Research project SR/67/147 (Research action SR)
In the design and development of remote sensors, a trade-off exists between SNR, spatial and spectral resolutions due to physical limitations and data-transfer requirements. This trade-off still hampers the use of remote sensing data for many applications. The Hyperion sensor currently offers the highest spectral resolution available from space. The spatial resolution of only 30 meters however restricts a proper use of the inherent potential of these data to easily distinct different objects. On the other hand, sensors such as Quickbird and WorldView-1&2 are able to offer very high resolution imagery, but this is at the expense of their spectral resolution: panchromatic at sub-meter spatial resolution, and 4 to 8 (Worldview2) spectral bands with approximately 2.5m spatial resolution. Although airborne acquisitions do allow capturing hyperspectral image data at a high spatial resolution (1-7 m) (e.g., AHS, APEX), their spatial coverage is limited. Moreover, many applications, such as detailed land cover mapping, studies focusing on vegetation dynamics and vegetation status assessment require even higher, i.e., cm level, spatial hyperspectral imagery.
Objective
In this project, we aim to narrow this spectral-spatial gap by further enhancement of state-of-the-art data fusion and unmixing techniques, in order to open up the potential in future applications.
Method
In order to obtain the highest spectral resolution possible, APEX data will be ordered and asked to be flown at different heights in the study areas. APEX is currently a top-level state-of-the-art sensor with a spectral resolution that is unmatched by other airborne sensors. To obtain very high spatial (centimeters) resolution imagery, a digital photogrammetric camera will be used onboard a micro-UAV. Both APEX and micro-UAV data will be gathered simultaneously in the summer of 2011.
After the necessary pre-processing, the image data will then be used in combination with existing and possibly new field data to develop, test and validate the two different approaches that are taken to enhance the spatial information in hyperspectral images: state-of-the-art data fusion and advanced unmixing techniques.
Result
- An adapted version of the fusion model recently developed by Visielab, to the fusion problem at hand.
- Development of new fusion techniques based on the principle of the existing model by (1) inclusion of a restoration of the hyperspectral image, and (2) using the same model to fuse spectral information into the hyperspatial image instead of vice versa. Ideally, both techniques would lead to the same final result, but due to differences in the practical application of both techniques, it is expected that one technique will perform better than the other.
- Adjustment of existing, and development of new spectral and spatial unmixing methods that enable the production of hyperspatial information products.
