Deep learning-based Doppler-spread target detection with attention mechanism

In radar systems, Doppler spread issues commonly arise in space small target detections when integration is adopted during a long coherent processing interval (CPI). Even with precise compensation for translational effects, phase variations introduced by changes in target observation attitudes (TOA) can still cause the target's echo energy to spread across multiple Doppler cells. As the TOA undergoes several periods over a CPI, the echo energy disperses into equidistant Doppler cells within a range cell in the range-Doppler (RD) map, which poses challenges for traditional methods and results in degraded detection performance. Leveraging the convolutional neural network's ability to autonomously extract data features, we propose a neural network based on the Doppler auto-correlation attention mechanism (DAAM) to enhance detection accuracy and precisely estimate target positions by exploiting the sparse characteristics of the target Doppler spectrum. Training and testing are conducted using an RD map dataset, and experimental results demonstrate the network's advanced performance in both target detection and position estimation.