臺灣博碩士論文加值系統

本論文針對可資源回收的瓶罐，透過嵌入式系統搭配深度學習影像辨識模型和機械手臂，實作一套可攜式瓶罐(含寶特瓶、易開罐、鋁箔包)挑選系統，並使用電動旋轉台模擬輸送帶的動態環境。影像辦識模型則採用了AlexNet和SSD深度神經網路為基礎，並加以修改為適合嵌入式系統使用的輕量化神經網路模型，名為AlexNet-SSD。
在網路模型訓練上，自行拍攝和收集了共約1200張相關瓶罐照片的資料集，隨機選擇1000張為訓練集而另外200張為測試集。最後將在桌上型電腦主機訓練完成的AlexNet-SSD神經網路模型，移植到嵌入式系統上進行測試。辨識的速度可高達26.8 FPS且凖確率為87.3%。

As people's awareness of environmental protection increases, improving resource reuse is a way to effectively reduce environmental pollution and achieve sustainable use of resources. But waste disposal is a complex process that requires a complex checking process before it enters the incineration. In addition, a large amount of human resources is required to pre-classify waste. The pre-classifying process greatly increases cost, reduces the efficiency, and even directly leads to abandon the classification action and directly incinerates.
In this research, we propose an automatic system for picking up bottles based on an embedded system with a deep learning model and a robotic arm. In order to make the system practical, the research used a rotating table to simulate a conveyor belt, and combined with a camera and image recognition. The image recognition model is based on the neural network architecture composed of AlexNet and SSD, and it is modified to be a lightweight neural network model for embedded system called AlexNet-SSD.
In the model training phase, nearly 1200 images of PET bottles, can, and tetra-pak were collected. Among which 1000 pieces were used as the training set and 200 pieces as the test set. Finally, the training results on the host are ported to the embedded system for testing. The testing results are 1) recognition speed can be as high as 26.8 frames per second, and 2) an accuracy of 87.3%.

中文摘要 I
Extended Abstract II
誌謝 VIII
章節目錄 IX
表目錄 XI
圖目錄 XII
第一章 緒論 1
1-1 研究背景與動機 1
1-2 研究目的 1
1-3 研究貢獻 2
1-4 論文架構 3
第二章 文獻探討 4
2-1 台灣廢棄物種類認知 4
2-2 廢棄物辨識相關研究 5
2-3 各國自動化廢棄物分類技術 12
2-4 自動化分類技術相關比較 14
第三章 嵌入式系統實現瓶罐挑選 15
3-1 實驗場景介紹 15
3-2 硬體設備 17
3-3 系統架構 21
3-4 影像辨識演算法說明 24
3-5 機械手臂和坐標定位說明 33
第四章 實作成果與討論 37
4-1 AlexNet-SSD神經網路模型訓練結果 37
4-2 瓶罐影像辨識實驗結果 38
4-3 機械手臂吸取實驗結果 42
4-4 討論 43
第五章 結論與未來工作 46
5-1 結論 46
5-2 未來工作 47
參考文獻 48
附錄 52
附錄1 uArm Swift SPEC 52
附錄2 系統程式碼 52

[1]A. Krizhevsky, I. Sutskever and G. E. Hinton, Imagenet Classification with Deep Convolutional Neural Networks, Proceedings of the Advances in Neural Information Processing Systems 25, January 2012
[2]AMP Robotics, Transforming Recycling, https://www.amprobotics.com/, last retrieve 14 October 2018
[3]B. M. Chinnathurai and R. Sivakumar Design and Implementation of a Semi-Autonomous Waste Segregation Robot, Proceedings of the SoutheastCon 2016, 30 March-3 April 2016, Norfolk, VA, USA
[4]D. Yudin and D. Slavioglo, Usage of fully convolutional network with clustering for traffic light detection, Proceedings of the IEEE 2018 7th Mediterranean Conference on Embedded Computing (MECO), 10-14 June 2018, Budva, Montenegro
[5]INSIDE, 資源回收也用上了AI，讓更多可重複利用資源不再淹沒垃圾掩埋場！, https://www.inside.com.tw/article/10477-ai-use-on-recycle, posted 11 September 2017, last retrieve 14 October 2018
[6]J. Canny, A Computational Approach to Edge Detection, Proceedings of the IEEE Transactions on Pattern Analysis and Machine Intelligence (Volume: PAMI-8, Issue:6) , pp. 679-698, November 1986
[7]J. Redmon, S. Divvala, R. Girshick and A.Farhadi, You Only Look Once: Unified, Real-Time Object Detection, Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 27-30 June 2016, Las Vegas, NV, USA
[8]J. Bai, S. Lian and D. Liu, Deep Learning Based Robot for Automatically Picking up Garbage on the Grass, Proceedings of the IEEE Transactions on Consumer Electronics (Volume:64, Issue:3), pp. 382-389, 25 July 2018
[9]K. Simonyan and A. Zisserman, Very Deep Convolutional Networks for Large-Scale Image Recognition, Proceedings of the 2015 International Conference on Learning Representations, 10 April 2015, San Diego, California
[10]K. He, X. Zhang, S. Ren and J. Sun, Deep Residual Learning for Image Recognition, Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 27-30 June 2016, Las Vegas, NV, USA
[11]K. Rungsuptaweekoon、V. Visoottiviseth and R. Takano, Evaluating the power efficiency of deep learning inference on embedded GPU systems, Proceedings of the IEEE 2017 2nd International Conference on Information Technology (INCIT), 2-3 November 2017, Nakhonpathom, Thailand
[12]K. Kozai and M. Hashimoto, Determining Robot Grasping-Parameters by Estimating “Picking Risk, Proceedings of the IEEE 2018 International Workshop on Advanced Image Technology (IWAIT), 7-9 January 2018, Chiang Mai, Thailand
[13]L. Juan Carlos Briñez and A. Rengifo Automatic Waste Classification using Computer Vision as an Application in Colombian High Schools, Proceedings of the 6th Latin-American Conference on Networked and Electronic Media, 23-25 September 2015, Medellin, Colombia
[14]Logitech, Logitech C930e, https://www.logitech.com/zh-tw/product/c930e-webcam, last retrieve 18 October 2018
[15]NIVDIA, Jetson TX2, https://www.nvidia.com/zh-tw/autonomous-machines/embedded-systems-dev-kits-modules/, last retrieve 17 October 2018
[16]OpenCV, Contours : More Functions, https://docs.opencv.org/3.4/d5/d45/tutorial_py_contours_more_functions.html, last retrieve 15 October 2018
[17]OpenCV, Platforms, https://opencv.org/platforms/, last retrieve 15 October 2018
[18]P. Karuppiah, H. Metalia and K. George, Automation of a Wheelchair Mounted Robotic Arm using Computer Vision Interface, Proceedings of the 2018 IEEE International Instrumentation and Measurement Technology Conference (I2MTC), 14-17 May 2018, Houston, TX, USA
[19]Qt, developers, https://www.qt.io/developers/, last retrieve 17 October 2018
[20]SADAKO, Max-AI™, http://www.sadako.es/max-ai/, last retrieve 16 October 2018
[21]tanford Vision Lab, IMAGENET Large Scale Visual Recognition Challenge 2012(ILSVRC2012), http://image-net.org/challenges/LSVRC/2012/index, last retrieve 18 October 2018
[22]TOMRA, Recycling technology, https://www.tomra.com/en/sorting/recycling/recycling-technology, last retrieve 14 October 2018
[23]UFACTORY, Quick Start Guide, https://cdn.sparkfun.com/assets/3/d/3/a/8/uArm-Swift-Pro-Quick-Start-Guide.pdf, last retrieve 19 October 2018
[24]UFACTOR, uArm Swift, https://www.ufactory.cc/#/cn/uarmswift, last retrieve 19 October 2018
[25]UFACTORY, uArm Swift SPEC, https://cdn.sparkfun.com/assets/9/8/a/c/0/uArm-Swift-Specifications-en.pdf, last retrieve 19 October 2018
[26]V. Kumar, Q. Wang, W. Minghua, S. Rizwan, S. Shaikh and X. Liu, Computer Vision Based Object Grasping 6DoF Robotic Arm Using Picamera, Proceedings of the IEEE 2018 4th International Conference on Control, Automation and Robotics (ICCAR), 20-23 April 2018, Auckland, New Zealand
[27]W. Liu, D. Anguelov, D. Erhan, C. Szegedy and S.Reed, SSD: Single Shot MultiBox Detector, Proceedings of the Computer Vision and Pattern Recognition, 8 December 2015
[28]W. Setiawan, A. Wahyudin and G. R. Widianto The Use of Scale Invariant Feature Transform (SIFT) Algorithms To Identification Garbage Images Based on Product Label, Proceedings of the 3rd International Conference on Science in Information Technology (ICSITech), 25-26 October 2017, Bandung, Indonesia
[29]Wiki, G-code, https://en.wikipedia.org/wiki/G-code, last retrieve 19 October 2018
[30]Z. Chen and H. Zou A Vision-based Robotic Grasping System Using Deep Learning for Garbage Sorting, Proceedings of the 36th Chinese Control Conference, 26-28 July 2017, Dalian, China
[31]行政院環境保護署, 一般廢棄物管理政策, https://www.epa.gov.tw/Page/420AD8B3A3EABE03, last retrieve 12 October 2018
[32]莊炳義, 垃圾強制分類政策推動之評估規劃與管理研究, 環境規劃與管理研究所, 國立臺北科技大學, 2006
[33]陳熙文, 台灣垃圾知多少? https://theme.udn.com/theme/story/7490/2965909, posted 03 February 2018, last retrieve12 October 2018
[34]程式前沿, 大話CNN經典模型: AlexNet, https://codertw.com/%E7%A8%8B%E5%BC%8F%E8%AA%9E%E8%A8%80/568398/, posted 28 July 2018, last retrieve 18 October 2018
[35]劉志堅, 台灣垃圾處理現況, http://www.taiwanwatch.org.tw/issue/garbage/DIEOSEN/diesa010.htm, last retrieve 11 October 2018
[36]賴品瑀、彭瑞祥, 無采工！那些上了回收車 卻進焚化爐的塑膠垃圾, https://e-info.org.tw/node/211681, posted 22 July 2018, last retrieve 12 October 2018