Drone Detection Using Velocity Estimation on Range-Doppler Diagrams

Detecting drones, amidst buildings or mountains is challenging. We propose an artificial intelligence-based algorithm that tracks the movements of a target on the range-Doppler diagram and extracts only actual targets. Since a low threshold is employed for detecting low-Radar Cross Section (RCS) targets, surrounding clutter and noise are also detected. Targets that were detected, along with their surroundings, were classified as targets, clutter, and noise using a three dimensional-deep convolutional neural networks (3D-DCNN) classification model. As a single target is detected as multiple due to sidelobes, the velocity was estimated using the 3D-DCNN regression model to eliminate detections in sidelobes.If the estimated velocity within the error range of the velocity value on the range-Doppler diagram, it is finally considered a target. The proposed models were trained using simulation data and verified by measuring actual drones. 1670 of the 1683 False Alarms (FAs) that occurred after constant false alarm rate (CFAR) were removed and all drones were detected. Using our proposed method, the number of misdetections in the sidelobe was reduced from 104 to 13, compared to when only the classification model was used.