Feature extraction from millimetre wave radar images

Abstract

This thesis describes research performed into the segmentation and classification of features on images of wound terrain generated from an airborne millimetre wave radar.
The principles of operation of the radar are established and it is shown how an image is produced from this particular radar. The parameters such as wavelength, antenna size and pulse length are related to the images and a mathematical description of the radar data is given.
The effectiveness of established image processing techniques is reviewed when applied to millimetre wave radar images and a statistical classification technique is seen to yield encouraging results. This method of segmentation and classification is then extended to make optimal use of the available information from the radar.
An orthogonal expansion of the Poincaré sphere representation of polarised radiation is established and it is shown how different terrain types cluster in the eigenspace of these spherical harmonics. Segmentation then follows from the clustering properties of pixels within this multidimensional eigenspace and classification from the locations of the clusters.