Deep Learning Approach to UAV Detection and Classification by Using Compressively Sensed RF Signal

Area of Science:

• Electrical Engineering
• Computer Science
• Aerospace Engineering

Background:

• Unmanned Aerial Vehicles (UAVs) misuse poses significant security challenges.
• Traditional radio frequency (RF) detection methods for UAVs face bandwidth and real-time processing limitations.
• Effective identification and detection of UAVs are crucial for security and airspace management.

Purpose of the Study:

• To develop an efficient and accurate method for detecting and identifying UAVs, including their type and flight patterns.
• To overcome the limitations of traditional RF detection methods by employing compressed sensing and deep learning.
• To improve the real-time processing and data transmission efficiency for UAV signal sampling.

Main Methods:

• Utilized compressed sensing and a multi-channel random demodulator for efficient UAV communication signal sampling.
• Proposed a multi-stage deep learning (DL) approach for UAV detection, identification, and flight pattern classification.
• Employed Deep Neural Network (DNN) for presence detection and Convolutional Neural Network (CNN) for type and flight pattern classification, validated using 10-fold cross-validation.

Main Results:

• Compressed sensing effectively sampled UAV and controller communication signals.
• The multi-stage DL detection method demonstrated superior efficiency and accuracy compared to existing techniques.
• The proposed method achieved over 99% accuracy in detecting UAV presence, type, and flight model.

Conclusions:

• Compressed sensing is an effective technique for sampling UAV communication signals.
• The multi-stage deep learning approach significantly enhances the accuracy and efficiency of UAV detection and identification.
• This method offers a robust solution for identifying UAV presence, type, and flight patterns, improving airspace security.