Abstract:Recently, computer-aided design models and electromagnetic simulations have been used to augment synthetic aperture radar (SAR) data for deep learning. However, an automatic target recognition (ATR) model struggles with domain shift when using synthetic data because the model learns specific clutter patterns present in such data, which disturbs performance when applied to measured data with different clutter distributions. This study proposes a framework particularly designed for domain-generalized SAR-ATR called IRASNet, enabling effective feature-level clutter reduction and domain-invariant feature learning. First, we propose a clutter reduction module (CRM) that maximizes the signal-to-clutter ratio on feature maps. The module reduces the impact of clutter at the feature level while preserving target and shadow information, thereby improving ATR performance. Second, we integrate adversarial learning with CRM to extract clutter-reduced domain-invariant features. The integration bridges the gap between synthetic and measured datasets without requiring measured data during training. Third, we improve feature extraction from target and shadow regions by implementing a positional supervision task using mask ground truth encoding. The improvement enhances the ability of the model to discriminate between classes. Our proposed IRASNet presents new state-of-the-art public SAR datasets utilizing target and shadow information to achieve superior performance across various test conditions. IRASNet not only enhances generalization performance but also significantly improves feature-level clutter reduction, making it a valuable advancement in the field of radar image pattern recognition.

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is: **How to achieve domain generalization (DG) of synthetic SAR (Synthetic Aperture Radar) data through a deep - learning model without using real - measurement data, in order to improve the performance of automatic target recognition (ATR)**. Specifically, the paper proposes a new framework named IRASNet, aiming to overcome the following challenges: 1. **Domain shift problem**: When an ATR model trained with synthetic data is applied to real - measurement data with different clutter distributions, its performance will decline. This is because the model has learned specific clutter patterns in the synthetic data, which do not exist or are different in the real data. 2. **Feature - level clutter reduction**: Traditional clutter reduction methods are usually carried out at the pixel level, which may lead to information loss and cannot adapt well to changes in different domains. Therefore, a method that can effectively reduce clutter at the feature level is required. 3. **Domain generalization without real data**: Many existing domain adaptation (DA) methods rely on real - measurement data for model adjustment, but in practical applications, it may be very difficult or impossible to obtain these data. Therefore, a solution that can achieve good generalization only relying on synthetic data is needed. To solve these problems, the paper proposes the following key points: - **Clutter Reduction Module (CRM)**: This module reduces the influence of clutter at the feature level by maximizing the signal - to - clutter ratio (SCR) on the feature map, while retaining target and shadow information, thereby improving ATR performance. - **Adversarial learning**: Combine CRM with adversarial learning to extract domain - invariant features for clutter removal. This method can narrow the gap between synthetic data and real data without using real - measurement data. - **Positional Supervision Task (PST)**: By introducing mask ground truth encoding, enhance the model's understanding of the position information of target and shadow areas, thereby increasing SCR and enhancing the inter - class discrimination ability. In summary, the main contribution of this paper is to propose a new DG framework, IRASNet, which not only improves the clutter reduction effect at the feature level, but also achieves domain generalization, thereby significantly improving the performance of SAR - ATR.

IRASNet: Improved Feature-Level Clutter Reduction for Domain Generalized SAR-ATR

Learning Invariant Representation Via Contrastive Feature Alignment for Clutter Robust SAR ATR

Soft Segmented Randomization: Enhancing Domain Generalization in SAR-ATR for Synthetic-to-Measured

Learning Invariant Representation via Contrastive Feature Alignment for Clutter Robust SAR Target Recognition

ADERLNet: Adaptive Denoising Enhancement Representation Learning for Low-Latency and High-Accurate Target Detection on SAR Sensors

SAR ATR for Limited Training Data Using DS-AE Network

Optimizing Target Recognition in Synthetic Aperture Radar Imagery: A Hyperparameter- Tuned Approach With Iterative Transfer Learning and Branched-Convolutional Neural Network

Target Recognition in SAR Images by Deep Learning with Training Data Augmentation

Capsule Broad Learning System Network for Robust Synthetic Aperture Radar Automatic Target Recognition with Small Samples

Weighted residual network for SAR automatic target recognition with data augmentation

Adaptive Convolutional Subspace Reasoning Network for Few-Shot SAR Target Recognition

Transductive Prototypical Attention Reasoning Network for Few-Shot SAR Target Recognition

Semi-Supervised SAR ATR Framework with Transductive Auxiliary Segmentation

Deep-IRTarget: An Automatic Target Detector in Infrared Imagery Using Dual-Domain Feature Extraction and Allocation

Crucial Feature Capture and Discrimination for Limited Training Data SAR ATR

SAR ATR Method with Limited Training Data via an Embedded Feature Augmenter and Dynamic Hierarchical-Feature Refiner

C-RISE: A Post-Hoc Interpretation Method of Black-Box Models for SAR ATR

Single-Scene SAR Image Data Augmentation Based on SBR and GAN for Target Recognition

An Integrated Network for SA-ISAR Image Processing with Adaptive Denoising and Super-Resolution Modules

RATIR-Net: Adaptive SAR Image Reconstruction Based on Transformer Architecture

Adaptive Scattering Feature Awareness and Fusion for Limited Training Data SAR Target Recognition