Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Anomaly detection in radio galaxy data with trainable COSFIRE filters

Steven Ndung'u, Trienko Grobler, Stefan J. Wijnholds, George Azzopardi

TL;DR

This work tackles the challenge of detecting rare, unknown anomalies in large radio-astronomy datasets by employing trainable COSFIRE filters to produce rotation-invariant morphological descriptors, followed by Local Outlier Factor scoring. The approach is semi-supervised, requiring only normal-pattern examples for training, and demonstrates superior anomaly-detection performance on the FRGADB FIRST benchmark (G-Mean ~0.79) compared with PCA+LOF and convolutional autoencoders. The method combines efficient, explainable morphology-based features with a robust outlier detector, offering scalable, fast processing suitable for next-generation telescopes like the SKA and enabling serendipitous discovery. Limitations include a relatively small dataset and a focus on a limited set of anomaly types, pointing to future work with larger, more diverse catalogs and additional rare morphologies.

Abstract

Detecting anomalies in radio astronomy is challenging due to the vast amounts of data and the rarity of labeled anomalous examples. Addressing this challenge requires efficient methods capable of identifying unusual radio galaxy morphologies without relying on extensive supervision. This work introduces an innovative approach to anomaly detection based on morphological characteristics of the radio sources using trainable COSFIRE (Combination of Shifted Filter Responses) filters as an efficient alternative to complex deep learning methods. The framework integrates COSFIRE descriptors with an unsupervised Local Outlier Factor (LOF) algorithm to identify unusual radio galaxy morphologies. Evaluations on a radio galaxy benchmark data set demonstrate strong performance, with the COSFIRE-based approach achieving a geometric mean (G-Mean) score of 79%, surpassing the 77% achieved by a computationally intensive deep learning autoencoder. By characterizing normal patterns and detecting deviations, this semi-supervised methodology overcomes the need for anomalous examples in the training set, a major limitation of traditional supervised methods. This approach shows promise for next-generation radio telescopes, where fast processing and the ability to discover unknown phenomena are crucial.

Anomaly detection in radio galaxy data with trainable COSFIRE filters

TL;DR

This work tackles the challenge of detecting rare, unknown anomalies in large radio-astronomy datasets by employing trainable COSFIRE filters to produce rotation-invariant morphological descriptors, followed by Local Outlier Factor scoring. The approach is semi-supervised, requiring only normal-pattern examples for training, and demonstrates superior anomaly-detection performance on the FRGADB FIRST benchmark (G-Mean ~0.79) compared with PCA+LOF and convolutional autoencoders. The method combines efficient, explainable morphology-based features with a robust outlier detector, offering scalable, fast processing suitable for next-generation telescopes like the SKA and enabling serendipitous discovery. Limitations include a relatively small dataset and a focus on a limited set of anomaly types, pointing to future work with larger, more diverse catalogs and additional rare morphologies.

Abstract

Detecting anomalies in radio astronomy is challenging due to the vast amounts of data and the rarity of labeled anomalous examples. Addressing this challenge requires efficient methods capable of identifying unusual radio galaxy morphologies without relying on extensive supervision. This work introduces an innovative approach to anomaly detection based on morphological characteristics of the radio sources using trainable COSFIRE (Combination of Shifted Filter Responses) filters as an efficient alternative to complex deep learning methods. The framework integrates COSFIRE descriptors with an unsupervised Local Outlier Factor (LOF) algorithm to identify unusual radio galaxy morphologies. Evaluations on a radio galaxy benchmark data set demonstrate strong performance, with the COSFIRE-based approach achieving a geometric mean (G-Mean) score of 79%, surpassing the 77% achieved by a computationally intensive deep learning autoencoder. By characterizing normal patterns and detecting deviations, this semi-supervised methodology overcomes the need for anomalous examples in the training set, a major limitation of traditional supervised methods. This approach shows promise for next-generation radio telescopes, where fast processing and the ability to discover unknown phenomena are crucial.

Paper Structure

This paper contains 9 sections, 1 equation, 2 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Data and Methodology
  3. Data set
  4. Method
  5. Evaluation Metrics
  6. Experimental Results
  7. Performance
  8. Comparison with previous work
  9. Discussion and Conclusion

Figures (2)

  • Figure 1: Examples of typical radio galaxies and both real and synthetic anomalous radio sources: FRI (edge-darkened source), FRII (edge-brightened source), XRG (x-shaped source), RRG (ring-shaped source) and synthetic sources with non-standard morphologies.
  • Figure 2: A schematic illustration of the proposed COSFIRE framework for detecting anomalous radio galaxies, showing training and inference phases of the workflow. The training phase comprises three key steps: configuring the COSFIRE filters, extracting descriptors from known radio galaxy images, and training a LOF model for anomaly detection. During the inference phase, new radio images undergo the same descriptor extraction before being evaluated by the trained LOF model to identify potential anomalies. The "Apply COSFIRE filters" and "Create descriptors" steps are shared between training and inference stages. The continuous arrows represent the training phase while dashed arrows indicate the inference phase.