在現代農業中，仰賴人工澆水會增加生產成本且耗費人力，使用溫室噴霧系統可以解決前述問題，卻會造成葉面浸潤，進而增加植物的疾病發生率。並且兩種方法均無法針對個別植株，控制其所需的澆水量以及澆水方向。

本研究研發一新型多致動龍門型澆水機器人，可應用於蝴蝶蘭植株栽培溫室中。首先藉由應用YOLOv4-tiny物件辨識技術，偵測植株位置，再經由計算圖像回歸直線以估測植株姿態。並且利用變形路徑演算法進行路徑規劃，以降低姿態估測的不確定性。最後透過控制龍門上的線性致動器，使致動器沿著規劃路徑移動，藉以避開葉面區域。若實際在致動器上安裝澆水器，而非實驗時使用雷射作為驗證，可望減少澆水時葉面上積水，降低植株病害的可能性。

本研究建構於機器人系統（ROS）上，並且透過Nvidia Xavier實現演算法。最終，與過往的研究比較，本研究利用雷射模擬澆水路徑，實現了82%的澆水成功率，根據使用者設定速度，可以達到每秒3個植株以上的速度進行精準澆水。

For watering plants, manual watering requires human resources and thus increase cost, while the overhead irrigation system leads to infiltration on the leaf surface, resulting in more unnecessary incidence rate of plants that may cause diseases on the plant. Both methods above cannot be used to precisely control the watering amount and watering direction on each plant.

This research develops a new type of gantry robot to water Phalaenopsis orchid seedlings in the greenhouse. The neural network model YOLOv4-tiny is employed to locate the orchid seedlings and to regress the cropped image to estimate the poses of the seedlings. Furthermore, the reactive deformation trajectory planning algorithm is utilized to generate the sprayer trajectory from the estimated poses of seedlings by YOLOv4-tiny and to reduce the uncertainty of pose estimation. Finally, multiple linear actuators are guided along the planned paths that is supposed to water the soil portion, rather than the leaf area of plants, if the water sprayer is built, instead of using laser beam in this work, in the future. Due to less watering in the leave, the diseases of seedlings can be reduced.

The gantry robot is implemented on the robot operating system (ROS) that is run on an Nvidia Xavier computer. Experiments are conducted and the results are compared to the earlier works, using laser beam to substitute the water sprayer, the proposed system shows that the defined watering success rate can reach 82% and the defined watering speed is at least three seedlings per second.

摘要 i
Abstract ii
Acknowledgements

Chapter 1 Introduction 1
1.1 Overview 1
1.2 Background and Motivation 3
1.3 Literature Review 6
1.3.1 Related Works 6
1.3.2 Related Methods 10
1.4 Thesis Structure 12

Chapter 2 Methodology 13
2.1 Seedling Pose Estimation 15
2.1.1 YOLOv4­-tiny and Orchid Seedlings Image Dataset 18
2.1.2 RGB Image Regression 20
2.1.3 Pose Similarity 33
2.1.4 Position Fusion 35
2.1.5 Coordinate Transformation 35
2.2 Reactive Deformation Trajectory Planning Algorithm 37

Chapter 3 Implementation 45
3.1 Hardware Design 45
3.2 Software Implementation 47
3.3 Actuator Control 50

Chapter 4 Experimental Results and Discussions 55
4.1 YOLOv4­-tiny Performance 55
4.2 Different methods of determining the seedling poses 57
4.3 Performance 68
4.3.1 Success watering percentage 71
4.3.2 Watering speed 71

Chapter 5 Conclusion & Future Work 73
5.1 Conclusion 73
5.2 Future Work 74

References 75
A Hardware List and Technical Specification 79
B Three­-view of the prototype XsY Gantry robot 80

[1] 行政院農委會, “農業就業人口統計.” https://agrstat.coa.gov.tw/sdweb/public/inquiry/InquireAdvance.aspx, accessed 2021­08­26.
[2] 黃 柏 軒, “以科技突破台灣農業挑戰今周刊,” May 2018. https://www.businesstoday.com.tw/article/category/80394/post/201805180035, accessed 2021­08­26.
[3] 行政院農委會, “單一農產品進出口量值 ─ 按國家別.” https://agrstat.coa.gov.tw/sdweb/public/trade/tradereport.aspx, accessed2021­08­26.
[4] A. Henn, “The plant doctor ­ watering and plant disease,” Oct 2016. http://extension.msstate.edu/publications/information-sheets/the-plant-doctor-watering-and-plant-disease, accessed 2021­08­26.
[5] C. Hong and G. Moorman, “Plant pathogens in irrigation water: Chal­ lenges and opportunities,” Critical Reviews in Plant Sciences ­CRIT REV PLANT SCI, vol. 24, pp. 189–208, 05 2005.
[6] J. S. Chang, “Automated counting system for orchid seedlings based on deep learning,” Master’s thesis, National Tsing Hua University, 2020.
[7] R. L. Aronson, “Humanity’s open­source automated precision farming machine,” in Creative Commons Attribution 4.0 International License, California, Aronson, R Landon, 2013.
[8] R. Jauhari, A. N. Jati, and F. Azmi, “Mechanical design of cnc for general farming automation,” in 2017 5th International Conference on Instrumen­ tation, Control, and Automation (ICA), pp. 47–50, 2017.
[9] A. Athukorala, N. Ranasinghe, K. Herath, P. Jayasekara, and T. D. Lalitharatne, “Scalable autonomous agronomical smartbot,” in 2018 14th IEEE/ASME International Conference on Mechatronic and Embedded Systems and Applications (MESA), pp. 1–6, 2018.
[10] M. Jawad, N. Saleem, S. A. Ahmed, M. B. Qureshi, S. M. Ali, Z. Ul­ lah, B. Khan, C. A. Mehmood, R. Asghar, and U. Fayyaz, “Design of a self­sustained farming system (sfs) for pakistan,” in 2019 International Conference on Engineering and Emerging Technologies (ICEET), pp. 1– 6, 2019.
[11] M. Erick, S. Fiestas, R. Sixto, and G. Prado, “Modeling and simulation of kinematics and trajectory planning of a farmbot cartesian robot,” in 2018 IEEE XXV International Conference on Electronics, Electrical Engineer­ ing and Computing (INTERCON), pp. 1–4, 2018.
[12] C. J. Choque Moscoso, E. M. Fiestas Sorogastúa, and R. S. Prado Gardini, “Efficient implementation of a cartesian farmbot robot for agricultural applications in the region la libertad­peru,” in 2018 IEEE ANDESCON, pp. 1–6, 2018.
[13] C. J. Choque M, P. E. Linares O, N. F. Alcorta S, J. L. Alva A, and S. R. Prado G., “Iot automated orchard in a domestic environment,” in 2019 IEEE 1st Sustainable Cities Latin America Conference (SCLA), pp. 1–6, 2019.
[14] K. Ramesh, K. T. Prajwal, C. Roopini, M. Gowda M.H., and V. V. S. N. S. Gupta, “Design and development of an agri­bot for automatic seed­ ing and watering applications,” in 2020 2nd International Conference on Innovative Mechanisms for Industry Applications (ICIMIA), pp. 686–691, 2020.
[15] N. Putu Devira Ayu Martini, N. Tamami, and A. Husein Alasiry, “Design and development of automatic plant robots with scheduling system,” in 2020 International Electronics Symposium (IES), pp. 302–307, 2020.
[16] S. L. C. Lian, R. Zahari, and T. H. Lim, “Precision analysis of bresenhams algorithm on low­power intelligent seeding system,” in 2020 International Conference on Artificial Intelligence in Information and Communication (ICAIIC), pp. 231–236, 2020.
[17] M. D. I. Sujon, R. Nasir, M. M. I. Habib, M. I. Nomaan, J. Baidya, and M. R. Islam, “Agribot: Arduino controlled autonomous multi­purpose farm machinery robot for small to medium scale cultivation,” in 2018 International Conference on Intelligent Autonomous Systems (ICoIAS), pp. 155–159, 2018.
[18] T.­W. Chang, W.­C. Wang, and R. Chen, “Intelligent control system to irrigate orchids based on visual recognition and 3d positioning,” Applied Sciences, vol. 11, no. 10, 2021.
[19] Y. Gal and Z. Ghahramani, “Dropout as a bayesian approximation: Rep­ resenting model uncertainty in deep learning,” in international conference on machine learning, pp. 1050–1059, PMLR, 2016.
[20] Y. Xiang, T. Schmidt, V. Narayanan, and D. Fox, “Posecnn: A convolu­ tional neural network for 6d object pose estimation in cluttered scenes,” arXiv preprint arXiv:1711.00199, 2017.
[21] B. Okorn, M. Xu, M. Hebert, and D. Held, “Learning orientation distribu­ tions for object pose estimation,” in 2020 IEEE/RSJ International Confer­ ence on Intelligent Robots and Systems (IROS), pp. 10580–10587, IEEE, 2020.
[22] A. Bochkovskiy, C.­Y. Wang, and H.­Y. M. Liao, “Yolov4: Optimal speed and accuracy of object detection,” arXiv preprint arXiv:2004.10934, 2020.
[23] T. Anand, “Reactive deformation of path for navigation among dynamic obstacles,” in 2009 IEEE International Conference on Robotics and Biomimetics (ROBIO), pp. 544–549, 2009.
[24] J. Redmon, “Darknet: Open source neural networks in c.” http://pjreddie.com/darknet/, 2013–2016.
[25] E. W. Weisstein, “Least squares fitting,” https://mathworld.wolfram.com/, 2002.
[26] S. G. A. F., Least Squares Estimation, p. 35–38. John Wiley &; Sons, 2003.
[27] S. G. A. F., Weighted Least Squares for the Straight Line, p. 150–151. John Wiley &; Sons, 2003.