臺灣博碩士論文加值系統

隨著物聯網技術的快速發展，物聯網提供了生活便利性，如何保護物聯網裝置、避免駭客竊取物聯網資料已成為一大挑戰。本研究針對物聯網環境下的WebSocket通訊安全問題，提出了一種基於深度學習的惡意封包檢測方法。本研究一開始使用GNS3虛擬化平台構建了一個物聯網WebSocket虛擬網路拓樸，並收集了大量的物聯網WebSocket數據集樣本。並且利用卷積神經網路（CNN）和循環神經網路（RNN）兩種不同的深度學習模型以及使用Adam和Adadelta兩種不同的優化器，訓練了多個二元分類模型，用於識別物聯網WebSocket流量是否為惡意類型。為了適應物聯網設備的硬體和提升物聯網執行的效率，本研究將訓練好的模型轉換為輕量且高效的tflite版本，並且分別部署至Raspberry pi 4、Raspberry pi 5進行惡意封包的預測。通過不同的模型、優化器的以及物聯網設備組合，在準確率以及執行效率當中等關鍵指標的表現，得出CNN Adam tflite模型在本研究場景下具有卓越的惡意封包檢測能力。本研究為物聯網領域提供了一種新穎的安全防護解決方案，並且能夠自動辨別出物聯網WebSocket惡意封包的行為，提高物聯網裝置的惡意防禦能力，降低物聯網裝置被入侵的風險，對於構建安全可靠的物聯網生態具有重要意義，並且可將其應用於實際的物聯網場景中。

With the rapid development of Internet of Things (IoT) technology, IoT provides convenience in our daily lives. However, protecting IoT devices and preventing hackers from stealing IoT data has become a significant challenge. This research addresses the security issue of WebSocket communication in the IoT environment and proposes a deep learning-based malicious packet detection method. At the beginning, this research employed the GNS3 virtualization platform to construct a virtual IoT WebSocket network topology and collected a large dataset of IoT WebSocket samples. Furthermore, two different deep learning models, Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN), were utilized in conjunction with two distinct optimizers, Adam and Adadelta, to train multiple binary classification models for identifying whether the IoT WebSocket traffic is malicious or not. To adapt to the hardware constraints and improve the execution efficiency of IoT devices, the trained models were converted into a lightweight and highly efficient tflite version, which was then deployed on Raspberry Pi 4 and Raspberry Pi 5 for malicious packet prediction. By evaluating the performance of different model, optimizer, and IoT device combinations in terms of accuracy and execution efficiency, among other key indicators, the CNN Adam tflite model demonstrated exceptional malicious packet detection capability in the context of this research. This study provides a novel security solution for the IoT. Our solution automatically identify malicious behaviors in IoT WebSocket packets, enhancing the malicious defense capabilities of IoT devices and reducing the risk of IoT device intrusion. The contribution of this research is proposing a secure and reliable AI model to identify potentially harmful packets and thereby increasing the robustness of IoT ecosystem.

第一章 緒論 1
第一節 研究背景 1
第二節 研究動機 2
第三節 研究目的 3
第四節 研究流程 4
第二章 文獻探討 5
第一節 物聯網 5
第二節 資訊安全 12
第三節 深度學習 17
第三章 研究方法 20
第一節 實驗流程 20
第二節 虛擬網路拓樸 21
第三節 相關工具 22
第四章 研究結果 25
第一節 物聯網網路拓樸建置 25
第二節 資料前處理 38
第三節 深度學習 43
第四節 比較差異 55
第五章 結論 59
第一節 研究結論 59
第二節 未來研究建議 59
參考文獻 61

[1] statists, "Number of Internet of Things (IoT) connected devices worldwide from 2019 to 2023," statists, 27 7 2023. [Online]. Available: https://www.statista.com/statistics/1183457/iot-connected-devices-worldwide/.
[2] CTIMES, “七大要點！強化物聯網設備安全驗證,” [線上]. Available: https://www.ctimes.com.tw/DispArt/tw/%E7%89%A9%E8%81%AF%E7%B6%B2%E8%A8%AD%E5%82%99/%E5%AE%89%E5%85%A8%E9%A9%97%E8%AD%89/221024144547.shtml.
[3] F. Kuan, “物聯網實施指南,” MOKO SMART, 13 01 2022. [線上]. Available: https://www.mokosmart.com/zh-tw/iot-implementation-guide/.
[4] 陳曉莉, “物聯智慧 Kalay 平臺含有重大漏洞，將允許駭客自遠端存取 IoT 裝置。,” ithome, 18 8 2021. [線上]. Available: https://www.ithome.com.tw/news/146258.
[5] Cve, "CVE-2021-28372," Cve, 13 03 2021. [Online]. Available: https://cve.mitre.org/cgi-bin/cvename.cgi?name=CVE-2021-28372.
[6] Faisal Hussain , Syed Ghazanfar Abbas , Ghalib A. Shah , Ivan Miguel Pires, Ubaid U. Fayyaz ,Farrukh Shahzad , Nuno M. Garcia and Eftim Zdravevski, "A Framework for Malicious Traffic Detection in IoT Healthcare Environment," 26 4 2021. [Online]. Available: https://www.mdpi.com/1424-8220/21/9/3025.
[7] A. S. Gillis, "internet of things (IoT)," TechTarget, 8 2023. [Online]. Available: https://www.techtarget.com/iotagenda/definition/Internet-of-Things-IoT.
[8] K. Söderby, "Getting Started with Arduino," Arduino, 05 01 2024. [Online]. Available: https://docs.arduino.cc/learn/starting-guide/getting-started-arduino/.
[9] Arduino, "What is Arduino?," Arduino, 25 01 2022. [Online]. Available: https://docs.arduino.cc/learn/starting-guide/whats-arduino/.
[10] E. Inc, "Espressif Announces the Launch of ESP32 Cloud on Chip and Funding by Fosun Group," Espressif Inc, 7 9 2016. [Online]. Available: https://web.archive.org/web/20170401061525/http://espressif.com/en/media_overview/news/espressif-announces-launch-esp32-cloud-chip-and-funding-fosun-group.
[11] 乐鑫科技, “芯片概览｜乐鑫科技,” 乐鑫科技, [線上]. Available:
https://www.espressif.com/zh-hans/products/socs.
[12] BBC, "BBC and partners unveil the landmark BBC micro:bit," BBC, 6 7 2015. [Online]. Available: https://www.bbc.co.uk/mediacentre/mediapacks/microbit.
[13] R. P. Foundation, "Welcome to Raspbian," Raspberry Pi Foundation, [Online]. Available: https://www.raspbian.org/.
[14] 台灣智能感測科技, “Home Assistant 是一個開源的智慧家庭平台,” 台灣智能感測科技, [線上]. Available: https://www.taiwansensor.com.tw/62334-2/.
[15] DrewAHendricks, "The Trouble with the Internet of Things," London Datastore, 10 8 2015. [Online]. Available: https://data.london.gov.uk/blog/the-trouble-with-the-internet-of-things/.
[16] Microsoft, “什麼是資訊安全 (資安)？,” microsoft, [線上]. Available: https://www.microsoft.com/zh-tw/security/business/security-101/what-is-information-security-infosec.
[17] E. B. K. B. R. G. Andrew Banks, "MQTT Version 5.0," mqtt-v5.0, 7 3 2019. [Online]. Available: https://docs.oasis-open.org/mqtt/mqtt/v5.0/mqtt-v5.0.html.
[18] AWS, “什麼是 MQTT？ – MQTT 協定介紹 – AWS,” AWS, [線上]. Available: https://aws.amazon.com/tw/what-is/mqtt/.
[19] "The WebSocket Protocol," [Online]. Available: https://www.rfc-editor.org/rfc/rfc6455.html.
[20] Google, "Colaboratory," Googlehttps://research.google.com/colaboratory/faq.html, [Online]. Available: https://research.google.com/colaboratory/faq.html.
[21] J. C. Neumann, "The book of GNS3 : build virtual network labs using Cisco, Juniper, and more," 2015. [Online]. Available: https://archive.org/details/thebookofgns3_201908.
[22] GNS3, "Getting Started with GNS3," GNS3, [Online]. Available: https://docs.gns3.com/docs/.
[23] E. Upton, "Ten millionth Raspberry Pi, and a new kit," 8 9 2016. [Online]. Available: https://www.raspberrypi.com/news/ten-millionth-raspberry-pi-new-kit/.
[24] "Docker," 29 1 2021. [Online]. Available: https://www.tensorflow.org/install/docker.
[25] NVIDIA, "Installing the NVIDIA Container Toolkit," NVIDIA, [Online].
63
Available: https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/latest/install-guide.html.
[26] Microsoft, "Developing inside a Container using Visual Studio Code Remote Development," Microsoft, 05 06 2024. [Online]. Available: https://code.visualstudio.com/docs/devcontainers/containers.
[27] fdeloche, "Compressed (left) and unfolded (right) basic recurrent neural network," 19 6 2017. [Online]. Available: https://en.wikipedia.org/wiki/Recurrent_neural_network#/media/File:Recurrent_neural_network_unfold.svg.
[28] Offnfopt, 12 April 2015. [Online]. Available: https://upload.wikimedia.org/wikipedia/commons/0/00/Multi-Layer_Neural_Network-Vector-Blank.svg.