A Neural Network with Convolutional Module and Residual Structure for Radar Target Recognition Based on...

Area of Science:

• Computer Science
• Artificial Intelligence
• Signal Processing

Background:

• Conventional deep neural networks require significant depth for high recognition accuracy.
• Deep networks can suffer from saturation, leading to decreased accuracy with increased layers.
• Recognition of High Range Resolution Profile (HRRP) signals presents challenges for traditional models.

Purpose of the Study:

• To propose a novel neural network model to address the limitations of conventional deep networks in recognition tasks.
• To enhance feature separability using a new loss function incorporating boundary constraints and center clustering.
• To improve the accuracy and robustness of HRRP signal recognition.

Main Methods:

• A neural network model incorporating a micro convolutional module and residual structure was developed.
• A novel loss function was designed, integrating boundary constraints and center clustering for improved feature separability.
• The model was evaluated using a simulated dataset of HRRP signals from thirteen 3D CAD object models.

Main Results:

• The proposed model achieved higher recognition accuracy compared to common network structures.
• The model demonstrated enhanced robustness in HRRP signal recognition.
• The micro convolutional module and residual structure contributed to fewer hyper-parameters and flexible extensibility.

Conclusions:

• The presented neural network model effectively overcomes the saturation problem in deep networks.
• The novel loss function significantly enhances feature separability, leading to improved recognition performance.
• The proposed approach offers a more accurate and robust solution for HRRP signal recognition.