1IMM, Department of Mathematical Modelling,
Technical University of Denmark, Building 321, DK-2800 Lyngby, Denmark
e-mail aa,firstname.lastname@example.org, http://www.imm.dtu.dk
2DCRS, Danish Center for Remote Sensing, Department
of Electromagnetic Systems,
Technical University of Denmark, Building 348, DK-2800 Lyngby, Denmark
e-mail email@example.com, http://www.dcrs.dtu.dk
This contribution describes and uses a statistical, orthogonal transformation of two sets of multivariate data based on the established canonical correlations analysis. The transformation concentrates information on change in multivariate data by looking at differences between the canonical variates ordered by ascending canonical correlations. As opposed to methods that work on simple difference images, this method is invariant to affine and linear transformations. This means that the change detected is insensitive to for instance 1) differences in gain and offset in a measuring device, and 2) radiometric and atmospheric corrections that are linear in the grey level numbers. The method applies to space- and airborne optical and radar data and is illustrated here with (airborne) polarimetric EMISAR data covering the Kalø area. Both C-band data (24 March 1996, 4 July 1996) and L-band data (22 March 1996, 4 July 1996) are used. The resulting images show a clear distinction between the `super-classes' cultivated fields and natural vegetation. The new variates are therefore useful for the construction of masks under which one can analyze these `super-classes' separately, for instance by applying the above method iteratively.