Reducing False Alarm in Target Sidelobes Using Time-Varying Trajectories in Range-Doppler Diagram With Conditional 3D-DCNN

The increasing use of drones in civil and defense applications poses threats to critical infrastructure, with low-radar cross section (RCS) targets detection in cluttered environments challenging conventional radar methods like constant false alarm rate (CFAR), which often result in high false alarm rates (FAs). This paper proposes a conditional-three dimensional-deep convolutional neural networks (C-3D-DCNN) that reduces FAs by utilizing the time-varying target trajectories observed over consecutive frames in the range-Doppler diagram with velocity information as a conditional input. By adjusting a Gaussian filter in accordance with the target’s velocity in the suggested C-3D-DCNN structure, the algorithm distinguishes real targets from clutter, noise, and target sidelobes that appear along the Doppler axis. The proposed method was validated using frequency-modulated continuous wave radar to detect low-RCS targets and demonstrated a reduction in FAs in clutter environments. Verification with 200 simulation datasets showed a 99.8 % reduction in FAs excluding the sidelobe region along the range axis compared to CFAR and a 95.22 % reduction compared to 3D-DCNN, while all FAs in the Doppler sidelobe region were eliminated (100 % reduction). We also verified the effectiveness of this method in experiments with real drones by removing 100 % of the FA in both the excluding sidelobe in range axis and Doppler sidelobe regions when tested with two drones. This approach simplifies previous two-step processes into a one-step solution, offering a more efficient framework for drone surveillance systems.