About

I am a Postdoctoral researcher in Deep Learning and Computer Vision at the University of Oxford, working with Prof Ben Sheldon. My research focuses on developing innovative Deep Learning models to advance feature representation learning, with an emphasis on salient feature extraction for segmenting complex and densely populated scenes. Additionally, I design architectures for multimodal deep learning and data fusion, combining diverse data sources — such as RGB, hyperspectral, and thermal imagery — to improve generalisation across various domains. My work bridges theoretical advancements in Deep Learning with practical applications in Remote Sensing, Environmental Monitoring, and Industrial Scene Analysis.

Prior to this, I completed an EPSRC-funded PhD at the University of Sussex, where I focused on Deep Learning for Remote Sensing systems. Under the supervision of Prof. Andy Philippides and Prof. Novi Quadrianto, my research involved developing novel Deep Learning models that extract salient feature representations for domain adaptation and semantic segmentation tasks.

As part of the Predictive Analytics Lab (PAL), I contributed to a British Academy funded project on Satellite/Aerial Image Scene Segmentation. I designed innovative Deep Learning models to classify land use from satellite/aerial imagery, with a focus on mapping peri-urban agriculture in Ghaziabad, India. The project aimed to integrate multiple data types into a web-based mapping and visualisation tool, with a goal of enhancing our understanding of land use patterns. Please see the demo below. demo Currently, I am developing an innovative semantic segmentation architecture that integrates a novel convolutional layer to enhance feature extraction and spatial context modelling on dense, complex aerial scenes. Furthermore, I propose a hybrid CNN-Transformer model that refines performance, demonstrating significant improvements over existing pure Transformer-based methods.

Additionally, I am working on a mid-level data fusion model for detecting cement plants. By integrating multi-temporal and multi-spectral satellite data, the model leverages Deep Learning to enhance industrial monitoring and improve detection in challenging environments.

I also developed a novel architecture for semantic segmentation in aerial images that extracts salient features and capturing broader context. This approach improves the detection of water bodies (lakes, rivers) from UAVs, outperforming current methods by being more resilient to shadows and effectively retrieving information occluded by canopy.

Furthermore, I investigated the Deep Learning robustness to domain shifts due to seasonal variations and proposed a novel architecture that extracts invariant feature representations between domains by combining salient texture feature extraction and a wider receptive field. This makes the model robust in the presence of domain shifts due to seasonal variations.

Part of my research also involves curating high-quality datasets tailored for classification, semantic segmentation, and instance segmentation tasks, contributing to the development of robust and scalable solutions across diverse domains.

During my research internship at Satellite Applications Catapult (under the supervision of Dr Cristian Rossi), I applied physics aware AI and Remote Sensing data to detect and classify cement plants in China by exploiting physical properties such as plant surrounding temperature and soil moisture.

Prior my PhD, I worked as a Control and Automation Design Engineer for the oil and gas industry, where I was exposed to the full software and hardware life cycle, systems sensing and control, communications, functional safety, project planning and management.

I am always up for new collaborations, drop me an email if you want to chat!

Dr Georgios Voulgaris

Recent News