臺灣博碩士論文加值系統

駕駛人視野清晰度是雨天行車安全與行駛速度最重要的關鍵，現今多數車輛仍然使用手動方式去調整雨刷運作頻率。而隨著近年電腦運算效能的進步，卷積神經網路於影像辨識上有許多突破，過去本團隊提出使用卷積神經網路進行雨水辨識的G(N)，其準確度為77%。由於G(N)使用完整影像集N進行訓練，面對多變的背景環境，需要收集大量且多樣數據訓練，於收集數據與整理則需花費大量時間。因此本文則提出四種系統化的數據生成方式，並將四種方式所生成的數據，分別利用卷積神經網路進行訓練，分別為：1.利用雨水模板生成雨水圖像集M訓練的G(M)；2.利用生成對抗網路所生成圖像H，但由於H圖片破損故未進行訓練；3.利用高斯裁切抽取現有數據庫C訓練的G(C)；4.利用柔化與銳利化改變圖像邊緣生成訓練集E並訓練出G(E)，且將訓練出來的辨識網路與G(N)進行比較。其中G(C)與G(N)進行比較有明顯進步，前者的準確度與後者相比提升了8%，且對於雨水圖片的召回率提升了15%。

As the computer performs better in recent years, convolutional neural networks have made many breakthroughs in image recognition. Our team proposed to use a convolutional neural network for rainwater identification G(N) and to obtain the result of accuracy of 77%. And the complete image training dataset N used by G(N) for training. It is necessary to collect a large amount of data and diverse data for training as a result of the diverse environment, and it takes a lot of time to collect and classify the data. Therefore, we propose four systematic data generation methods and put the result of generation images into convolutional neural networks to train the data. 1. Use the rainwater template to generate M and train a CNN as G(M). 2. The H images are generated by the Generative Adversarial Network, but the training is not performed because the H images are corrupt. 3. Use Gaussian extraction to extract the image to generate a training set C, and to do a CNN training as G(C). 4. Put images into the softening and sharpening process to generate a training set E and train it as G(E). And we compare the result of new network models with G(N). As a result, the G(C) is better than G(N) apparently. The accuracy of the former has increased by 10% compared with the latter, and the recall rate of rainwater pictures has increased by 15%.

摘要 i
ABSTRACT ii
誌 謝 iii
目 錄 iv
表目錄 vi
圖目錄 vii
第一章 緒論 1
1.1前言 1
1.2研究動機 2
1.3研究目標 3
1.4研究流程 4
第二章 文獻探討 5
2.1視覺雨水辨識 5
2.2卷積神經網路 6
2.3深度殘差學習網路 7
2.4雨水模擬 9
第三章 訓練數據生成與增強 10
3.1傳統影像收集法 10
3.2人造雨水模擬實驗 12
3.3生成對抗網路(GAN)雨水模板生成 15
3.4高斯隨機抽取 20
3.5柔化與銳利化 24
第四章 雨水辨識器訓練 27
4.1深度殘差學習網路訓練 27
4.1.1人造雨水模擬訓練 27
4.1.2高斯抽樣訓練 29
4.1.3柔化與銳利化訓練 31
第五章 雨水辨識測試實驗與結果 33
5.1測試集驗證 33
5.1.1原始圖片驗證 34
5.1.2裁切圖片驗證 45
5.1.3柔化與銳利化驗證 50
5.2網路泛化實驗 52
第六章 結論與未來展望 60
6.1結論 60
6.2未來展望 61
參考文獻 62

1.S. Shladover, "Review of the state of development of advanced vehicle control systems (AVCS)," Vehicle Syst. Dyn., vol. 24, pp. 551-595, July 1995.
2.A.H.M Fazle Elahi, M. S. Rahman, "Intelligent windshield for automotive vehicles," 17th International Conference on Computer and Information Technology (ICCIT), 2014.
3.C. C. Lai, C. H. G. Li, "Video-Based Windshield Rain Detection and Wiper Control Using Holistic-View Deep Learning," IEEE 15th International Conference on Automation Science and Engineering (CASE), 2019.
4.H. Kurihata, T. Takahashi, I. Ide, Y. Mekada, H. Murase, Y. Tamatsu, T. Miyahara, "Rainy Weather Recognition from In-Vehicle Camera Images for Driver Assistance," IEEE Proceedings, Intelligent Vehicles Symposium, 2005.
5.J. H. Park, M.H. Kim, H. J. Im, K. C . Lee,S. Lee,"Development of Vision based Control Smart Windshield WiperSystem for Intelligent Vehicle," SICE-ICASE International Joint Conference, 2006, pp.4398-4403.
6.F. A.Machot, M. Ali,A. H. Mosa, C. Schwarzlmüller, M. Gutmann, "Real-time raindrop detection based on cellular neural networks for ADAS," Journal of Real-Time Image Processing, 2016, pp.1-13.
7.A. Krizhevsky, I.Sutskever, and G. E. Hinton, "ImageNet Classification with Deep Convolutional Neural Networks," in Proc. Advances in Neural Information Processing Systems (NIPS), 2012.
8.S. Albawi, T. Mohammed and S. Al-Zawi,"Understanding of a convolutional neural network", Proceedings of ICET, Aug 2017.
9.J. Deng, A.C. Berg, K. Li, Fei-Fei Li., "What does classifying more than 10,000 image categories tell us?," in ECCV 2010, Lecture Notes in Computer Science, Springer, Berlin, Heidelberg, 2010.
10.Q. Le, M. Ranzato, R. Monga, M. Devin, K. Chen, G. Corrado, J. Dean, A. Ng, "Building high-level features using large scale unsupervised learning," International Conference in Machine Learning, 2012.
11.S. Levine, C. Finn, T. Darrell and P. Abbeel, "End-to-end training of deep visuomotor policies", J. Mach. Learn. Res., vol. 17, no. 39, pp. 1-40, 2016.
12.C. N. Dos Santos and M. Gatti, "Deep convolutional neural networks for sentiment analysis of short texts," Proc. Int. Conf. Computational Linguistics, pp. 69-78, 2014.
13.R. Johnson and T. Zhang, "Effective use of word order for text categorization with convolutional neural networks," Proc. Hum. Lang. Technol. Annu. Conf. North Amer. Chapter, pp. 103-112, 2015.
14.K. He, X. Zhang, S. Ren, J. Sun, "Deep Residual Learning for Image Recognition," In Proceedings of CVPR, 2016, pp. 770-778.
15.中央氣象局，大豪雨降雨量標準，雨量分級定義與警戒事項說明資料，交通部，台灣，2020。
16.I. J. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. C. Courville, and Y. Bengio, "Generative adversarial nets," In Proceedings of NIPS, pages 2672– 2680, 2014.
17.Y. H. Huang, C. H. G. Li, Y. M. Chang, "Illumination-Robust Object Coor-dinate Detection by Adopting Pix2Pix GAN for Training Image Generation," In TAAI, 2019, pp. 1-6.
18.P. Isola, J.-Y. Zhu, T. Zhou and A. A. Efros, "Image-to-image translation with conditional adversarial networks," Proc. IEEE Conf. Comput. Vis. Pattern Recognit., pp. 5967-5976, 2017.
19.R. V. Hogg, A. T. Craig. Introduction to Mathematical Statistics (5th Edition)(M). Prentice Hall ,2004.
20.陳奎文，執行深度學習視覺偵測之崁入式運算平台，碩士論文，國立臺北科技大學機電整合研究所，臺北，2020。