A Novel Radar HRRP Recognition Method with Accelerated T-Distributed Stochastic Neighbor Embedding and Density-Based...

Area of Science:

• Radar Signal Processing
• Machine Learning for Target Recognition

Background:

• High-resolution range profiles (HRRPs) are valuable for radar target analysis but conventional methods require prior target knowledge and struggle with real-time processing.
• Existing algorithms often fail to classify unlabeled samples or handle large datasets efficiently, limiting their practical application.

Purpose of the Study:

• To develop a novel, automatic high-resolution range profile (HRRP) recognition method for classifying unlabeled samples.
• To address the limitations of conventional algorithms by enabling real-time processing and eliminating the need for prior target information.
• To improve recognition performance under challenging conditions such as large azimuth angle ranges and low signal-to-noise ratio (SNR).

Main Methods:

• Preprocessing of high-resolution range profiles (HRRPs).
• Dimensionality reduction using Principal Component Analysis (PCA).
• Visualization and further dimensionality reduction via t-distributed Stochastic Neighbor Embedding (t-SNE) with Barnes-Hut approximation.
• Classification using density-based clustering.

Main Results:

• The proposed method successfully classifies unlabeled HRRP samples automatically, even when the number of categories is unknown.
• Dimensionality reduction using PCA and t-SNE significantly improved computation speed.
• Density-based clustering demonstrated superior recognition performance compared to conventional algorithms, particularly under large azimuth angle variations and low SNR.

Conclusions:

• The novel HRRP recognition method offers an effective solution for automatic target classification without prior knowledge.
• The combined use of PCA and t-SNE enhances computational efficiency for processing large HRRP datasets.
• The proposed approach provides robust performance in complex radar environments, outperforming traditional methods.