Extended object detection and extraction in a high resolution polarimetric surveillance radar

Detection in most surveillance radars is based on assumptions on point targets against a more or less homogeneous background. Currently, the resolution of many new types of radar is increasing, at least in the range dimension. Therefore many objects no longer can be considered as points. Also, the background is getting more diverse, in statistical terms, since many individual objects in the background are no longer mixed coherently. The current paper proposes as staged approach to detection and parameter assessment of extended objects and adds classification based on the polarimetric features. The evaluation of the approach is based on recordings with a polarimetric radar of real natural scenes and both real and artificially inserted extended objects. It has been observed that a multi-stage detection outperforms a classical method, in terms of the ratio between wanted and unwanted alarms.