Regularisation in multi- and hyperspectral remote sensing change detection

    Research output: Chapter in Book/Report/Conference proceedingArticle in proceedingsResearchpeer-review

    115 Downloads (Pure)


    Change detection methods for multi- and hypervariate data look for differences in data acquired over the same area at different points in time. These differences may be due to noise or differences in (atmospheric etc.) conditions at the two acquisition time points. To prevent a change detection method from detecting uninteresting change due to noise or arbitrary spurious differences the application of regularisation also known as penalisation is considered to be important. Two types of regularisation in change detected by the multivariate alteration detection (MAD) transformation are considered: 1) ridge regression type and smoothing operators applied to the estimated weights in the MAD transform; and 2) pre-processing (before applying the MAD transformation) by noise reducing orthogonal transformations where the number of retained transformed variables can be considered a regularisation parameter. Regularisation by the former methods smooth the weights given to the individual bands in the MAD transformation and thus it penalises weights that fluctuate wildly as a function of wavelength; regularisation by the latter methods tends to smooth in the image domain. Also, regularisation may be necessary to prevent numerical instability especially when working on hyperspectral data.
    Original languageEnglish
    Title of host publication6th Geomatic Week Conference
    Publication date2005
    Publication statusPublished - 2005

    Cite this