Signal Property Information-Based Target Detection with Dual-Output Neural Network in Complex Environments

Area of Science:

• Radar Signal Processing
• Deep Learning Applications
• Target Detection Algorithms

Background:

• Traditional Constant False-Alarm Ratio (CFAR) algorithms struggle with complex environments, multiple targets (MT), and clutter edges (CE) due to inaccurate background noise estimation.
• Single-input single-output neural networks exhibit performance degradation in dynamic scenes because of fixed threshold mechanisms.

Purpose of the Study:

• To introduce a novel single-input dual-output network detector (SIDOND) that overcomes the limitations of traditional and single-output network detectors.
• To enhance the robustness and accuracy of target detection in challenging radar scenarios using data-driven deep neural networks (DNN).

Main Methods:

• Developed a single-input dual-output network detector (SIDOND) utilizing deep neural networks (DNN).
• Implemented one output for signal property information (SPI)-based estimation of the detection sufficient statistic.
• Utilized a second output to establish a dynamic-intelligent threshold mechanism based on the threshold impact factor (TIF).

Main Results:

• The proposed SIDOND demonstrated superior robustness and performance compared to conventional model-based and single-output network detectors.
• Experimental validation confirmed the effectiveness of SIDOND in complex environments with multiple targets and clutter edges.
• The threshold impact factor (TIF) effectively simplified target and background environment information for dynamic thresholding.

Conclusions:

• SIDOND offers a significant advancement in radar target detection, particularly in complex and dynamic environments.
• The dual-output network architecture and dynamic thresholding provide a more adaptable and accurate detection solution.
• Visual explanation techniques were employed to elucidate the operational principles of the SIDOND.