3-D electromagnetic-model-based absolute attitude estimation using a deep neural network

With the improvement of range resolution, wideband radars can achieve not only the range and velocity but also signature information of the target, such as the target RCS and the distribution of the scattering centres. Since radar target signature is closely related to target's geometric structure and attitude, it is possible to estimate attitude from wideband radar echoes. To this end, we propose a novel end-to-end Radar Target Attitude Estimation Network (RTAENet) to estimate the absolute attitude of the radar target. A difficulty for training the network is that a large dataset with ground truth is required. To solve this issue, we introduce the three-dimensional (3-D) electromagnetic model to generate training samples. Since the RTAENet requires very little feature engineering by hand and can take advantage of increases in the amount of available data, it achieves high accuracy for attitude estimation. Experiments using both data predicted by a high-frequency electromagnetic code and data measured in an anechoic chamber demonstrate the feasibility of the proposed method.