/RadSeg

RadSeg is a synthetic radar pulse activity dataset submitted to the proceedings of the 2024 IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP 2024).

MIT LicenseMIT

Radar Segmentation Dataset (RadSeg)

RadSeg is a synthetic radar dataset designed for building semantic segmentation models for radar activity recognition. Unlike existing radio classification datasets that only provide signal-wise annotations for short and isolated IQ sequences, RadSeg provides sample-wise annotations for interleaved radar pulse activities that extend across a long time horizon. This makes RadSeg the first annotated public dataset of its kind for radar activity recognition. This dataset is released to the public under the MIT License.

Huang, Zi, Akila Pemasiri, Simon Denman, Clinton Fookes, and Terrence Martin. "Multi-stage Learning for Radar Pulse Activity Segmenation" In Proceedings of the 2024 IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP 2024) (under review).

Dataset Details

RadSeg contains pulsed radar signals at varying signal-to-noise ratios (SNRs) between -20 to 20 dB with a resolution of 0.5 dB. This repository provides the RadSeg dataset which consists of three parts:

  • RadSeg-Train contains 60,000 radar signals for model training;
  • RadSeg-Valadation contains 10,000 radar signals for model validation; and
  • RadSeg-Test contains 10,000 radar signals held out for testing.

This dataset comprises a total of 5 radar signal types, which include:

  • Barker codes, up to a code length of 13;
  • Polyphase Barker codes, up to a code length of 13;
  • Frank codes, up to a code length of 16;
  • Linear frequency-modulated (LFM) pulses; and
  • Coherent unmodulated pulse trains.

Additional dataset characteristics:

  • The sampling rate used in RadSeg is 3.2 MHz.
  • Each radar signal contains 32,768 complex, baseband IQ samples.
  • Annotations are provided as channel-wise binary masks where each channel corresponds to a signal type.

Please refer to the associated conference paper (TBA) for further details on RadSeg.

Download Links

The RadSeg datasets can be downloaded from the following links:

Raw IQ Data

Segmentation Masks (Channel-wise Annotations)

SNR Labels

Citation

Please cite both the dataset and the conference paper (TBA) if you find them helpful for your research.

@inproceedings{huang2024radseg,
  title={Multi-stage Learning for Radar Pulse Activity Segmenation},
  author={Huang, Zi and Pemasiri, Akila and Denman, Simon and Fookes, Clinton and Martin, Terrence},
  booktitle={Proceedings of the 2024 IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP 2024) (under review)},
  year={2024},
}