Intelligent Understanding and Processing of Airborne Sense and Avoid Radar Data with Antenna Diversities

A technology for knowledge-based detection and classification of aerospace hazards through a new generation of on-board radar sensor is discussed. The airborne radar system has to perform in various situations where types, shapes and sizes of hydrometeors may vary significantly, and existing field measurements from radar usually lack of ground truths and clean content. A single cell Monte-Carlo sampling-based simulation approach is developed. The simulation tool has the ability to handle large amount of uncertain variables at different frequencies, and incorporate a numerical weather prediction model (NWP). Dual-polarization radar signatures of hydrometeors and other hard targets are generated based on electromagnetic scattering theories. Knowledge gained from the Monte-Carlo simulation is learned using Gaussian Mixture Model (GMM), which is able to model mutual correlations among multidimensional, dual-polarization variables. A Bayesian classifier is then built upon the GMM models. Five classes of hydrometeors including rain, snow, melting snow, hail and melting hail are considered.

Simulated “clean” scenarios of aviation weather hazard encounters, simulated radar scans based on NWP models with mixing species and attenuation effects, and actual radar observations of a winter storm are combined to validate the proposed scheme. Very good accuracy (>98%) is achieved for distinguishing rain, snow and hail. Though the system is affected by attenuation, it is still able to identify most regions in simulated radar scans where hail presents, which satisfies the key requirement of the application. Classification results of actual ground-based radar observations of the winter storm are also match the observed ground truth. The hazard types in the studies so far are mainly precipitations and hydrometeors, but can also be extended to air-traffic, biological targets, etc.