臺灣博碩士論文加值系統

在木材上有木節、裂痕等缺陷時會影響到木材整體結構力等，而不同種類的木節影響木材的程度也不盡相同；因而本論文研究目標有主要兩項：(1)對單一木節圖片數據集做正確分類。在此以Xception卷積類神經網路(Convolution Neural Networks)做為影像辨識與分類。(2)在雲杉木材上偵測出缺陷的存在，找出其位置且識別出正確的木節種類；
我們新設計的識別模型之構想基底來自遮罩型區域卷積類神經網路(Mask Region based Convolution Neural Networks,Mask-RCNN)。根據木材缺陷資料的特性，在某些木節與無缺陷木材過於相似下，無法精準的判斷是否為目標物，因而以Focal Loss取代原網路Cross-Entropy；比起得到更加精準的區域建議框框選大小，我們更加關心區域建議框內是否為缺陷，因此我們利用增加調整參數的方式改良損失函數(Loss Function)，以改良過往的木節識別模型提升其偵測精準度。

In this thesis we focus on two topics：(1) Classification of single wood knot image datasets. Here, the Xception convolutional neural network is used for image recognition.(2) Detection of defects on spruce wood by identifying the area of the mask and classifying the type of knots. Our idea is based on Mask Region based Convolution Neural Networks (Mask-RCNN). According to the nature of wood defects dataset, we modify the loss function in Mask-RCNN by replacing cross-entropy loss with focal loss as well as an additional tuning parameter to improve the detection performance.

1 緒論 1

2 文獻回顧 3
2.1 卷積類神經網路 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2 區域卷積類神經網路 . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.2.1 網絡設計 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.2.2 訓練 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.3 快速型區域卷積類神經網路 . . . . . . . . . . . . . . . . . . . . . . . 13
2.3.1 網絡設計 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.3.2 訓練 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.4 更快速型區域卷積類神經網路 . . . . . . . . . . . . . . . . . . . . . . 17
2.4.1 網路設計 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.4.2 訓練 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.5 遮罩區域卷積類神經網路 . . . . . . . . . . . . . . . . . . . . . . . . 22
2.5.1 網路設計 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.5.2 訓練 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3 研究方法 26
3.1 單一木節圖像分類 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.2 木節偵測暨分類 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.2.1 錨點改善 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.2.2 模型改良 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

4 實驗 31
4.1 資料來源與敘述 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.2 單一木節圖像分類 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
4.3 木節偵測暨分類 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
4.3.1 木材缺陷新標記 . . . . . . . . . . . . . . . . . . . . . . . . . 33
4.3.2 評量指標 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.3.3 模型改良 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

5 結論與未來展望 41

參考文獻 42

[1] Ross Girshick, Jeff Donahue, Trevor Darrell, and Jitendra Malik. Rich feature hierarchies for accurate object detection and semantic segmentation. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2014.

[2] Ross Girshick. Fast R-CNN. In Proceedings of the International Conference on Computer Vision (ICCV), 2015.

[3] Shaoqing Ren, Kaiming He, Ross Girshick, and Jian Sun. Faster R-CNN: Towards real-time object detection with region proposal networks. In Neural Information Processing Systems (NIPS), 2015.

[4] Kaiming He, Georgia Gkioxari, Piotr Dollár, and Ross Girshick. Mask R-CNN. In Proceedings of the International Conference on Computer Vision (ICCV), 2017.

[5] Tsung-Yi Lin, Priyal Goyal, Ross Girshick, Kaiming He, and Piotr Dollár. Focal loss for dense object detection. IEEE transactions on pattern analysis and machine intelligence, 2018.

[6] Steve Lawrence, C Lee Giles, Ah Chung Tsoi, and Andrew D Back. Face recognition: a convolutional neural-network approach. IEEE transactions on neural networks, 8(1):98–113, 1997.

[7] Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hinton. Imagenet classification with deep convolutional neural networks. In Advances in neural information processing systems, pages 1097–1105, 2012.

[8] François Chollet. Xception: Deep learning with depthwise separable convolutions. arXiv preprint, pages 1610–02357, 2017.

[9] Yann LeCun, Léon Bottou, Yoshua Bengio, and Patrick Haffner. Gradientbased learning applied to document recognition. Proceedings of the IEEE, 86 (11):2278–2324, 1998.

[10] Mate Szarvas, Akira Yoshizawa, Munetaka Yamamoto, and Jun Ogata. Pedestrian detection with convolutional neural networks. In Intelligent vehicles symposium, 2005. Proceedings. IEEE, pages 224–229. IEEE, 2005.

[11] Jasper RR Uijlings, Koen EA Van De Sande, Theo Gevers, and Arnold WM Smeulders. Selective search for object recognition. International journal of computer vision, 104(2):154–171, 2013.

[12] Karen Simonyan and Andrew Zisserman. Very deep convolutional networks for large-scale image recognition. arXiv preprint arXiv:1409.1556, 2014.

[13] Sinno Jialin Pan and Qiang Yang. A survey on transfer learning. IEEE Transactions on knowledge and data engineering, 22(10):1345–1359, 2010.

[14] Pedro F Felzenszwalb, Ross B Girshick, David McAllester, and Deva Ramanan.Object detection with discriminatively trained part-based models. IEEE trans-
actions on pattern analysis and machine intelligence, 32(9):1627–1645, 2010.

[15] Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Spatial pyramid pooling in deep convolutional networks for visual recognition. In European conference on computer vision, pages 346–361. Springer, 2014.

[16] Jonathan Long, Evan Shelhamer, and Trevor Darrell. Fully convolutional networks for semantic segmentation. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 3431–3440, 2015.

[17] Yann LeCun, Bernhard Boser, John S Denker, Donnie Henderson, Richard E Howard, Wayne Hubbard, and Lawrence D Jackel. Backpropagation applied to handwritten zip code recognition. Neural computation, 1(4):541–551, 1989.

[18] Yann A LeCun, Léon Bottou, Genevieve B Orr, and Klaus-Robert Müller. Efficient backprop. In Neural networks: Tricks of the trade, pages 9–48. Springer, 2012.

[19] Nitish Srivastava, Geoffrey Hinton, Alex Krizhevsky, Ilya Sutskever, and Ruslan Salakhutdinov. Dropout: a simple way to prevent neural networks from overfitting. The Journal of Machine Learning Research, 15(1):1929–1958, 2014.

[20] Olli Silven. Visual inspection of lumber, 2000. URL http://www.ee.oulu.fi/ ~olli/Projects/Lumber.Grading.html.

[21] Olli Silvén, Matti Niskanen, and Hannu Kauppinen. Wood inspection with non-supervised clustering. In COST action E10 Workshop - Wood properties for industrial use, 2000.

[22] Teuvo Kohonen. The self-organizing map. Proceedings of the IEEE, 78(9): 1464–1480, 1990.

[23] A. Dutta, A. Gupta, and A. Zissermann. VGG image annotator (VIA), 2016. URL http://www.robots.ox.ac.uk/~vgg/software/via/.