臺灣博碩士論文加值系統

目的 : 為了利於牧場對乳牛身分的管理，本研究使用卷積神經網路進行乳牛臉部影像識別取代傳統識別方式，以協助牧場管理工作之進行。
材料與方法 : 本研究共蒐集54頭乳牛976張照片，採用Tensorflow平台模型為訓練卷積神經網路的核心架構，監督式學習使用LabelImg軟體套件標記出乳牛臉部位置及編號，並以Adobe Lightroom修正部分影像曝光不足、銳利化臉部模糊區域，以Python撰寫之驗證腳本載入訓練完成時於凍結圖所儲存的模型權重值，以108張影像依序驗證模型辨識能力。
結果 : 結果發現使用卷積神經網路模型應用於乳牛辨識，以432張影像作為訓練資料集其模型有效辨識率為76.92%，若將訓練集影像增加至847張則模型有效辨識率達到86.14%，兩組資料集於Epoch=60、Epoch=80無顯著差異。
結論 : 由於採用非侵入式也能避免傳統烙印造成傷害，或標示牌所產生之異物感、牛隻互咬、傷口感染等危害。雖然辨識率略低於RFID技術，其原因為深度學習模型仰賴圖片中的線條去分類不同類別，於不同角度拍攝立體物件會產生截然不同的平面資訊，偶爾因角度偏差所造成辨識錯誤，但在有效辨識率86.14%的情況下，針對同一隻牛的影片進行連續圖片擷取再送入模型辨識，因大部分的照片都能辨識正確，故可以解決少部分照片因角度偏差造成的錯誤辨識。

Purpose: In order to facilitate the management of dairy cattle identity recognition in the ranch, this study used convolutional neural networks (CNN) to replace the traditional recognition methods for face recognition to assist the management.
Materials and methods: A total of 976 images of 54 dairy cattle were collected in this study. The Tensorflow platform model was used as the core architecture for training CNN. LabelImg software suite was used to mark dairy cattle faces and numbers for supervised learning, and Adobe Lightroom was used to correct the underexposure and sharpen blurred areas on the dairy cattle faces. The verification script written in Python was loaded with the model’s weights and biases values stored in the inference graph when the training was completed. One hundred and eight images were used to sequentially verify the model recognition ability.
Results: It was found that the effective recognition rate of this CNN based model for dairy cattle identification was 76.92% with 432 images as the training data set. If the number of images was increased to 847, the effective recognition rate is 86.14%. There was no significant difference between the two sets of data at Epoch = 60 and Epoch =80.
Conclusions: With image recognition model, dairy cattle can be identified quickly and accurately. The non-invasive method also avoided the harm caused by traditional branding or the foreign body sensation, mutual bites of dairy cattle, and wound infection caused by tags. The recognition rate of the image recognition model was slightly lower than that of RFID technology. The reason was that the deep learning model relied on the lines in the image to classify different categories. Shooting a three-dimensional object from different angles will produce completely different graphic information which can lead to recognition errors occasionally. However, the effective recognition rate of 86.14% was obtained by continuous image capture of the same cattle's video and was fed into the recognition model. Since most of the images can be correctly identified, it can solve the misidentifications caused by angular deviation.

摘要
Abstract
致謝
目錄
表目錄
圖目錄
符號縮寫
第一章 緒論
1.1 動機
1.2 目的
1.3 文獻探討
第二章 材料與方法
2.1 前言
2.2 資料收集
2.3 資料前處理
2.4 監督式學習
2.5 凍結圖(Inference graph)
2.6 多層感知器(Fully-connected multilayer perceptron)
2.7 激勵函數(Activate function)
2.8 卷積神經網路(Convolutional Neural Network)
2.8.1 卷積層(Convolution layer)
2.8.2 池化層(Pooling layer)
2.8.3 全連接層(Fully connected layers)
2.9 Faster region-based Convolutional Neural Network
2.9.1 Region-based CNN
2.9.2 Fast region-based CNN
2.9.3 Faster region-based CNN
第三章 結果與討論
3.1 前言
3.2 資料前處理 - 影像銳利化
3.3 資料前處理 - 增強曝光
3.4 模型驗證統計資料
3.5 模型辨識影響因素
3.5.1 拍攝角度影響模型辨識能力
3.5.2 全黑牛隻不易辨識
3.5.3 裁切像素
3.5.4 增加資料集
3.6 背景分離、補色
3.7 類別之間特徵過於相似
3.8 影像轉換對辨識率之影響
3.9 修正後模型辨識率
3.10 修正裁切影像後模型辨識率
3.11 影像辨識速度
3.12 研究限制與未來研究方向
第四章 結論
參考文獻
自傳

[1]張兆鋒, 蔡方喻, 宋政杰, 柯景瀚, 卓芝余, and 許文玲, "自動化牧場," 2012.
[2]陳泰元 and 阮喜文, "乳牛場線上記錄與管理系統," 2004.
[3]R. Want, "An introduction to RFID technology," IEEE pervasive computing, pp. 25-33, 2006.
[4]耿雪霏, "RFID 技術在物流管理中的應用," 包裝工程, vol. 26, pp. 118-119, 2005.
[5]張榮輝, "RFID 應用於圖書館的尋書定位服務," 成功大學工程科學系碩士在職專班學位論文, pp. 1-65, 2008.
[6]T.-M. LIN, "RFID 技術應用於停車管理之規劃," National Central University, 2007.
[7]莊士德, "調查牛乳房炎導致產業之經濟影響及建立台灣乳牛因應氣候變遷之健康管理標準模式," 2012.
[8]陳政廷, "乳房炎乳牛於抗生素治療之發炎-緩解過程全身及乳房自主防禦功能之變化與乳房之重整," 中興大學動物科學系所學位論文, pp. 1-85, 2014.
[9]O. M. Parkhi, A. Vedaldi, and A. Zisserman, "Deep face recognition," bmvc, p. 6, 2015.
[10]J. Liu, Y. Deng, T. Bai, Z. Wei, and C. Huang, "Targeting ultimate accuracy: Face recognition via deep embedding," arXiv preprint arXiv:1506.07310, 2015.
[11]V. John, K. Yoneda, B. Qi, Z. Liu, and S. Mita, "Traffic light recognition in varying illumination using deep learning and saliency map," 17th International IEEE Conference on Intelligent Transportation Systems (ITSC), pp. 2286-2291, 2014.
[12]G. Litjens, T. Kooi, B. E. Bejnordi, A. A. A. Setio, F. Ciompi, M. Ghafoorian, et al., "A survey on deep learning in medical image analysis," Medical image analysis, vol. 42, pp. 60-88, 2017.
[13]B. Huval, T. Wang, S. Tandon, J. Kiske, W. Song, J. Pazhayampallil, et al., "An empirical evaluation of deep learning on highway driving," arXiv preprint arXiv:1504.01716, 2015.
[14]A. S. Voulodimos, C. Z. Patrikakis, A. B. Sideridis, V. A. Ntafis, and E. M. Xylouri, "A complete farm management system based on animal identification using RFID technology," Computers and electronics in agriculture, vol. 70, pp. 380-388, 2010.
[15]陳季希, "為視覺障礙人士開發之深度追跡導盲機器犬," 交通大學電機與控制工程系所學位論文, pp. 1-64, 2018.
[16]陳昱丞 and 周志成, "基於卷積神經網路之車牌辨識系統," 2017.
[17]M. Dong, D. He, C. Luo, D. Liu, and W. Zeng, "A CNN-Based Approach for Automatic License Plate Recognition in the Wild," BMVC, 2017.
[18]K.-N. Hsieh, "使用深度與彩色影像的卷積神經網路做倒車障礙物偵測," National Central University, 2018.
[19]S. Kumar, A. Pandey, K. S. R. Satwik, S. Kumar, S. K. Singh, A. K. Singh, et al., "Deep learning framework for recognition of cattle using muzzle point image pattern," Measurement, vol. 116, pp. 1-17, 2018.
[20]李哲緯, "基於卷積神經網路之手語影像辨識," 2016.
[21]曾定章, 謝鈞惟, and 黎尚昆, "縮小更快速區域卷積神經網路的前車偵測與辨識," TANET2018 臺灣網際網路研討會, pp. 1316-1321, 2018.
[22]Y.-C. Tsai, "乳牛熱緊迫影像監控系統之建置與資料分析," 臺灣大學生物產業機電工程學研究所學位論文, pp. 1-100, 2018.
[23]A. Krizhevsky, I. Sutskever, and G. E. Hinton, "Imagenet classification with deep convolutional neural networks," Advances in neural information processing systems, pp. 1097-1105, 2012.
[24]D. C. Ciresan, U. Meier, J. Masci, L. M. Gambardella, and J. Schmidhuber, "Flexible, high performance convolutional neural networks for image classification," Twenty-Second International Joint Conference on Artificial Intelligence, 2011.
[25]M. A. Hearst, S. T. Dumais, E. Osuna, J. Platt, and B. Scholkopf, "Support vector machines," IEEE Intelligent Systems and their applications, vol. 13, pp. 18-28, 1998.
[26]J. A. Suykens and J. Vandewalle, "Least squares support vector machine classifiers," Neural processing letters, vol. 9, pp. 293-300, 1999.
[27]S. S. Mohamed, W. A. Mohamed, A. Khalil, and A. Mohra, "Deep Learning Face Detection and Recognition." , 2015.
[28]R. A. Vyas and S. Shah, "Comparision of PCA and LDA techniques for face recognition feature based extraction with accuracy enhancement," International Research Journal of Engineering and Technology (IRJET), vol. 4, pp. 3332-3336, 2017.
[29]C. K. Tran, T. F. Lee, L. Chang, and P. J. Chao, "Face description with local binary patterns and local ternary patterns: improving face recognition performance using similarity feature-based selection and classification algorithm," International Symposium on Computer, Consumer and Control, pp. 520-524, 2014.
[30]邱彥豪, 吳昆儒, 梁家銘, 陳建志, and 曾煜棋, "應用於智慧農場之影像電子圍籬," TANET2018 臺灣網際網路研討會, pp. 84-85, 2018.
[31]林晏僖, "中文名詞組的辨識 : 規則式判別, 監督式, 半監督式與非監督式學習法的實驗," 臺灣大學資訊網路與多媒體研究所學位論文, pp. 1-61, 2010.
[32]M. W. Gardner and S. Dorling, "Artificial neural networks (the multilayer perceptron)—a review of applications in the atmospheric sciences," Atmospheric environment, vol. 32, pp. 2627-2636, 1998.
[33]J. Tang, C. Deng, and G.-B. Huang, "Extreme learning machine for multilayer perceptron," IEEE transactions on neural networks and learning systems, vol. 27, pp. 809-821, 2015.
[34]J. Hosang, R. Benenson, P. Dollár, and B. Schiele, "What makes for effective detection proposals?," IEEE transactions on pattern analysis and machine intelligence, vol. 38, pp. 814-830, 2015.
[35]R. Girshick, J. Donahue, T. Darrell, and J. Malik, "Rich feature hierarchies for accurate object detection and semantic segmentation," IEEE conference on computer vision and pattern recognition, pp. 580-587, 2014.
[36]R. Girshick, "Fast r-cnn," IEEE international conference on computer vision, pp. 1440-1448, 2015.
[37]S. Ren, K. He, R. Girshick, and J. Sun, "Faster r-cnn: Towards real-time object detection with region proposal networks," Advances in neural information processing systems, pp. 91-99, 2015.
[38]A. Salvador, X. Giró-i-Nieto, F. Marqués, and S. i. Satoh, "Faster r-cnn features for instance search," IEEE Conference on Computer Vision and Pattern Recognition Workshops, pp. 9-16, 2016.
[39]陳奕瑄, "以 CNN 進行植物圖片的辨識以及經過處理後的再辨識," 臺灣大學資訊工程學研究所學位論文, pp. 1-51, 2018.
[40]謝斯宇, "基於臉部偵測及 CNN 模型之硬體臉部辨識系統," 2019.