PhD thesis defense to be held on July 8, 2024, at 12:00 (Conference Room, New ECE building)

Picture Credit: George Routis

Thesis title: Knowledge Extraction and Security in Internet of Things Systems

Abstract: It is more than obvious that the Internet is not the same as it was a few decades ago. It has evolved and brought new results and factors that have changed it and made it compatible with the current needs. We have witnessed a revolution and the birth of technologies like the Internet of Things, widely known as IoT nowadays. The traditional Internet has penetrated many areas of our everyday life and we no longer follow the initial paradigm, where the user turns on a desktop or laptop PC and connects to the internet. The idea now is that “things” such as Single Board Computers (SBCs) or low-cost processing modules are being used in areas such as Smart Agriculture or Vehicles and via wireless (2G, 3G, 4G, 5G, Wi-Fi (Wireless Fidelity), LoRa (Long Range), Zigbee, Sigfox, Bluetooth, Satellite...) and stable connections (LAN – Local Area Network, optic fiber, …) the user can interact with sensors and actuators, in order to observe, data log measurements, and act accordingly. The SBCs and the low-cost processing modules are equipped with sensors, so they are ideal for many areas where there is need for sensing and datalogging over one or more connections. However, the usage of IoT involves so much more than just datalogging via sensors. Modules can work independently of user supervising. They can operate autonomously via the use of programs that have been stored on them. For instance, in the Internet of Vehicles (IoV) which is a sub-category of IoT, the nodes (vehicles) obey to the IoT rationale. Furthermore, they execute more complex programs, implement changing of their position, and analyze the environment for the safety of the users that are inside the vehicles. IoT can interface with Clouds and various Internet Services, so a user who lives in Germany can control the smart Energy Meter which is placed in a house in Greece.

With the evolution of Machine Learning (ML) it is feasible to execute ML code in SBCs and provide more processing power to the IoT devices. It is not uncommon to attach a camera to a low-cost processing unit and via the use of a trained image ML model to inference live image processing targeting needs such as pest control in a farm field or soil salinity analysis and leaf disease analysis. Unmanned aerial vehicles (UAVs) are connected wirelessly with base stations and can capture images from a farm field in order to identify essential problems in farm field.

This thesis documents our work on real world Internet of Things applications in different areas. First, we analysed the rationale and various details of IoT and Machine Learning in precision agriculture. A scheme was implemented in order to sense and evaluate different factors in a laboratory experiment. We sensed and logged temperature, Ultra Violet (UV) radiance, soil moisture and air humidity. Through the use of a sophisticated Recurrent Neural Network - Long Short Term Memory (RNN-LSTM), we were able to forecast weather conditions, so the user could identify when there was need to irrigate the plant or farm field (in cases of scaling up). This way the user could save water resources and money by avoiding unnecessary/excess irrigation. Energy is a valuable/scarce resource in farms, therefore we also proceeded to an analysis of different IoT modules and the related wireless systems, in order to identify ways to optimize energy consumption.

Secondly, we performed experiments within the realm of the Internet of Vehicles (IoV), where security is a critical factor. More precisely, we simulated an IoV network of vehicles in an ns-3 simulator, where different asymmetric cryptographic protocols Number Theory Research Unit (NTRU), Elliptic Curve Cryptography (ECC), Hyper Elliptic Curve Cryptography – genus 2 (HECC-g2), Hyper Elliptic Curve Cryptography – genus 3 (HECC-g3), Rivest Shamir Adleman (RSA) were analysed. We observed metrics of encryption/decryption times, message sizes, signature generation times, signature verification times, exchange handshake sizes, and pseudonym exchange times, while we also examined how the energy of the nodes (vehicles) was affected when executing each asymmetric protocol.

Thirdly, we elaborated on the effects of Machine Learning models, and more precisely how the Convolutional Neural Network (CNN) model behaves when executed in different processing architectures. We used 3 SBCs that incorporated different processing units: Central Processing Unit (CPU), Graphics Processing Unit (GPU) Tensor Processing Unit (TPU) used in the inference part on image analysis related to leaves’ diseases. Our research focused mainly on CPU-, Random Access Memory (RAM)-, and swap memory usage, as well as temperature and energy consumption.

Fourthly, we experimented extensively with Arduino IoT modules in rice and maize farms, in cooperation with Machine Learning and more specifically CNNs and RNN-LSTMs. Linear Regression and Multiple Regression were used for farm metrics' analysis, especially in rice fields farms. There is also a pioneer device analyzed towards resin and rubber collection presented based on Arduino microcontroller and various sensors, that transmits to the end user information about the environmental conditions of the resin/rubber collection via GSM/GPRS or via Xbee Zigbee.

Lastly, a pioneer mailbox for hardcopy letters, was invented, with the ability to inform the user via Short Message/Messaging Service (SMS) messages if a letter is received. It uses an InfraRed sensor which senses the reception of a new letter in order to identify when there is a new letter inside the mailbox. It also incorporates a Liquid Crystal Display (LCD) screen, and keypad in order to control some functions such as the end user’s mobile number, the current consumption and the GSM/GPRS (Global System for Mobile Communications/General Packet Radio Service) signal strength. The user can also check the life of the battery. The device has been patented in the Hellenic Industrial Property Organisation (“OBI” in greek).

Supervisor: Professor Ioanna Roussaki

PhD Student: George Routis