DR. Fereshteh Nayyeri

Project Title: Using AI and cameras to identify and monitor litter

The project aimed to address the pressing issue of marine debris by leveraging deep learning techniques to estimate the volume of floating litter in waterways and identify key hotspots. By evaluating existing management systems, the goal was to contribute to the preservation of healthy water ecosystems.

Key Achievements:

• Managed the extensive image dataset captured by trail cameras strategically positioned across various waterways in Sydney, Australia.
• Created a comprehensive catalogue detailing different types of litter observed in the waterways, categorizing them into distinct classes for further analysis and classification.
• Developed ground truth annotations for a large image dataset. This process involved rigorous verification and validation to ensure accuracy and reliability.
• Implemented and train state-of-the-art artificial intelligence models to detect and classify marine litter types effectively.
• Contributed to the project's overarching goal by providing insights and recommendations based on findings from the analysis, ultimately aiming to enhance existing management systems and promote healthier waterway environments.
• Women in AI Australia: Keynote Speaker
• Ending Plastic Waste Symposium: Presentation
• News: 7News Australia

Project Title : Development of Local Machine Web Application for Image Analysis and Visualization.

As part of my role, I spearheaded the creation of a robust web-based application designed to operate exclusively on local machines. Leveraging Flask, I engineered a comprehensive solution for visualizing images, running pre-trained machine learning models for image detection and classification, and presenting data analysis results through intuitive visual interfaces.

Key Achievements:

• Developed a user-friendly web application using Flask, tailored to meet the specific requirements of local machine deployment.
• Integrated pre-trained machine learning models into the application, enabling users to analyze images for detection and classification purposes.
• Implemented functionalities to visualize detection results, providing users with insightful representations of analyzed data.
• Incorporated features to locate images on Google Maps based on metadata such as latitude and longitude, enhancing the spatial understanding of image capture locations.
• Conducted thorough testing and debugging to ensure the application's stability and reliability, addressing any issues promptly to deliver a robust and error-free user experience.
• Collaborated closely with stakeholders to gather requirements and incorporate feedback, ensuring the application met their expectations and provided value in facilitating image analysis and visualization tasks on local machines.

Project Title : Glassdoor Employee Sentiment Analysis.

I collaborated with the Department of Business Strategy and Innovation at Griffith Business School on a funded research project analysing changes in employee work preferences during the COVID-19 period using large-scale review data.

Key Achievements:

• Built a Python tool to scrape and process 14,200+ Glassdoor reviews, extracting metadata and sentiment indicators across roles, ratings, and narratives.
• Analysed temporal trends across pre-pandemic, pandemic, and post-pandemic phases to uncover shifts in employee priorities and workplace perceptions.
• Demonstrated ability to use social media data for organisational research and digital labor market insights.
• Contributed to a peer-reviewed journal publication: link
• More info about the analysis: link

Project Title: Predicting Koala Road Crossing Behaviours using AI-Powered Observation Network

As a casual researcher, I contributed to a pilot study focused on training artificial intelligence (AI) systems for koala "face recognition" at crossing locations throughout South East Queensland. The project aimed to leverage facial recognition technology to enhance koala conservation efforts and improve safety measures at these critical crossings.

Key Achievements:

• Conducted research as part of the pilot study, investigating methodologies for training AI algorithms to recognize and classify koalas based on facial features.
• Implemented a methodology centered around artificial neural networks and deep learning algorithms, utilizing the Python programming language and TensorFlow framework. This facilitated the detection and classification of koalas, enabling the application of facial expression analysis techniques to understand koala behavior and usage patterns at crossing locations.
• Collaborated with interdisciplinary teams to ensure the integration of AI technologies with conservation efforts, contributing to the development of research-based strategies for koala conservation and habitat protection.
• Contributed insights and findings to support the project's objectives, including potential applications of AI-driven facial recognition in wildlife monitoring and conservation management practices.
• News: LinkedIn , ABS , DW News

Project title: An Automated System for the Analysis of Road Safety and Conditions

This ARC Linkage project was a collaboration between CQUinversity and DTMR (Department of Transport and Main Roads) in Queensland, Australia. The main aim of this project is to develop an automatic system for the detection of road safety attributes and distances to improve road infrastructure and reduce fatalities on the roads. The deep learning based technique is developed and evaluated using the digital video road data which is collected over every state road in Queensland annually. The system finnaly tested on state-controlled roads in Queensland to assess road safety and conditions automatically. The major challenges of current manual systems are to accurately detect, segment and classify all road objects and also calculate the distance between objects. Deep learning has the ability to address such major challenges.

Key Achievements:

• Established and maintained a comprehensive database comprising road attributes and signage extracted from DTMR videos, meticulously annotating the collected data.
• Implemented methodology based on Artificial Neural Network and Deep Learning algorithms with Python programming language and Tensorflow framework for detection and classification of road speed limit signs.
• Executed experiments leveraging deep learning architectures on High-Performance Computing (HPC) systems to analyze vast datasets effectively.

Project Title :Foreground-Background Classification for Crack Detection

Key Achievements:

• Developed and programmed the innovative deep learning architecture using Python and TensorFlow programming languages, ensuring seamless integration with existing frameworks.
• Conducted extensive experimentation and optimization to refine the architecture's performance, achieving significant advancements in crack detection accuracy and efficiency.
• Collaborated with fellow researchers to analyze and interpret results, contributing insights crucial for the project's success and further advancements in the field.
• Published findings in reputable academic journals and presented results at conferences, garnering recognition and acclaim within the research community for innovative contributions to dimensionality reduction in semantic segmentation for crack detection.
• Principal Supervisor
• Associate Supervisor
• Conferral, page: 39

Project Title : Designing and developing the algorithm to correct respiratory motion from PET/CT lung cancer images

As a Graduate Research Assistant within the Biomedical Imaging & Signal Processing research group at the School of Electrical and Electronic Engineering, I played an integral role in advancing the field of biomedical imaging. My work primarily focused on exploring innovative techniques for image reconstruction and addressing challenges related to respiratory motion correction in PET/CT imaging systems.

Key Achievements:

• Conducted in-depth research into biomedical imaging techniques, with a particular emphasis on image reconstruction and respiratory motion correction.
• Identified technological limitations and deficiencies within existing PET/CT imaging systems, processes, and methodologies, contributing valuable insights towards enhancing imaging capabilities.
• Designed and developed sophisticated algorithms for biomedical image reconstruction, leveraging cutting-edge computational techniques to improve image quality and accuracy.
• Implemented and programmed various biomedical image reconstruction, processing, and analysis techniques using the MATLAB programming language, ensuring seamless integration with existing systems.
• Conducted rigorous testing, debugging, and diagnosis of errors and faults within the applications, guaranteeing optimal performance and adherence to specifications. Through meticulous attention to detail, I facilitated the refinement of algorithms and processes for enhanced biomedical imaging outcomes.

Project Title: Image matching using dimensionally reduced embedded Earth Mover's Distance

• Thesis as a book.
• Supervisor