Research on UAV Classification and Recognition Technology Based on Radio Frequency Signals and Convolutional Neural Networks

This GitHub open-source project demonstrates how to use radio frequency signal processing and convolutional neural network technology to achieve automatic classification and recognition of UAVs. It includes complete MATLAB code for spectrogram generation and Python implementation of CNN models, providing a full workflow from signal acquisition to model deployment. It is a practical reference resource for the application of combining radio frequency signal processing with deep learning.

With the popularization of UAV technology, its abuse brings security risks such as illegal intrusion and privacy violations, making effective classification and recognition of UAVs an important issue. Traditional methods like radar and optical cameras are costly and greatly affected by the environment. Recognition technology based on radio frequency signals has advantages such as non-contact, all-weather operation, and low cost, which has become a research hotspot in recent years.

Core solution: Capture UAV remote control/video transmission signals → Generate time-frequency spectrograms with MATLAB → Classify using CNN models. Compared with traditional methods of manually designing features (such as carrier frequency, modulation mode), CNN can automatically learn discriminative features in spectrograms, improving classification accuracy and robustness.

Use MATLAB code to implement Short-Time Fourier Transform (STFT), converting one-dimensional time-domain signals into two-dimensional time-frequency spectrograms. Signal differences between different UAVs form unique visual patterns. Key parameters (window function, window length, overlap rate) affect the quality of spectrograms, and the generated images are used as input data for CNN.

Build CNN using Python and TensorFlow/PyTorch, including convolutional layers (extracting local features), pooling layers (dimensionality reduction and enhancing invariance), and fully connected layers (classification decision). Design considerations include input size matching, number of categories, and overfitting issues (solved via data augmentation, Dropout, and regularization).

The project records the complete workflow, and the open-source code facilitates reproduction and expansion. The technology can be integrated into UAV defense systems to achieve real-time detection and recognition. Compared with traditional radar, it has advantages such as flexible deployment, easy upgrade, and controllable cost, making it suitable for security needs in scenarios like airports and military bases.

The repository contains graduation thesis documents, MATLAB and Python code. Learning directions include: radio frequency signal characteristics and preprocessing, MATLAB signal analysis and visualization, application of CNN in signal classification, and end-to-end deep learning system design. Interdisciplinary integration provides references for fields like intelligent radio.