Multi-Aspect SAR Target Recognition Based on Prototypical Network with a Small Number of Training Samples

Area of Science:

• Computer Science
• Electrical Engineering
• Remote Sensing

Background:

• Synthetic Aperture Radar (SAR) automatic target recognition (ATR) is crucial for military and civilian applications.
• SAR image recognition is challenging due to significant appearance variations of targets across different imaging aspects.
• Current deep learning methods for SAR ATR struggle with limited, expensive-to-acquire datasets, hindering model training.

Purpose of the Study:

• To develop a robust and reliable multi-aspect SAR target recognition method.
• To address the challenge of small sample sizes in SAR datasets for deep learning models.
• To enhance recognition accuracy for targets in SAR imagery with limited training data.

Main Methods:

• Proposes a novel multi-aspect SAR target recognition method utilizing a prototypical network architecture.
• Incorporates multi-task learning and multi-level feature fusion techniques to boost performance.
• Evaluates the method's effectiveness on the challenging MSTAR dataset.

Main Results:

• The proposed prototypical network-based method achieves high recognition accuracy, comparable to models trained on complete datasets.
• Demonstrates significant improvements in multi-aspect SAR target recognition, especially under small sample conditions.
• Confirms the method's applicability to various feature extraction models for small sample learning.

Conclusions:

• The developed prototypical network approach offers a viable solution for SAR ATR with limited data.
• Multi-task learning and feature fusion effectively enhance recognition robustness and accuracy.
• This method provides a promising direction for future research in few-shot SAR target recognition.