Machine Learning-Based Radar Signal Target Classification System: Intelligent Identification of UAVs, Vehicles, and Personnel

This project is the open-source Radar-Signal-Classification on GitHub (author: sehosr, released on June 2, 2026). It uses machine learning and neural network technologies to automatically identify three types of targets—UAVs, vehicles, and personnel—from radar signal features. Adopting an end-to-end design approach, it covers the complete workflow from signal preprocessing and feature engineering to classification models, providing technical support for intelligent radar decision systems and applicable to multiple scenarios such as UAV management and intelligent transportation.

Traditional radar relies on manual judgment of target types, which is inefficient and prone to subjective influences. With the development of AI and machine learning, radar signal processing is transitioning to intelligence. Automated target classification systems can extract features from complex echoes and identify targets in real time, which is of great significance for scenarios such as UAV management and border surveillance.

The project adopts a classic machine learning pipeline design: 1. Signal preprocessing (noise reduction, feature extraction); 2. Feature engineering (converting raw signals into vectors understandable by models); 3. Classification models (neural networks or other algorithms for target identification). The layered architecture ensures interpretability and facilitates subsequent optimization and expansion.

Target classification relies on the following features and technologies: 1. Micro-Doppler features: Doppler frequency shifts from different parts of the target (e.g., UAV rotors, limb movements of personnel); 2. Radar Cross Section (RCS): Differences in RCS among different targets (UAVs have small and highly variable RCS due to posture changes; vehicles have stable and large RCS; personnel have the smallest RCS with periodic fluctuations); 3. Time-frequency analysis: Methods like STFT or wavelet transform to convert signals into 2D time-frequency images; 4. Neural network architectures: Possible use of CNN (to learn spatial patterns), RNN/LSTM (to model temporal dependencies), or Transformer (to capture long-range dependencies).

The technology is applied in multiple domains: 1. UAV management: Distinguish between birds and UAVs to reduce false alarms; 2. Intelligent traffic monitoring: Vehicle classification helps with traffic flow analysis and road planning; 3. Border and perimeter security: Distinguish between personnel, vehicles, and animals for precise response; 4. Autonomous driving assistance: Improve the reliability of obstacle identification in severe weather conditions.

Current challenges and directions: 1. Data scarcity: Need for transfer learning or synthetic data generation; 2. Real-time performance: Model compression, quantization, and edge deployment; 3. Adversarial robustness: Improve stability under interference; 4. Multi-sensor fusion: Combine cameras and LiDAR to enhance perception capabilities.

The open-source nature of this project provides learning resources for researchers worldwide, accelerating technology dissemination and iteration. The combination of radar signal processing and machine learning is an important direction for perceptual intelligence. With the decline in hardware costs and algorithm optimization, it is expected to be popularized in more civilian scenarios, helping to build a safer and smarter society.