在先進的產業中，建立一套CAD模型資料庫並藉由CAD模型搜索，減少產品之開發時間之應用已相當普遍。然而，隨著CAD模型機構設計的複雜化，與CAD模型種類與數量的增加，藉由人工方式進行CAD模型搜索變得越來越困難。本研究的主要目的是開發應用於CAD模型資料庫上的計算CAD模型多視角影像演算法與深度學習影像辨識演算法，將CAD模型搜索問題定義為影像辨識問題，訓練一個自動化分類CAD模型的深度學習神經網路。在本研究中，建立了一個總數量為360個，90分類的CAD模型資料庫，其中訓練用CAD模型有90個，測試用CAD模型有270個，並利用計算CAD模型多視角影像演算法，建立一套完整的深度學習數據集。深度學習方面則是利用VGG16神經網路進行遷移學習、超參數調優與合併不同視角的深度學習模型等方式，最佳化深度學習模型。本研究最主要的貢獻為利用深度學習演算法，成功的分類CAD模型，並且藉由組合不同視角的深度學習模型，提高辨識的正確率。

Industries nowadays, it becomes mandatory for establishing CAD model database for the purpose of reducing development time of the product design. The database will ease the user to search the similarity between the developing and database models. The time needed for searching the similarity is corresponding to the complexity and various types of CAD models. Therefore, the purpose of this study is to provide a method and procedures to reduce the time needed for CAD model searching. This task can be achieved by combining a multi-view imaging technique for CAD models and a deep learning image classification algorithm. The first technique is for generating an image to represent the distribution of the depth in a view for a CAD model, and then the second technique is employed to apply multi-view image data for the training of the AI model. These two techniques can be combined to classify CAD models automatically based on its similarity. In this study, we provide 360 CAD-model databases and 90 classes generated by using multi-view images and deep learning classification algorithm. The 360 CAD database consists of 90 CAD models for training the AI models and 270 CAD models for testing the AI models. Eventually, using the proposed multi-view imaging algorithm to build a complete deep learning database. For deep learning, it employs VGG16 transfer learning, hyperparameter tuning, and assemble deep learning models from different views to optimize the deep learning models. The main contribution of this study is that we develop a multi-view imaging algorithm to convert a CAD model into images, and successfully classify CAD models and improve the accuracy of the learning by assembling the deep learning models from different views.

參考文獻
[1] J. Y. Lai, S. H. Shu and Y. C. Huang, “A cell subdivision strategy for r-nearest neighbors computation”, J. Chin. Inst. Eng, Vol. 29, pp. 953–965, 2006.
[2] A. Krizhevsky, I. Sutskever and G. E. Hinton, “Imagenet classification with deep convolutional neural networks”, Advances in Neural Information Processing Systems, Vol. 25, pp. 1097-1105, 2012.
[3] K. Simonyan and A. Zisserman, “Very deep convolutional networks for large-scale image recognition.”, International Conference on Learning Representations , 2015.
[4] C. Szegedy, W. Liu, Y. Jia, P. Sermanet, S. Reed, D. Anguelov, ... and A. Rabinovich, “Going deeper with convolutions”, IEEE Conference On Computer Vision and Pattern Recognition, pp. 1-9, 2015.
[5] K. He, X. Zhang,S. Ren and J. Sun, “Deep residual learning for image recognition”, IEEE Conference on Computer Vision and Pattern Recognition,
pp. 770-778, 2016.
[6] H. Su, S. Maji, E. Kalogerakis, and E. Learned-Miller, “Multi-view convolutional neural networks for 3d shape recognition”, IEEE International Conference on Computer Vision, pp.945-953, 2015.
[7] B. T. Phong, “ Illumination for computer generated pictures”, Commun. ACM, pp. 6, 1975.
[8] Q. Yu, C. Yang, H. Fan and H. Wei, “Latent-MVCNN: 3D shape recognition using multiple views from pre-defined or random viewpoints ”, Neural Processing Letters, Vol. 52, pp. 581-602, 2020.
[9] Y. Feng, H. You, X. Zhao and Y. Gao, “Meshnet: mesh neural network for 3D shape representation”, AAAI Conference on Artificial Intelligence, Vol. 33, No. 01, pp. 8279-8286, 2019.
[10] C. R. Qi, H. Su, K. Mo and L. J. Guibas, “Pointnet: Deep learning on point sets for 3d classification and segmentation”, IEEE Conference On Computer Vision and Pattern Recognition, pp. 652-660, 2017.
[11] C. Zhang and G. Zhou, “A view-based 3D CAD model reuse framework enabling product lifecycle reuse”, Advances in Engineering Software, Vol. 127, pp. 82-89, 2019.
[12] S. Dong, K. Zou and W. Li, “Fine-grained recognition of 3D shapes based on multi-view recurrent neural network”, International Conference on Machine Learning and Computing, pp. 152-156, 2020.
[13] Y. Feng, Z. Zhang, X. Zhao, R. Ji and Y. Gao, “Gvcnn: group-view convolutional neural networks for 3D shape recognition”, IEEE Conference on Computer Vision and Pattern Recognition, pp 264-272, 2018.
[14] T. Yu, J. Meng and J. Yuan, “Multi-view harmonized bilinear network for 3D object recognition”, IEEE Conference on Computer Vision and Pattern Recognition, pp. 186-194, 2018
[15] Z. Yang and L. Wang, “Learning relationships for multi-view 3D object recognition”, IEEE/CVF International Conference on Computer Vision , pp. 7505-7514, 2019.
[16] J. Jiang, D. Bao, Z.Chen, X. Zhao and Y. Gao, “ MLVCNN: Multi-loop-view convolutional neural network for 3D shape retrieval”, AAAI Conference on Artificial Intelligence, Vol. 33, pp. 8513-8520, 2019.
[17] Rhinoceros , Website:http://www.rhino3d.com, Accessed 18 June 2021.
[18] openNURBS, Website: http://www.rhino3d.com/tw/opennurbs, Accessed 18 June 2021.
[19] PyTorch, Website:https://pytorch.org/, Accessed 18 June 2021.
[20] Numpy, Website:https: https://numpy.org/, Accessed 18 June 2021.
[21] Matplotlib,Website:https://matplotlib.org/, Accessed 18 June 2021.
[22] 曾定璋,深度學習電腦視覺,Website:http://ip.csie.ncu.edu.tw/i09.htm, Accessed 18 June 2021.
[23] 卷積神經網路,Website:https://chtseng.wordpress.com/2017/09/12/%E5%88%9D%E6%8E%A2%E5%8D%B7%E7%A9%8D%E7%A5%9E%E7%B6%93%E7%B6%B2%E8%B7%AF/,Accessed 18 June 2021.
[24] 動手學深度學習,Website:https://zh.d2l.ai/chapter_introduction/deep-learning-intro.html, Accessed 18 June 2021.