Synthetic Image Generation for Training and Testing DNN-based Anomaly Detection Systems

Deep neural networks (DNNs) are increasingly employed in vision-based systems, extending well beyond self-driving vehicles. With the rise of the Internet of Things (IoT) and distributed vision sensing, vision-based systems have found diverse applications across various domains. These include detecting hazards on work sites and anomalies in critical infrastructure, utilizing, for example, video and image data captured by drones. In collaboration with SmartInside AI, this project aims to address the identification of equipment anomalies in the power grid using video and image data. The primary challenge that this project aims to tackle is the limited availability of high-quality labeled data for developing and testing DNN-based approaches. To overcome this challenge, we will capitalize on recent advancements in generative AI to create synthetic images. This approach will enable us to adapt SmartInside AI’s existing solution, originally developed using South Korean data, to suit the Canadian environment. Specifically, we will employ generative AI techniques to modify the existing DNN-based solution, accommodating the variations in equipment types and weather conditions between South Korea and Canada…

Faculty Supervisor:

Shiva Nejati;Mehrdad Sabetzadeh

Student:

Partner:

SMARTINSIDE AI, INC.

Discipline:

Computer science

Sector:

Professional, scientific and technical services

University:

University of Ottawa

Program: