A Low Angle Target Altitude Measurement Method Based on Transfer Learning and Deep Neural Networks

When tracking low-altitude targets, the presence of complex multipath signals can cause significant amplitude and phase distortions in the radar's received signals, leading to degraded performance of physics-driven model algorithms. In this paper, we propose a low-angle target direction of arrival (DOA) estimation method based on transfer learning and deep neural networks (DNN) to mitigate amplitude and phase distortions, thereby enhancing the DOA estimation performance of physics-driven algorithms. In the training phase, transfer learning is incorporated into the DNN training process, with samples from different terrain conditions trained sequentially. This approach allows for better training of the network in complex terrain scenarios, even with limited data. The trained network model is used to optimize the covariance matrix of the test dataset, which is then employed for DOA estimation. The proposed method improves DOA estimation performance without requiring additional data and effectively addresses the impact of varying terrain conditions on target angle measurements. The effectiveness of the method is validated through real data.