臺灣博碩士論文加值系統

近年來由於無人飛行載具盛行，許多行業都開始跟隨了這波熱潮，航空攝影測量是最近幾年流行的測量技術，解決了以往傳統測量的不方便性，並透過高精度的模型等比例還原該地區的面貌，進而做出測量，但是單純的航拍攝影無法掌控模型的精確程度，因此需要在拍攝前去佈置航測標，來提高建置模型的精度與時間，而自從多視點影像技術出現後，許多相關的軟體對於這些標記點能夠高精準度的標記航測標的中心位置，透過這樣的功能解決以往傳統航空攝影測量的不方便性，例如建模前需要花費大量人力時間進行人工刺點，但是多視點影像技術不是毫無缺點，因為主要演算法採用SIFT，此算法的缺點在於辨識上需要許多條件才能達成所謂的高精度辨識。

深度學習是現今的熱潮，許多行業都結合此技術開發相關應用，但目前測量業並未有相關應用，原因在於台灣相關行業目前還是以傳統測量為主，航空攝影測量為輔，再加上需要具備相關專業知識，因此本研究嘗試使用深度學習中的物體檢測技術來辨識航測標，模擬自動化標記點且改善拼接軟體中的缺點，本研究採用YOLO作為物體檢測的模型，擁有快速且完整的權重模型，現今的技術講求的是高幀數的辨識，對於錨點與目標範圍還有待加強，而航測標的技術是透過照片來生成模型，因此可以不需改變YOLO內部結構。本研究以四分割法來提高模型效能，透過自定義非極大值抑制(NMS)將多餘錨點消除，對於講求精度的測量界來說，直觀上置信度高低也就決定精度高低，如果讓模型的置性區間提高，預測位置與交集面積相互重疊率越高，則獲取物體中心點的機率就越高。

In recent years, Unmanned Aerial Vehicle have been prevalent in the world and have been applied in many industries. Aerial photogrammetry using the newest measured approach to solve the inconvenient measurement previously, and return to the original features through one to one scale high precision models. However, because general aerial photogrammetry cannot control precision of models, we need to do targeting to increase accuracy and time in model development before taking the pictures. Since Image Synthesis Technology for Multi-view 3D Display appeared, numerous software can mark targeting central position correctly, and it reduce the time of artificially marked point. Nevertheless, Image Synthesis Technology still have a disadvantage, its algorithm adapts SIFT system. This mean that there must have plenty of recognizable conditions so that it can achieve high accuracy identification.

Nowadays, deep learning is the main trend that variety of industries apply; however, measurement industry has not executed any kinds of event. The reason is that related industries mainly rely on traditional measurement supplemented by Aerial photogrammetry in Taiwan. As a result, this research using object detection with deep learning to identify targeting, also to simulate mark automatically, and improve disadvantage of SILF. YOLO having fast and completed weight model; nevertheless, it bounding box function cannot achieve high frame identification for object detection. People do not need to change YOLO internal structure. Increasing efficacy by four spilt method, and removing unnecessary anchor points by Non-Maximum Suppression (NMS).

中文摘要 i
英文摘要 ii
誌謝 iv
目錄 v
表目錄 viii
圖目錄 x
第一章、緒論 1
1.1研究背景 1
1.2研究動機 3
1.3研究之重要性 3
1.4研究目的 4
1.5研究流程與架構 4
第二章 文獻探討 8
2.1 航測標記點(Aerial Photogrammetry Target) 8
2.2 三維正射影像建置 9
2.3人工智慧(Artificial Intelligence, AI) 10
2.4 機器學習(Machine Learning, ML) 11
2.4.1 K-平均演算法 (K-means Clustering, K-means） 12
2.4.2 機器學習模型評估 13
2.5 深度學習(Deep Learning ,DL) 14
2.5.1 深度卷積網路(Convolutional Neural Networks，CNN) 15
2.5.2 深度殘差網路（Deep Residual Network，DRN） 18
2.5.3 YOLO(You Only Look Once)與Faster R-CNN 20
2.5.4 雅卡爾指數(lntersection over Union , IoU) 22
2.5.5 平均精度均值 (Mean Average Precision，mAP) 23
第三章 研究方法 24
3.1 航測參與 24
3.2 資料庫收集 26
3.3 模型架構 27
3.4改良型YOLOv3檢測介紹-四分割法 29
3.4.1 坐標說明: 29
3.4.2 處理被分割目標坐標: 30
3.4.3 四分割法驗證 31
3.5系統環境 32
3.6系統架構說明 33
3.7 蒐集PhotoScan的資料方法 35
3.7.1 文獻蒐集與使用者訪談 35
第四章 系統實作與研究結果 37
4.1 實驗步驟 37
4.2 PhotoScan刺點使用測試 39
4.2.1 因素測試1-低光源 40
4.2.2 因素測試2-水平平移 41
4.2.3 因素測試3-反光背景 42
4.2.4 因素測試2-多幾何圖形背景 43
4.2.5 PhotoScan結論 44
4.3 YOLO模型建置 46
4.3.1格式的建置說明 46
4.3.2 標籤生成 47
4.3.3 資料強化設定 47
4.3.4 訓練次數 48
4.3.5 YOLOv3低光源測試 48
4.3.6 YOLOv3的水平移動測試 50
4.3.7 YOLOv3的多幾何圖形背景測試 51
4.3.8 YOLOv3噪點測試 52
4.3.9 YOLOv3 10公尺距離測試 54
4.3.10 YOLOv3總結 56
4.4 四分割法模型建置 57
4.4.1 自定義NMS 59
4.4.2 四分割法低光源測試 60
4.4.3 四分割法水平移動測試 62
4.4.4 四分割法多幾何圖形背景測試 64
4.4.5 四分割法噪點測試 66
4.4.6 四分割法10公尺距離測試 67
4.4.7 四分割法15公尺與20公尺距離測試 69
4.4.8四分割法總結 72
4.5系統比較 73
4.5.1 模型距離系統距離比較 73
4.5.2模型效能指標比較 74
4.5.3 模型系統性能比較 75
4.5.4 辨識度比較 76
4.5.5占比率比較 77
4.5.6 實際航拍結果 77
4.5.7比較總結 81
第五章 討論 82
5.1研究限制與範圍 82
5.2 資料庫的限制 82
5.3 硬體的限制 82
5.4 標記點偵測的缺陷 83
5.5 深度學習網路架構的使用 83
5.6 四分割法改良型 83
第六章 結論與未來方向 85
文獻參考 86

[1]李欣宜，UAS航測於地籍圖重測效率提升之研究 -以苗栗縣竹南鎮鹽館前

段為例，台南:嘉南藥理大學應用空間資訊系，2019。

[2]黃建倫，UAS航測加速地籍圖整理之研究-以台東縣蘭嶼鄉椰油村為館前

段為例，台南:嘉南藥理大學應用空間資訊系，2019。

[3]陳癸杏，不同航測標於影像自動標記之研究。台南:嘉南藥理大學應用空間資訊系，2018。

[4]Hung-Lin Lin, Hsin-Yi Lee, Hung-I Huang, Kuo-Hsiang Tseng, Yi-Chan Tsai
. "Comparison of Horizontal Precision of Various Aerial Survey Marks in
Orthophoto Production. " Yunlin, Taiwan, Nov 16-18. 2018.

[5]Y. Lai and S. Lai, "Emotion-Preserving Representation Learning via
Generative Adversarial Network for Multi-View Facial Expression Recognition
," 2018 13th IEEE International Conference on Automatic Face & Gesture
Recognition (FG 2018), Xi''an, 2018, pp. 263-270, doi:10.1109/FG.2018.00046.

[6]Abhay Kumar. K-Means Clustering Algorithm , https://bit.ly/2yfVJUe,

Accessed on: 2018/7/13.

[7]Irfan Danish , Introduction to Deep Learning , https://reurl.cc/5gZm Ev,
Accessed on: 2019/6/1.

[8]Albelwi, S.; Mahmood, A. A Framework for Designing the Architectures of
Deep Convolutional Neural Networks. Entropy 2017, 19, 242.

[9]Jason Brownlee, How Do Convolutional Layers Work in Deep Learning
Neural Networks?, https://reurl.cc/qD47Ep, Accessed on: 2019/9/26.

[10]Anh H. Reynolds, Convolutional Neural Networks (CNNs),https:// reurl.cc/E
KMa9g, Accessed on: 2019.

[11]Yeh Janes，[資料分析&機器學習]第5.1講:卷積神經網絡介紹(Convolutional
Neural Network)，https://reurl.cc/W4M5rD，上網日期:2017/11/25。

[12]Michael Klear, The Sparse Future of Deep Learning, https://reurl.cc/gv3jAR
, Accessed on: 2018/11/25.

[13]K. He, X. Zhang, S. Ren and J. Sun, "Deep Residual Learning for Image
Recognition, " 2016 IEEE Conference on Computer Vision and Pattern
Recognition (CVPR), Las Vegas, NV, 2016, pp. 770-778, doi: 10.1109/ CVPR.
2016.90.

[14]J. Redmon and A. Farhadi, "YOLO9000: Better, Faster, Stronger," 2017 IEEE
Conference on Computer Vision and Pattern Recognition (CVPR), Honolulu,
HI, 2017, pp. 6517-6525, doi: 10.1109/CVPR.2017.690.

[15]Q. Xu, R. Lin, H. Yue, H. Huang, Y. Yang and Z. Yao, "Research on Small
Target Detection in Driving Scenarios Based on Improved Yolo Network,"
in IEEE Access, vol. 8, pp. 27574-27583, 2020, doi: 10.1109/ ACCESS 2020
.2966328.

[16]Aurelien Geron, Hands-on Machine Learning with Scikit-Learn, Keras &
TensorFlow, O’REILLY, 2nd Edition, 2019.

[17]M. Ju, J. Luo, P. Zhang, M. He and H. Luo, "A Simple and Efficient Network
for Small Target Detection," in IEEE Access, vol. 7, pp. 85771-85781, 2019,
doi: 10.1109/ACCESS.2019.2924960.

[18]Wen Wu and Hao Zhou . "Data-Driven Diagnosis of Cervical Cancer With

Support Vector Machine-Based Approaches", Data-Driven Monitoring, Fault
Diagnosis and Control of Cyber-Physical Systems. IEEE Access(5), 842 –862,

2017.

[19]Yaseen Alwesabi,Avishek Choudhury and Daehan Won. "Classification of

Cervical Cancer Dataset", IISE Annual Conference and Expo 2018, Orlando,
2018.

[20]Hung-Lin Lin etc., "Efficiency Enhancement on Urban Planning Stake in
urban planning using High-precision True-Orthphoto", IEEE, ICASI, Sapporo,
Japan, May 13-17, 2017.

[21]Wenbo Lan , Jianwu Dang , Yangping Wang and Song Wang, "Pedestrian

Detection Based on YOLO Network Model", IEEE, ICMA, Changchun, China
, Aug 5-8,2018.

[22]Wang Yang ,Zheng Jiachun, "Real-time face detection based on YOLO",

IEEE, ICKII, Jeju, South Korea, Jul 23-27, 2018.

[23]Yuanyuan Hu ,Xinjian Wu ,Guangdi Zheng and Xiaofei Liu, "Object

Detection of UAV for Anti-UAV Based on Improved YOLO v3",IEEE,CCC,
Guangzhou, China, Jul 27-30, 2019.

[24]Zhi Xu , Haochen Shi , Ning Li , Chao Xiang and Huiyu Zhou, "Vehicle

Detection Under UAV Based on Optimal Dense YOLO Method", IEEE,
ICSAI, Nanjing, China, Nov 10-12, 2018.

[25]Alhussein Fawzi and Horst Samulowitz , Deepak Turaga , Pascal Frossard,

"Adaptive data augmentation for image classification", IEEE, ICIP, Phoenix,
AZ, USA, Sept 25-28, 2016.

[26]Xuhong Wang , Hongyi Huang , Yue Hu and Yupu Yang, "Partial Discharge

Pattern Recognition with Data Augmentation based on Generative Adversarial
Networks", IEEE, CMD, Perth, WA, Australia, Sept 23-26, 2018.

[27]Hung-Lin Lin, Chien-Lun Huang, Zhi-Yuan Huang, Shun-Jie Jhang and Liang-
Ting Wu, "Study on the Efficiency of Automatic Matching for Various Aerial
Survey Marks", IEEE, ICAM, Yunlin, Taiwan, Nov 16-18, 2018.

[28]Ying-Hsiu Lai and Shang-Hong Lai, "Emotion-Preserving Representation

Learning via Generative Adversarial Network for Multi-View Facial
Expression Recognition", IEEE, FG, Xi''an, China, May 15-19, 2018.

[29]Hongjiang Shi , Lu Wang , Guangtai Ding , Fenglei Yang and Xiaoqiang Li

, "Data Augmentation with Improved Generative Adversarial Networks", IEEE,
ICPR, Beijing, China, Aug 20-24, 2018.