Patch distribution modeling framework adaptive cosine estimator (PaDiM-ACE) for anomaly detection and localization in synthetic aperture radar imagery

2025-04-10

AI Summary

To address the challenge of detecting and precisely localizing weak and small anomalies in synthetic aperture radar (SAR) imagery, this paper proposes PaDiM-ACE: the first integration of the Adaptive Cosine Estimator (ACE) into the patch-wise distribution modeling pipeline of PaDiM, replacing the unbounded Mahalanobis distance with a bounded cosine similarity for anomaly scoring. The method synergistically combines Vision Transformer (ViT)-extracted pre-trained features, multi-scale patch-level statistical modeling, covariance-adaptive normalization, and ACE-based scoring, thereby significantly enhancing the robustness and discriminability of anomaly responses. Evaluated on multiple public SAR datasets, PaDiM-ACE achieves state-of-the-art image-level and pixel-level AUROC scores—surpassing both the original PaDiM and leading unsupervised anomaly detection methods—particularly improving localization accuracy for weak and small targets. The source code is publicly available.

Technology Category

• Computer Vision: Adversarial Attacks & Robustness
• Machine Learning: Adversarial Learning & Robustness
• Data Mining & Knowledge Management: Anomaly/Outlier Detection

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Security and Privacy: Large-scale security measurements
• Graph Algorithms and Modeling for the Web: Algorithms and analysis for heterogeneous, signed, attributed, multi-relational, temporal, higher-order, and annotated Web-related graphs

Abstract

This work presents a new approach to anomaly detection and localization in synthetic aperture radar imagery (SAR), expanding upon the existing patch distribution modeling framework (PaDiM). We introduce the adaptive cosine estimator (ACE) detection statistic. PaDiM uses the Mahalanobis distance at inference, an unbounded metric. ACE instead uses the cosine similarity metric, providing bounded anomaly detection scores. The proposed method is evaluated across multiple SAR datasets, with performance metrics including the area under the receiver operating curve (AUROC) at the image and pixel level, aiming for increased performance in anomaly detection and localization of SAR imagery. The code is publicly available: https://github.com/Advanced-Vision-and-Learning-Lab/PaDiM-LACE.

Problem

• Detects anomalies in SAR imagery using adaptive cosine estimator
• Improves localization accuracy with bounded anomaly scores
• Evaluates performance via AUROC at image and pixel levels

Innovation

• Adaptive cosine estimator for anomaly detection
• Bounded anomaly scores via cosine similarity
• Publicly available code for SAR imagery