Domain Adaptation

These works provide an overview of my research on domain shifts in aerial scenes due to seasonal variations between wet and dry seasons in the Global South. The aim of this work is to perform an aerial scene analysis and investigate which features Deep Learning models rely on when classifying aerial scenes. Based on the findings, we propose Deep Learning architectures that extract invariant feature representations across wet and dry season domains.

Semantic Segmentation

These works provide an overview of my research in semantic segmentation of aerial scenes.

Remote Sensing Scene Classification

These works provide an overview of my research in using remote sensing data and Deep Learning architectures to detect pollutant plants in China. I used Landsat-8 remote sensors (Thermal Infrared and Operational Land Imager) to create datasets consisting of thermal infrared (bands 10 and 11), short wave infrared (bands 6 and 7), and a geological ratio of the short-wave infrared image chips for two classes (cement plants and the surrounding land cover). I then trained and validated various state-of-the-art deep learning architectures (VGG, ResNet, EfficientNet) on the created datasets.

PhD Thesis

This research explores transferable features between domains, proposes Deep Learning architectures that combine salient feature extraction with a wider receptive field and highlights the importance of choosing appropriate feature priors for better model generalisation for domain adaptation and semantic segmentation tasks.