臺灣博碩士論文加值系統

在過去，以傳統機器學習辨識圖像的成敗跟特徵值的擷取往往有著密不可分的關係，假如沒能有效的提取出所要的特徵勢必會影響其結果的好壞，而隨著電腦硬體設備及運算能力的大幅提升，讓深度學習有了嶄露頭角的機會，其強大的自動抽取特徵能力，能夠有效地找尋出好的特徵來提高模型識別系統的效能。有鑑於此，本研究之目的在於應用多種二值化前處理演算法，並透過基於深度學習的卷積神經網路模組GoogLeNet模型來實作出一套影像辨識系統，我們將使用ImageNet這個全球最大的圖像識別數據庫來當作我們的訓練集資料，並選擇其中之五類圖庫集包括汽車、飛機、馬、船隻、卡車等來進行實驗。此系統嘗試使用多種演算法將圖片作灰階化及二值化處理，再利用GoogLeNet的Inception classification V1、Inception classification V2以及Inception classification V3等三個模型來進行訓練並試圖提高其原模型辨識結果之精確度。系統將分成兩個部分來實現，第一部分是影像前處理部分，將使用OpenCV所提供的五種不同選取閥值函數的演算法及大津演算法等六種方法對圖像進行前處理，第二部分則是將已經做完前處理的圖片集當作上述三組模型的訓練資料來進行訓練與測試，透過調整及更新參數來分析比較實驗之結果，最後我們將使用TensorBoard來觀察訓練過程中的各種視覺化資料，幫助我們理解、調整模型來優化出最佳的網路模型，並於實驗結果中顯示經過影像前處理的圖像是能有效提升模組辨識之正確率。

In the past, the success or failure of traditional machine learning to identify images is often inextricably linked to the extraction of feature values. The tremendous increase in equipment and computing power has given depth learning an opportunity to emerge, and its powerful automatic extraction feature capability can effectively find good features to improve the performance of the model recognition system. The purpose of this study is to apply a variety of binarization pre-processing algorithms, and to implement an image recognition system through the deep learning based convolutional neural network module GoogLeNet. We will use ImageNet, the world's largest image recognition database, and select five of them includes cars, airplanes, horses, boats and trucks as our training input data. The system attempts to grayscale and binarize the image using a variety of algorithms, and then uses GoogLeNet's Inception classification V1, Inception classification V2 and Inception classification V3 to train and try to improve the original model identification results accuracy. It will be implemented in two parts. First, the image pre-processing part, the image will be pre-processed using six different methods of selecting the threshold function provided by OpenCV and the Otsu algorithm. Second, using three different versions models to train and test the data that have already been image pre-processing. Through adjusting and updating the parameters to analyze and compare the results of the experiment. Finally we will use TensorBoard to observe the training process. Various visual data help us to understand and adjust the model to find out the optimal network model, and show that the image pre-processed image can effectively improve the correct rate of model identification.

摘要
Abstract
致謝
目錄
表目錄
圖目錄
一、緒論
1.1 研究背景
1.2 研究動機與目的
1.3 論文架構
二、文獻回顧
2.1 機器學習
2.1.1 類神經網路
2.1.2 卷積神經網路
2.2 深度學習
2.2.1 深度學習歷史
2.3 影像前處理
三、研究方法
3.1 前置處理
3.2 影像前處理方法
3.3 GoogLeNet Inception訓練模組
四、研究結果分析與比較
4.1 研究結果
五、結論與未來展望
參考文獻

[1]A. M. Turing, "Computing Machinery and Intelligence. Mind 49: 433-460. " ,950
[2]Silver, D. et al.,“Mastering the Game of Go with Deep Neural Networks and Tree Search”, Nature, vol. 529, pp. 484-489, 2016.
[3]C. Affonso, A. L. D. Rossi, F. H. A. Vieira, and A. C. P. de Leon Ferreira de Carvalho, ‘‘Deep learning for biological image classification,’’ Expert Systems with Applications, 85, 114-122, 2017.
[4]Y. LeCun, Y. Bengio, and G. Hinton, ‘‘ Deep learning,’’ Nature, 521(7553), 436-444, 2015.
[5]Russakovsky, Olga, et al. "Imagenet large scale visual recognition challenge." International Journal of Computer Vision 115.3 (2015): 211-252.
[6]Krizhevsky, Alex, Ilya Sutskever, and Geoffrey E. Hinton. "Imagenet classification with deep convolutional neural networks." Advances in neural information processing systems. 2012.
[7]Bengio, Yoshua, Patrice Simard, and Paolo Frasconi. "Learning long-term dependencies with gradient descent is difficult." IEEE transactions on neural networks 5.2 (1994): 157-166.
[8]Russell, Stuart J., and Peter Norvig. Artificial intelligence: a modern approach. Malaysia; Pearson Education Limited,, 2016.
[9]Hastie, Trevor, Robert Tibshirani, and Jerome Friedman. "Unsupervised learning." The elements of statistical learning. Springer, New York, NY, 2009. 485-585.
[10]N. Gupta, ‘‘Artificial neural network,’’ Network and Complex Systems, 3(1), 24-28, 2013.
[11]W. Cao, X. Wang, Z. Ming, and J. Gao, ‘‘A review on neural networks with random weights,’’ Neurocomputing, 2017.
[12]J. A. Freeman and D. M. Skapura, ‘‘Algorithms, Applications, and Programming Techniques,’’ Addison-Wesley Publishing Company, USA, 1991.
[13]斎藤康毅，2017，Deep Learning :用python進行深度學習的基礎理論實作，吳嘉芳，初版，碁峰資訊股份有限公司，台北市
[14]Danielsson, Per-Erik. "Euclidean distance mapping." Computer Graphics and image processing 14.3 (1980): 227-248.
[15]林大貴，2017，TensorFlow + Keras 深度學習人工智慧實務應用，初版，博碩文化股份有限公司，新北市
[16]R. Tauler, B. Walczak, and S. D. Brown, ‘‘Comprehensive chemometrics: chemical and biochemical data analysis,’’. Elsevier, 2009.
[17]F. Sadoughi, M. Ghaderzadeh, M. Solimany, and R. Fein, ‘‘An Intelligent System Based on Back Propagation Neural Network and Particle Swarm Optimization for Detection of Prostate Cancer from Benign Hyperplasia of Prostate,’’ J Health Med Informat, 5(158), 2, 2014.
[18]高揚，2018，白話深度學習與TensorFlow，初版，碁峰資訊股份有限公司，台北市
[19]黃文堅、唐源，2017，實戰 TensorFlow Google 深度學習，蔡勝文，初版，碁峰資訊股份有限公司，台北市
[20]Deng, Jia, et al. "Imagenet: A large-scale hierarchical image database." Computer Vision and Pattern Recognition, 2009. CVPR 2009. IEEE Conference on. Ieee, 2009. (http://dx.doi.org/10.1109/CVPR.2009.5206848)
[21]Y . Lecun.Bottou,Y.Bengio,and P.Haffner(1998):Gradient-based learning applied to document recognition.Proceedings of the IEEE86,11(November 1998),2278-2324.DOI: (http://dx.doi.org/10.1109/5.726791)
[22]Karen Simonyan and Andrew Zisserman(2014) : Very Deep Convolutional Networks for Large-Scale Image Recognition.aiXiv: 1409.1556[cs](September 2014).
[23]Christian Szegedy et al (2015) :Going Deeper with Convolutions .In The IEEE Conference on Computer Vision and Pattern Recongnition(CVPR).
[24]Szegedy, Christian, et al. "Going deeper with convolutions." Proceedings of the IEEE conference on computer vision and pattern recognition. 2015.
[25]Ioffe, Sergey, and Christian Szegedy. "Batch normalization: Accelerating deep network training by reducing internal covariate shift." arXiv preprint arXiv:1502.03167 (2015).
[26]Szegedy, Christian, et al. "Rethinking the inception architecture for computer vision." Proceedings of the IEEE conference on computer vision and pattern recognition. 2016.
[27]Otsu, Nobuyuki. "A threshold selection method from gray-level histograms." IEEE transactions on systems, man, and cybernetics 9.1 (1979): 62-66.