Benchmarking Suite for Synthetic Aperture Radar Imagery Anomaly Detection (SARIAD) Algorithms

TL;DR

Problem: There is no standardized benchmark for anomaly detection in synthetic aperture radar (SAR) imagery. Approach: We present SARIAD, a modular benchmarking suite built atop Anomalib that provides a SAR-specific data module, normal-data generation for datasets lacking normal examples, preprocessing options including despeckling, a model pipeline, and OpenVINO deployment. Contributions: integration of SAR datasets (MSTAR, HRSID), a configurable evaluation workflow, and zero-shot evaluation of PaDiM, DFM, and WinCLIP with multiple metrics and visuals. Findings: PaDiM and WinCLIP show strong performance depending on metric and dataset, while HRSID reveals challenges in pixel-level localization, underscoring the need for SAR-tailored representations. Significance: Enables reproducible SAR anomaly detection research and accelerates development by standardizing datasets, metrics, and tooling; public repo at the provided URL.

Abstract

Anomaly detection is a key research challenge in computer vision and machine learning with applications in many fields from quality control to radar imaging. In radar imaging, specifically synthetic aperture radar (SAR), anomaly detection can be used for the classification, detection, and segmentation of objects of interest. However, there is no method for developing and benchmarking these methods on SAR imagery. To address this issue, we introduce SAR imagery anomaly detection (SARIAD). In conjunction with Anomalib, a deep-learning library for anomaly detection, SARIAD provides a comprehensive suite of algorithms and datasets for assessing and developing anomaly detection approaches on SAR imagery. SARIAD specifically integrates multiple SAR datasets along with tools to effectively apply various anomaly detection algorithms to SAR imagery. Several anomaly detection metrics and visualizations are available. Overall, SARIAD acts as a central package for benchmarking SAR models and datasets to allow for reproducible research in the field of anomaly detection in SAR imagery. This package is publicly available: https://github.com/Advanced-Vision-and-Learning-Lab/SARIAD.

Figures (4)

• Figure 1: Example images of pure background from HRSID that are used as "normal" data.
• Figure 2: Example images of generating "normal" data for 5 of 10 classes in the MSTAR dataset. MSTAR only has target chips and does not have background images that are needed for SARIAD. The first row is the input images of the target chip. The second row shows the segmentation results of identifying the targets. The last row displays the "normal" images generated by sampling from the background pixels to fill in the targets.
• Figure 3: Example output resulting from applying PaDiM to a target chip in MSTAR. Despite not having pixel-level ground truth, the model predicts a region of the image where a possible anomaly or target is present.
• Figure 4: Example output resulting from applying PaDiM to a target chip in HRSID.