MaLeFiSenta: Machine Learning for FilamentS Identification and orientation in the ISM

TL;DR

This paper introduces a neural network-based method combining Mask R-CNN and U-Net architectures for filament identification and orientation in the interstellar medium, reducing user dependency and parameter tuning.

Contribution

The study develops a neural network approach that accurately identifies filaments and their orientations without extensive parameter tuning, using a small training dataset.

Findings

Neural networks can effectively identify filaments in astronomical maps.

The combined Mask R-CNN and U-Net model outperforms traditional methods.

Training on only ~100 maps is sufficient for accurate filament detection.

Abstract

Filament identification became a key step to tackling fundamental problems in various fields of Astronomy. Nevertheless, existing filament identification algorithms are critically user-dependent and require individual parametrization. In this study, we aimed at adapting the neural networks approach to elaborate the best model for filament identification that would not require fine-tuning for a given astronomical map. First, we created training samples based on the most commonly used maps of the interstellar medium obtained by Planck and Herschel space telescopes and the atomic hydrogen all-sky survey HI4PI. We used the Rolling Hough Transform, a widely used algorithm for filament identification, to produce training outputs. In the next step, we trained different neural network models and discovered that a combination of the Mask R-CNN and U-Net architecture is most appropriate for filament identification and determination of their orientation angles. We showed that neural network training might be performed efficiently on a relatively small training sample of only around 100 maps. Our approach eliminates the parametrization bias and facilitates filament identification and angle determination on large data sets.