Conference on Industrial Electronics and Applications (ICIEA), pages 1447–1452 IEEE.
Bissi, L., Baruffa, G., Placidi, P., Ricci, E., Scorzoni, A., and Valigi, P. (2013). Automated defect detection in uniform and structured fabrics using gabor filters and pca. Journal of Visual Communication and Image Representation, 24(7):838–845.
Bodnarova, A., Bennamoun, M., and Latham, S. (2002). Optimal gabor filters for textile flaw detection. Pattern recognition, 35(12):2973–2991.
Cao, Y., Xu, J., Lin, S., Wei, F., and Hu, H. (2020). Global context networks. IEEE Transactions on Pattern Analysis and Machine Intelligence.
Deng, J., Dong, W., Socher, R., Li, L.-J., Li, K., and Fei-Fei, L. (2009). Imagenet: A large-scale hierarchical image database. In 2009 IEEE conference on computer vision and pattern recognition, pages 248–255. Ieee.
Deshpande, A. M., Minai, A. A., and Kumar, M. (2020). One-shot recognition of manufacturing defects in steel surfaces. Procedia Manufacturing, 48:1064–1071.
Di, H., Ke, X., Peng, Z., and Dongdong, Z. (2019). Surface defect classification of steels with a new semi-supervised learning method. Optics and Lasers in Engineering, 117:40–48.
Dogandžić, A., Eua-anant, N., and Zhang, B. (2005). Defect detection using hidden markov random fields. In AIP conference proceedings, volume 760, pages 704–711. American Institute of Physics.
Dong, H., Song, K., He, Y., Xu, J., Yan, Y., and Meng, Q. (2019). Pga-net: Pyramid feature fusion and global context attention network for automated surface defect detection. IEEE Transactions on Industrial Informatics, 16(12):7448–7458.
Fang, F., Xu, Q., and Lim, J.-H. (2022). Hierarchical defect detection based on reinforcement learning. In 2022 IEEE International Conference on Image Processing (ICIP), pages 791–795. IEEE.
Fu, G., Zhang, Z., Le, W., Li, J., Zhu, Q., Niu, F., Chen, H., Sun, F., and Shen, Y. (2023). A multi-scale pooling convolutional neural network for accurate steel surface defects classification. Frontiers in Neurorobotics, 17:1096083.
He, K., Zhang, X., Ren, S., and Sun, J. (2016). Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 770–778.
He, Y., Song, K., Dong, H., and Yan, Y. (2019). Semi-supervised defect classification of steel surface based on multi-training and generative adversarial network. Optics and Lasers in Engineering, 122:294–302.
Hu, J., Shen, L., and Sun, G. (2018). Squeeze-and-excitation networks. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 7132–7141.
Iandola, F. N., Han, S., Moskewicz, M. W., Ashraf, K., Dally, W. J., and Keutzer, K. (2016). Squeezenet: Alexnet-level accuracy with 50x fewer parameters and < 0.5mb model size. arXiv preprint arXiv:1602.07360.
Kang, X. and Zhang, E. (2020). A universal and adaptive fabric defect detection algorithm based on sparse dictionary learning. IEEE Access, 8:221808–221830. Kingma, D. P. and Ba, J. (2014). Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980.
Koch, G., Zemel, R., Salakhutdinov, R., et al. (2015). Siamese neural networks for one-shot image recognition. In ICML deep learning workshop, volume 2. Lille.
Krizhevsky, A., Sutskever, I., and Hinton, G. E. (2012). Imagenet classification with deep convolutional neural networks. Advances in neural information processing systems, 25.
Li, J., Su, Z., Geng, J., and Yin, Y. (2018). Real-time detection of steel strip surface defects based on improved yolo detection network. IFAC-PapersOnLine, 51(21):76–81.
Li, P., Zhang, H., Jing, J., Li, R., and Zhao, J. (2015). Fabric defect detection based on multi-scale wavelet transform and gaussian mixture model method. The Journal of The Textile Institute, 106(6):587–592.
Li, W.-C. and Tsai, D.-M. (2012). Wavelet-based defect detection in solar wafer images with inhomogeneous texture. Pattern Recognition, 45(2):742–756.
Liu, Y., Xu, K., and Xu, J. (2019). An improved mb-lbp defect recognition approach for the surface of steel plates. Applied Sciences, 9(20):4222.
Liu, Y., Yuan, Y., Balta, C., and Liu, J. (2020). A light-weight deep-learning model with multi-scale features for steel surface defect classification. Materials, 13(20):4629.
Luo, Q., Fang, X., Liu, L., Yang, C., and Sun, Y. (2020). Automated visual defect detection for flat steel surface: A survey. IEEE Transactions on Instrumentation and Measurement, 69(3):626–644.
Luong, M.-T., Pham, H., and Manning, C. D. (2015). Effective approaches to attentionbased neural machine translation. arXiv preprint arXiv:1508.04025.
Martins, L. A., Pádua, F. L., and Almeida, P. E. (2010). Automatic detection of surface defects on rolled steel using computer vision and artificial neural networks. In IECON 2010-36th Annual Conference on IEEE Industrial Electronics Society, pages 1081–1086. IEEE.
Neogi, N., Mohanta, D. K., and Dutta, P. K. (2014). Review of vision-based steel surface inspection systems. EURASIP Journal on Image and Video Processing, 2014(1):1–19.
Parikh, A. P., Tackstr ¨ om, O., Das, D., and Uszkoreit, J. (2016). A decomposable attention model for natural language inference. arXiv preprint arXiv:1606.01933.
Pernkopf, F. (2004). Detection of surface defects on raw steel blocks using bayesian network classifiers. Pattern Analysis and Applications, 7:333–342.
Shi, T., Kong, J.-y., Wang, X.-d., Liu, Z., and Zheng, G. (2016). Improved sobel algorithm for defect detection of rail surfaces with enhanced efficiency and accuracy. Journal of Central South University, 23(11):2867–2875.
Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, Ł., and Polosukhin, I. (2017). Attention is all you need. Advances in neural information processing systems, 30.
Wang, Z. and Zhu, D. (2019). An accurate detection method for surface defects of complex components based on support vector machine and spreading algorithm. Measurement, 147:106886.
Woo, S., Park, J., Lee, J.-Y., and Kweon, I. S. (2018). Cbam: Convolutional block attention module. In Proceedings of the European conference on computer vision (ECCV), pages 3–19.
Wu, X., Gao, S., Sun, J., Zhang, Y., and Wang, S. (2022). Classification of alzheimer’s disease based on weakly supervised learning and attention mechanism. Brain Sciences, 12(12):1601.
Yan, Y., Kawahara, J., and Hamarneh, G. (2019). Melanoma recognition via visual attention. In Information Processing in Medical Imaging: 26th International Conference, IPMI 2019, Hong Kong, China, June 2–7, 2019, Proceedings 26, pages 793–804. Springer.
Yao, H., Zhang, X., Zhou, X., and Liu, S. (2019). Parallel structure deep neural network using cnn and rnn with an attention mechanism for breast cancer histology image classification. Cancers, 11(12):1901.
Yeung, C.-C. and Lam, K.-M. (2022). Efficient fused-attention model for steel surface defect detection. IEEE Transactions on Instrumentation and Measurement, 71:1–11.
Yi, L., Li, G., and Jiang, M. (2017). An end-to-end steel strip surface defects recognition system based on convolutional neural networks. steel research international, 88(2):1600068.
Yin, M., Yao, Z., Cao, Y., Li, X., Zhang, Z., Lin, S., and Hu, H. (2020). Disentangled non-local neural networks. In Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part XV 16, pages 191–207. Springer.
Zhang, H., Goodfellow, I., Metaxas, D., and Odena, A. (2019). Self-attention generative adversarial networks. In International conference on machine learning, pages 7354–7363. PMLR.
Zhang, Z., Wan, X., Li, L., and Wang, J. (2021). An improved dcgan for fabric defect detection. In 2021 IEEE 4th International Conference on Electronics and Communication Engineering (ICECE), pages 72–76. IEEE.
Zhao, C., Shu, X., Yan, X., Zuo, X., and Zhu, F. (2023). Rdd-yolo: A modified yolo for detection of steel surface defects. Measurement, 214:112776.
Zheng, Y. and Cui, L. (2023). Defect detection on new samples with siamese defectaware attention network. Applied Intelligence, 53(4):4563–4578.
Zhou, J., Semenovich, D., Sowmya, A., and Wang, J. (2014). Dictionary learning framework for fabric defect detection. The Journal of The Textile Institute, 105(3):223–234.
Zorić, B., Matić, T., & Hocenski, Ž. (2022). Classification of biscuit tiles for defect detection using fourier transform features. ISA transactions, 125:400–414.