ON-BOARD MULTISPECTRAL ALIGNMENT FOR IMPROVED CLASSIFICATION AND COMPRESSION RESULTS

摘要

In the future the significance of on -board processing of acquired images will increase. The reason is less a changed objective in the user community but rather the rising impact of engineering limitations. Prominent exam ples are given by micro-, mini-, and small-satellites that are currently gaining in popularity. One of the most substantial implications in the design of these satellites is a limited downlink bandwidth for various reasons, e.g. a restricted power budget and / or a small number of ground receiving stations. Therefore on-board processing becomes almost mandatory since it enables the utilisation of the raw imagery and reduces the amount of data by computing high-level products, which then again eases the requirements on the communication link. The advan tages of this approach are pre -dominant for highly specialised missions. However, even for general imaging missions on-board processing can provide additional benefits. This paper illustrates the gains of on-board processing by choosing multispectral alignment as a demonstrating application. The multispectral band-to-band alignment is one key factor in remote sensing and a mismatch between the different spectral bands inevitably leads to de creased accuracy for every application that makes explicitly use of the spectral information. The causes for the displacement are mainly determ ined by the design of the acquisition unit, i.e. the arrangement of the charge coupled devices. This paper proposes a technique to measure the actual displacement. Subsequently the multispectral scenes are aligned using a parallel processing unit that is currently developed for the mini-satellite mission X-Sat. The results show that satellite missions profit by the on -board alignment since the data reliability is improved. This is of major interest for any further processing of the image data, which will continuously be pushed towards space over the next few years.