A Convolutional Neural Network-Based Method for Discriminating Shadowed Targets in Frequency-Modulated...

Area of Science:

• * Radar Systems Engineering
• * Machine Learning Applications
• * Signal Processing

Background:

• * The radar shadow effect hinders accurate target detection when one object obstructs another.
• * Frequency-Modulated Continuous-Wave (FMCW) radars are increasingly used due to their cost-effectiveness and compact size.
• * Reliable target discrimination is crucial for various applications, including automotive and surveillance.

Purpose of the Study:

• * To develop a novel method for overcoming the radar shadow effect in FMCW radar systems.
• * To enhance the capability of FMCW radars in discerning targets within the shadow region of other objects.
• * To improve the reliability of target discrimination in complex radar environments.

Main Methods:

• * Utilized Short-Time Fourier Transform (STFT) analysis to process radar-received signals.
• * Developed a Convolutional Neural Network (CNN) model that analyzes spectrograms.
• * Trained the CNN to differentiate between targets in shadow regions and those clearly visible.

Main Results:

• * The proposed CNN-based method achieved a test accuracy of 92%.
• * The method demonstrated a low standard deviation of 2.86%, indicating consistent performance.
• * Successfully discerned whether targets were located within the shadow region of other objects.

Conclusions:

• * The novel CNN approach effectively addresses the radar shadow effect in FMCW systems.
• * The method offers a significant improvement in target discrimination accuracy.
• * This technique enhances the practical utility of low-cost FMCW radars in challenging scenarios.