Hyperspectral Data Fusion
Abstract : A new era of spaceborne hyperspectral remote sensing (or imaging spectroscopy) has begun with the launch and operation of hyperspectral satellite missions (e.g., DESIS and PRISMA). Continuous spectral signatures of hyperspectral imagery enable the detection and identification of Earth surface materials and processes at a more detailed level that is not easy to achieve with conventional multispectral sensors. Spaceborne hyperspectral missions are expected to make an impact in various application fields, such as mineral mapping and environmental assessment; however, there are limitations in spatial resolution, observation coverage, and revisit cycle due to sensor design constraints. Data fusion is the key to extend the resolution and analysis range while fully utilizing the rich spectral information of hyperspectral images. This talk provides an overview of data fusion technology to overcome the limitations by fusing hyperspectral images with other data sources (e.g., multispectral images and LiDAR-derived digital surface model). We introduce recent data fusion techniques based on coupled matrix/tensor decomposition, co-learning, and non-convex optimization for four different tasks, namely, super-resolution, land cover classification, change detection, and spectral unmixing, followed by discussions on remaining challenges and future directions.