A Feature Extraction Framework Based on Relative Polarization Information of Polarimetric Radar

Polarization constitutes one of the fundamental dimensions of electromagnetic waves. The acquisition of polarization information holds great significance for deciphering the target scattering mechanism, thereby facilitating target detection and recognition. In the current landscape of polarization information acquisition and processing, existing methods hinge on precision measurement. These approaches involve comparing the polarization information of measured echoes with theoretical models or empirical observations to classify targets. Nevertheless, practical radar systems suffer from polarization measurement errors. Polarization calibration comes at a high cost, and achieving precise polarization measurement has long remained an arduous challenge. Moreover, in the realm of target detection and recognition, the essential purpose is to obtain the differential characteristics of different targets. Consequently, leveraging relative polarization differences to distinguish among diverse targets within the constraints of limited polarization measurement accuracy emerges as a feasible technique. In this paper, we propose a framework for relative polarization information processing. Based on the metric of polarization difference, the polarization characteristic in different dimensions is utilized to extract the multi-dimensional features, and to realize the target recognition. The proposed relative polarization information processing framework has lower requirements on polarization measurement, which shows superior performance.