臺灣博碩士論文加值系統

菇類的生長和栽培管理技術有很大的關聯，透過最佳化栽培管理技術以達到均勻生長及縮短栽培時間是非常重要之目標。基於深度學習之影像視覺技術在菇類栽培管理上有很大之應用範疇，可以將生長影像透過深度學習建立生長模型，用來預測袋栽杏鮑菇支出菇生長狀態，達到栽培管理之目的，更進一步可以達成提高生產品質及優化生產管理等效益。本研究以杏鮑菇為研究對象，利用光學、雷射等先進的影像設備生成三維點雲資料，並應用深度學習技術進行實例分割，實例分割評估結果平均精度(AP)為0.38與0.47，分割準確率為71%與86%。分割完後透過Open3D、Numpy等模組框取子實體邊框資訊。同時，我們開發了一個簡便易用的網頁操作系統，使用者只需上傳分割好的檔案，便能輕鬆觀察和讀取測量杏鮑菇子實體的邊框數據。此外，本實驗建立了一個生長模型，該模型可用於預測杏鮑菇的生長情況。該模型基於本研究收集的三維生長數據，能夠預測杏鮑菇在第九天到第十二天的生長狀態，使栽培者能夠更精確地計畫收穫時間。該模型的方程式為y = 2.4042x + 4.2292。透過本研究的方法，我們能夠快速且準確地獲取杏鮑菇子實體的表型資訊，為菇類的栽培提供了精準且具再現性的數據支持。這對於栽培者來說，有助於監測和評估杏鮑菇的生長情況，並能夠及時調整栽培環境和管理策略，從而提高生產效率並確保產品品質的一致性。

The growth of mushrooms is closely related to cultivation and management techniques. It is a very important goal to achieve uniform growth and shorten cultivation time through optimized cultivation and management techniques. Image vision technology based on deep learning has a wide range of applications in mushroom cultivation and management. Growth images can be used to build a growth model through deep learning, which can be used to predict the growth status of bag-grown king oyster mushrooms and achieve the purpose of cultivation management. , further can achieve benefits such as improving production quality and optimizing production management. This study takes Pleurotus eryngii as the research object, using advanced imaging equipment such as optics and lasers to generate 3D point cloud data, and applying deep learning technology for instance segmentation. The average precision (AP) of instance segmentation evaluation results is 0.38 and 0.47, and the segmentation The accuracy rates are 71% and 86%. After segmentation, obtain the border information of the child entities through modules such as Open3D and Numpy. At the same time, we have developed an easy-to-use web operating system. Users only need to upload the divided files to easily observe and read the frame data of the fruiting bodies of Pleurotus eryngii. In addition, this experiment established a growth model that can be used to predict the growth of Pleurotus eryngii. Based on the three-dimensional growth data collected in this study, the model is able to predict the growth status of Pleurotus eryngii from the ninth day to the twelfth day, allowing growers to more accurately plan harvest time. The equation for this model is y = 2.4042x + 4.2292. Through the method of this study, we can quickly and accurately obtain the phenotypic information of the fruiting bodies of Pleurotus eryngii, which provides accurate and reproducible data support for the cultivation of mushrooms. For growers, it is helpful to monitor and evaluate the growth of Pleurotus eryngii, and to adjust the cultivation environment and management strategies in time, thereby improving production efficiency and ensuring the consistency of product quality.

摘要...........................................................................i
Abstract....................................................................ii
誌謝.........................................................................iii
目錄.........................................................................iv
表目錄.....................................................................vi
圖目錄.....................................................................vii
第一章 緒論..............................................................1
1.1前言.....................................................................1
1.2研究背景..............................................................1
1.3研究動機與目的...................................................2
1.4論文架構..............................................................2
第二章 文獻回顧.......................................................3
2.1袋栽菇類栽培出菇與菇類智慧化生產..................3
2.2電腦視覺在菇類栽培之應用.................................5
2.2.1三維重建...........................................................6
2.2.2 三維資料型態...................................................7
2.2.3點雲特性...........................................................8
2.3實例分割技術與應用............................................8
2.3.1 PointNet........................................................10
2.3.2 PointNet++....................................................11
2.3.3 SGPN............................................................12
第三章 研究方法.....................................................13
3.1三維點雲取像裝置.............................................14
3.1.1動作流程........................................................14
3.1.2掃描設備........................................................15
3.1.3掃描軟體介面.................................................19
3.2點雲前處理與標註.............................................20
3.2.1訓練點雲標記.................................................20
3.2.2資料集分配.....................................................21
3.3深度學習............................................................22
3.3.1 SGPN訓練環境設置.......................................22
3.4邊框計算............................................................23
3.5 網頁UI介面設計................................................25
3.5.1 Flask介紹......................................................25
3.5.2 Three.js介紹..................................................26
3.6生長模型建立....................................................27
第四章 結果與討論.................................................30
4.1掃描點雲數據....................................................30
4.2訓練模型成果....................................................32
4.2.1模型分割結果.................................................32
4.3生長模型結果....................................................36
4.3.1生長模型點雲數據..........................................36
4.3.2實驗生長趨勢分析..........................................46
4.3.3杏鮑菇大小與數量變化..................................47
4.4使用者介面.......................................................49
4.4.1功能說明........................................................50
第五章 結論與未來展望.........................................52
5.1結論..................................................................52
5.2未來展望...........................................................53
參考文獻................................................................54
Extended Abstract.................................................57

[1]Schwab, K., 2017, The Fourth Industrial Revolution, First ed, New York.
[2]周榮源、石信德、黃文信，2020，袋栽菇類製包技術回顧與智慧化發展初探，
行政院農委會農業科技決策資訊平台。2022年10月17日，取自：
https://agritech-foresight.atri.org.tw/article/contents/3144
[3]陳美杏、呂昀陞、李瑋崧、彭金騰，2017，杏鮑菇栽培技術(一)。行政院農業委員會農業知識入口網。2023年6月29日，取自：
https://kmweb.coa.gov.tw/theme_data.php?theme=news&sub_theme=variety&id=54785
[4]陳美杏、李瑋崧、呂昀陞，2017，菇類基本特性與栽培原理(二)。行政院農業委員會農業知識入口網。2023年6月29日，取自：
https://kmweb.coa.gov.tw/theme_data.php?theme=news&sub_theme=variety&id=54729
[5]莊老達，2012，溫度及替代介質對杏鮑菇生長及發育之影響，國立中興大學園藝學系研究所博士論文，台中市。
[6]林明毅，2020，杏鮑菇自動化採收機構之研究，國立虎尾科技大學機械設計系碩士論文，雲林縣。
[7]顏靖峰，2022，袋栽杏鮑菇完株採收機構之研究，國立虎尾科技大學機械設計系碩士論文，雲林縣。
[8]邱弘丞，2021，自主移動式智慧物聯模組在菇類栽培之研究，國立虎尾科技大學機械設計系碩士論文，雲林縣。
[9]紀錦川，2023，應用於菇類栽培感測與資料擷取之 ROS自主移動機器人研發，國立虎尾科技大學機械設計碩士論文，雲林縣。
[10]Wikipedia.,2022,電腦視覺,Retrieved from https://zh.wikipedia.org/zh-tw/計算機視覺 (Nov 9, 2021).
[11]Duffy Benjamin F, Flynn Daniel R, 2017, A Year in Computer Vision, The M Tank,
Retrieved from https://www.themtank.org/a-year-in-computer-vision
[12]黃文信，2019，基於深度學習之杏鮑菇自動化分級系統之研究，國立虎尾科技大學機械設計系碩士論文，雲林縣。
[13]古丙駿，2020，杏鮑菇採收深度學習物件辨識系統之研究，國立虎尾科技大學機械設計系碩士論文，雲林縣。
[14]胡原愷，2021，袋栽菇類製包流程智慧檢測系統之研究，國立虎尾科技大學機械設計系碩士論文，雲林縣。
[15]劉宇珉，2020，應用高光譜影像之微分分析法判斷杏鮑菇新鮮程度，國立臺北科技大學碩士論文，台北市。
[16]徐祥睿，2022，杏鮑菇資料集之開發暨生長預測，國立陽明交通大學智慧計算與科技研究所碩士論文，台南市。
[17]D. Geronimo, et al,2010, IEEE transactions on Pattern Analysis and Machine Intelligence, 32 (7), 1239.
[18]B. Langmann,2014, Wide Area 2D/3D Imaging Development, Analysis and Applications , 1 st ed. Switzerland, SpringerLink.
[19]M. Weinmann,2016, Reconstruction and Analysis of 3D Scenes, 1 st ed. Switzerland, SpringerLink.
[20]Mescheder, L.M., Oechsle, M., Niemeyer, M., Nowozin, S., & Geiger, A, 2018, Occupancy Networks: Learning 3D Reconstruction in Function Space, IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 4455-4465.
[21]Park, J.J., Florence, P.R., Straub, J., Newcombe, R.A., & Lovegrove, S, 2019, DeepSDF: Learning Continuous Signed Distance Functions for Shape Representation, IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 165-174.
[22]常見三維表示方法。網路資料。2023年7月2日，取自：https://reurl.cc/r5EYEb
[23]Bello, S.A., Yu, S., & Wang, C, 2020, Review: deep learning on 3D point clouds. ArXiv, abs/2001.06280.
[24]Niccolò Cacciotti, 2022, Computer vision: the ultimate guide on the 4 main tasks,
Retrieved from https://www.smart-interaction.com/2022/07/14/computer-vision-the-ultimate-guide-on-the-4-main-tasks
[25]Guo, Y., Wang, H., Hu, Q., Liu, H., Liu, L., & Bennamoun (2019). Deep Learning for 3D Point Clouds: A Survey. IEEE Transactions on Pattern Analysis and Machine Intelligence, 43, 4338-4364.
[26]He, Y., Yu, H., Liu, X., Yang, Z., Sun, W., Wang, Y., Fu, Q., Zou, Y., & Mian, A.S. (2021). Deep Learning based 3D Segmentation: A Survey. ArXiv, abs/2103.05423.
[27]Qi, C., Su, H., Mo, K., & Guibas, L.J. (2016). PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation. 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 77-85.
[28]Qi, C., Yi, L., Su, H., & Guibas, L.J. (2017). PointNet++: Deep Hierarchical Feature Learning on Point Sets in a Metric Space. NIPS.
[29]Wang, W., Yu, R., Huang, Q., & Neumann, U. (2017). SGPN: Similarity Group Proposal Network for 3D Point Cloud Instance Segmentation. 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2569-2578.
[30]ASUS(2022).M640MB Datasheet.
[31]REVOPOINT(2022).POP 3D Scanner.Datasheet.
[32]CloudCompare(2022).Datasheet.
[33]Wikipedia.(2022).CloudCompare.Retrieved from
https://en.wikipedia.org/wiki/CloudCompare(July 14, 2022).
[34]Wang, Weiyue,2018,SGPN. Retrieved October, 2022, from
https://github.com/laughtervv/SGPN(April 15, 2022).
[35]Wikipedia.(2023).Flask. Retrieved from
https://zh.wikipedia.org/zh-tw/Flask(September 5, 2022).
[36]正郁農場。杏鮑菇簡介。檢自
http://www.focusline.com.tw/vegetable/summary.aspx (2022年1月20日)。