Machine Learning Based Object Classification and Identification Scheme Using an Embedded Millimeter-Wave Radar...

Area of Science:

• Electrical Engineering
• Machine Learning
• Radar Systems

Background:

• Object detection and classification are critical for radar sensor design.
• Key parameters include target movement and radar cross-sections (RCS).
• Existing methods may require complex signal processing toolboxes.

Purpose of the Study:

• To evaluate the feasibility and effectiveness of a 24 GHz radar system for object detection.
• To develop and apply a supervised machine learning model for target classification based on RCS.
• To compare different Support Vector Machine (SVM) classification strategies.

Main Methods:

• Utilized a 24 GHz radar with low-noise microwave amplifiers.
• Trained a supervised machine learning model (SVM) using recorded target data.
• Classified targets into four categories based on RCS and distance.
• Compared one-against-all, one-against-one, and directed acyclic graph SVM methods.

Main Results:

• Achieved approximately 96.6% accuracy and a 96.5% F1-score using the one-against-one SVM with an RFB kernel.
• Demonstrated effective classification of objects at varying distances.
• The proposed contactless radar and SVM approach showed high performance.

Conclusions:

• The 24 GHz contactless radar system combined with an SVM algorithm is effective for real-time target detection and categorization.
• The one-against-one SVM method with an RFB kernel provides superior classification performance.
• This approach eliminates the need for a signal processing toolbox for real-time applications.