研究目的
To employ radar micro-Doppler for detection and classification of small drones, distinguishing between drones and birds (target detection) and types of drones (target classification).
研究成果
Micro-Doppler radar has the potential for reliable small drone target detection and is also promising for classifying the type of drone. The boosting classifier achieved excellent target detection (99.4% correct detection) and good drone type classification (90% correct classification). The boosting classifier has some advantages over SVM, such as being easier to use and providing insight into the classifier and important features.
研究不足
The integration time, time step, and observation interval were suitable for helicopters but not for birds. The physical Kim Ling features are designed for humans and large land animals and may not be optimal for all types of drones and birds.
1：Experimental Design and Method Selection:
The study used radar micro-Doppler for detection and classification of small drones. Physical features were extracted from TVDs (Time Velocity Diagrams), and a boosting classifier was used for classification.
2：Sample Selection and Data Sources:
Radar measurements of small drones and birds were used. The drones included model helicopters, a fixed-wing model aircraft, and a quadrocopter. Birds of prey and a flock of common starlings were also measured.
3：List of Experimental Equipment and Materials:
A stationary radar operating at
4：7 GHz (X-band) was used. The radar was a CW (continuous-wave) radar with a sufficiently high sampling frequency to avoid aliasing in Doppler. Experimental Procedures and Operational Workflow:
Manual object detection and selection of time intervals where the object signal is present were performed. Object sequences were split into smaller ones with observation intervals 50 ms – 70 ms long. A clutter filter removed signal contents with radial velocities close to zero. For each object sequence, a TVD was created by overlapping short-time Fourier transforms.
5：Data Analysis Methods:
Physical features were extracted from the TVDs, and a boosting classifier was used for classification. The performance was compared with an SVM classifier.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
