Design of deep learning models
Deep Learning Innovations in Hyperspectral Imaging and Remote Sensing.
My research lies at the intersection of Computer Vision, Deep Learning, Remote Sensing, and Hyperspectral Imaging, with a focus on developing intelligent algorithms for analyzing complex Earth observation and imaging data. I work on advancing deep learning methodologies for hyperspectral image processing, environmental monitoring, and remote sensing applications. A significant part of my research focuses on hyperspectral image denoising, where I developed UNFOLD, a hybrid architecture that combines 3D U-Net, 3D CNN, and Transformer networks to effectively model both local spatial features and long-range spectral dependencies. Building upon this work, I introduced VISIONARY, a novel spatial-spectral attention mechanism that jointly analyzes spatial and spectral information to improve feature representation and denoising performance. To address the challenge of limited labeled data, I proposed RECREATE, a framework that integrates supervised contrastive learning and image inpainting to enhance feature discrimination and model generalization. Beyond denoising, I actively investigate hyperspectral methane plume segmentation for environmental monitoring. My recent work, FUMESNet, explores frequency-based Transformer architectures that leverage magnitude and phase information to jointly model spatial and spectral-frequency characteristics, enabling more accurate methane plume detection from satellite imagery. I further enhanced feature propagation through improved skip connections and efficient channel attention mechanisms. My research interests also extend to semantic segmentation, foundation models, and remote sensing applications, including adaptive pseudo-labeling for semi-supervised learning, satellite image analysis, and UAV-based environmental monitoring. Currently, I am exploring hyperspectral imaging for mining-related contamination assessment, developing deep learning models to analyze chemical compounds and quantify contamination effects using spectral signatures. Through these contributions, I aim to bridge the gap between advanced machine learning techniques and real-world remote sensing challenges, enabling more reliable, scalable, and impactful solutions for environmental monitoring and Earth observation.
