本研究針對工業產品識別的需求，提出了一套方法，包含以下三個部分：(1)新型多視角卷積神經網路模型，(2)三維物體的遷移學習，(3)增量學習的演算法。由於工廠環境下取得的資料容易有翻轉、背景雜亂等問題，傳統的二維影像辨識方法效果不佳，本研究提出了一種新的網路架構，稱為自注意力殘差網路，具有優秀的背景分割能力和全局特徵理解，可幫助多視角的二維圖像進行三維物體識別。另一個問題是在現實中取得足以訓練好模型的資料量需要極高的成本，需要對多視角卷積神經網路進行遷移學習以提升辨識效果。此外，本研究針對無法儲存過去資料的情境之類別增量學習，提出局部深度模型融合。實驗表明，使用自注意力殘差卷積神經網路作為特徵擷取層，性能大幅領先目前最先進的卷積神經網路，並透過多視角卷積神經網路之遷移學習，在缺乏訓練資料的情形下進一步提高準確率。本研究提出的局部深度模型融合，對比其他增量學習演算法同樣取得較佳的效果。最後我們使用真實的工業產品拍攝，模擬在實際辨識情況會遇到的困境，並演示模型之效果。

This study addresses the need for industrial product identification by proposing a method that includes the following three components: (1) a novel multi-view convolutional neural network model, (2) transfer learning for three-dimensional objects, and (3) an incremental learning algorithm. Traditional 2D image recognition methods perform poorly due to issues like flipping and cluttered backgrounds commonly found in factory environments. This study introduces a new network architecture, called the “SARNet”, which excels in background segmentation and global feature comprehension, aiding in the recognition of 3D objects from multi-view 2D images. Another challenge is the high cost of obtaining sufficient data to train a model effectively in real-world scenarios, which necessitates the use of transfer learning for the multi-view convolutional neural network to enhance recognition performance. Additionally, this study proposes “Partial Deep Model Consolidation” for class incremental learning scenarios where storing past data is not feasible. Experiments demonstrate that using a “SARNet” as the feature extraction layer significantly outperforms the current state-of-the-art convolutional neural networks. Transfer learning with the multi-view convolutional neural network further improves accuracy in situations with limited training data. The proposed “Partial Deep Model Consolidation” also achieves better results compared to other incremental learning algorithms. Finally, we use real industrial product photographs to simulate the challenges encountered in actual recognition scenarios and demonstrate the effectiveness of the model.

[1] Bengio, Y. "Deep learning of representations for unsupervised and transfer learning." Proceedings of ICML Workshop on Unsupervised and Transfer Learning. JMLR Workshop and Conference Proceedings, 2012, 17-36.
[2] Bengio, Y., F. Bastien, A. Bergeron, N. Boulanger–Lewandowski, T. Breuel, Y. Chherawala, M. Cisse, M. Côté, D. Erhan, J. Eustache, X. Glorot, X. Muller, S. P. Lebeuf, R. Pascanu, S. Rifai, F. Savard, G. Sicard "Deep learners benefit more from out-of-distribution examples." Proceedings of the Fourteenth International Conference on Artificial Intelligence and Statistics. JMLR Workshop and Conference Proceedings, 2011, 164-172.
[3] De Lange, M., R. Aljundi, M. Masana, S. Parisot, X. Jia, A. Leonardis, G. Slabaugh, T. Tuytelaars "A continual learning survey: Defying forgetting in classification tasks." IEEE Transactions on Pattern Analysis and Machine Intelligence 44.7, 2021, 3366-3385.
[4] Deng, J., W. Dong, R. Socher, L. J. Li, K. Li, & L. Fei-Fei "Imagenet: A large-scale hierarchical image database." 2009 IEEE Conference on Computer Vision and Pattern Recognition. IEEE, 2009, 248-255.
[5] Goodfellow, I. J., M. Mirza, D. Xiao, A. Courville, Y. Bengio "An empirical investigation of catastrophic forgetting in gradient-based neural networks." arXiv preprint arXiv:1312.6211, 2013.
[6] He, K., X. Zhang, S. Ren, J. Sun "Deep residual learning for image recognition." Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2016, 770-778.
[7] Hinton, G., O. Vinyals, & J. Dean. "Distilling the knowledge in a neural network." arXiv preprint arXiv:1503.02531, 2015.
[8] Kanezaki, A., Y. Matsushita, & Y. Nishida. "Rotationnet for joint object categorization and unsupervised pose estimation from multi-view images." IEEE Transactions on Pattern Analysis and Machine Intelligence 43.1, 2019, 269-283.
[9] Kim, T., J. Oh, N.Y. Kim, S. Cho, S.Y. Yun "Comparing kullback-leibler divergence and mean squared error loss in knowledge distillation." arXiv preprint arXiv:2105.08919, 2021.
[10] Krizhevsky, A., I. Sutskever, & G. E. Hinton. "Imagenet classification with deep convolutional neural networks." Advances in Neural Information Processing Systems, 2012, 25.
[11] Li, Z., & D. Hoiem. "Learning without forgetting." IEEE Transactions on Pattern Analysis and Machine Intelligence 40.12, 2017, 2935-2947.
[12] Obst, P., W. Nasser, S. Rink, G. Kleinpeter, B. Szost, D. Rietzel, G. Witt "Complexity and economical value of Artificial Intelligence for automated and industrialized recognition of additive manufactured components." Proc. 17th Rapid. Tech 3D Conf, 2021, 141-152.
[13] Phong, B. T. "Illumination for computer generated pictures." Seminal Graphics: Pioneering Efforts that Shaped the Field, 1998, 95-101.
[14] Qi, C. R., H. Su, K. Mo, L. J. Guibas "Pointnet: Deep learning on point sets for 3d classification and segmentation." Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017, 652-660.
[15] Qi, S., X. Ning, G. Yang, L. Zhang, P. Long, W. Cai, W. Li "Review of multi-view 3D object recognition methods based on deep learning." Displays 69, 2021, 102053.
[16] Ramasesh, V. V., E. Dyer, & M. Raghu. "Anatomy of catastrophic forgetting: Hidden representations and task semantics." arXiv preprint arXiv:2007.07400, 2020.
[17] Schuh, G., G. Lukas, S. Hohenstein, J. M. Schäfer, J. L. Drescher "Part Recognition in Additive Production Systems using a Computer-vision Approach." 2022 IEEE International Conference on Industrial Engineering and Engineering Management (IEEM). IEEE, 2022, 96-101.
[18] Simonyan, K., & A. Zisserman. "Very deep convolutional networks for large-scale image recognition." arXiv preprint arXiv:1409.1556, 2014.
[19] Su, H., S. Maji, E. Kalogerakis, E. Learned-Miller "Multi-view convolutional neural networks for 3d shape recognition." Proceedings of the IEEE International Conference on Computer Vision, 2015, 945-953.
[20] Wu, Z., S. Song, A. Khosla, F. Yu, L. Zhang, X. Tang, J. Xiao "3d shapenets: A deep representation for volumetric shapes." Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2015, 1912-1920.
[21] Yosinski, J., J. Clune, Y. Bengio, H. Lipson "How transferable are features in deep neural networks?." Advances in Neural Information Processing Systems, 2014, 27.
[22] Zhang, H., I. Goodfellow, D. Metaxas, A. Odena "Self-attention generative adversarial networks." International Conference on Machine Learning. PMLR, 2019, 7354-7363.
[23] Zhang, J., J. Zhang, S. Ghosh, D. Li, S. Tasci, L. Heck, H. Zhang, C. C. J. Kuo "Class-incremental learning via deep model consolidation." Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, 2020, 1131-1140.