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

TL;DR

The proposed 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.

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.

Figures (11)

• Figure 1: Visualization of input image contributions in DL models using Shapley additive explanations (SHAP) lundberg2017unified for ResNet resnet and DANN dann.
• Figure 2: Visualization of algorithms for reducing the impact of clutter according to domain. (a) Pixel-level clutter reduction algorithm. (b) Clutter-robust learning algorithm. (c) Feature-level clutter reduction algorithm. The red bounding box shows pixel-level errors, and the yellow bounding box shows feature-level errors.
• Figure 3: Overall pipeline of the proposed IRASNet framework. The CRM operates differently during the training and testing phases. The augmented and synthetic datasets are combined using a novel CRM and adversarial learning to derive domain-invariant, clutter-reduced features.
• Figure 4: Paired synthetic and measured images in the SAMPLE dataset.
• Figure 5: Visualization of feature distribution using t-SNE embeddings. (a) ResNet50 resnet. (b) DANN dann. (c) IRASNet. Features extracted from samples of the same category are shown in the same color, with synthetic (represented by o) and measured (represented by x) images.