One-Dimensional Convolutional Neural Network for Object Recognition Through Electromagnetic Backscattering in the...

Area of Science:

• Electromagnetics
• Machine Learning
• Signal Processing

Background:

• Item recognition traditionally relies on radar or computer vision.
• Machine and deep learning, particularly Convolutional Neural Networks (CNNs), are now state-of-the-art for image-based recognition.
• Vision-based methods face limitations like data unavailability, poor image quality, and privacy concerns.

Purpose of the Study:

• To investigate a novel machine learning object recognition approach using electromagnetic backscattering in the frequency domain.
• To address the limitations of vision-based object recognition systems.

Main Methods:

• Employed a 1D Convolutional Neural Network (CNN) to analyze electromagnetic backscattered signals.
• Collected data through backscattering measurements in the millimeter-wave (mmWave) band within controlled environments.
• Utilized signal generators and spectrum analyzers for reliable data acquisition.

Main Results:

• Achieved 100% accuracy in object detection.
• Attained 84% accuracy in object recognition for two object classes.
• Identified a trade-off between accuracy and processing speed based on signal bandwidth and frequency steps.

Conclusions:

• Electromagnetic-based object recognition shows promise as a complement or alternative to vision-based systems, especially when vision is impractical.
• The proposed approach demonstrates flexibility and potential for real-time applications due to adjustable accuracy-speed trade-offs.