Skip to content

Latest commit

 

History

History
73 lines (58 loc) · 4.4 KB

Readme.md

File metadata and controls

73 lines (58 loc) · 4.4 KB
Raw

Oceanship: A Large-Scale Dataset for Underwater Audio Target Recognition

oceanship

The recognition of underwater audio plays a significant role in identifying a vessel while it is in motion. Classifying underwater vessels involves training a network to extract features from audio data and predict the vessel type. The current UATR dataset exhibits shortcomings in both duration and sample quantity. In this paper, we propose Oceanship4, a large-scale and diverse underwater audio dataset. This dataset comprises 15 categories, spans a total duration of 65 hours, and includes comprehensive annotation information such as coordinates, velocity, vessel types, and timestamps. We compiled the dataset by crawling and organizing original communication data from the Ocean Communication Network (ONC5) database between 2021 and 2022. While audio retrieval tasks are well-established in general audio classification, they have not been explored in the context of underwater audio recognition. Leveraging the Oceanship dataset, we introduce a baseline model named Oceannet for underwater audio retrieval. This model achieves a recall at 1 (R@1) accuracy of 67.11% and a recall at 5 (R@5) accuracy of 99.13% on the Deepship dataset.

Oceanship are available here:

Way to extract wav files from multiple multi-compressed files

  • The original downloaded files should be formatted like this:
ocean_dataset_archive.aa  ocean_dataset_archive.af  
ocean_dataset_archive.ab  ocean_dataset_archive.ag 
ocean_dataset_archive.ac  ocean_dataset_archive.ah
ocean_dataset_archive.ad  ocean_dataset_archive.ai 
ocean_dataset_archive.ae  ocean_dataset_archive.aj
  • Then please run the unzip code below:
cat ocean_dataset_archive.aa ocean_dataset_archive.ab ocean_dataset_archive.ac ocean_dataset_archive.ad ocean_dataset_archive.ae ocean_dataset_archive.af ocean_dataset_archive.ag ocean_dataset_archive.ah ocean_dataset_archive.ai ocean_dataset_archive.aj > ocean_dataset_archive.tar
tar -xvf ocean_dataset_archive.tar
  • Next step, move these files to your desired location
mkdir /path to your file/oceanship
find /path to your file/cluster/home/lizeyu/oceandil/dataset/ocean_dataset/v100_preprocessed_89_09_31/ -type f -exec mv {} /path to your file/oceanship/ \;
  • Finally, clean up the useless folders
rm -rf /path to your file/cluster
rm -rf /path to your file/ocean_dataset_archive*

Data-fetching

  • Warning: Make sure your disk have more than enough storage space to save these raw files;
# starting to download and preprocess
cd way_to_download_from_onc
bash jack.sh
  • The download speed depends on the internet speed and the pre-processing speed depends on the CPU. therefore, it is recommended to use a high performance computing device for crawling;
  • The downloaded results will be found in way_to_download_from_onc/underwater after running;

Enhanced information based on MMSI

I have obtained the MMSI information. By crawling shipsfind艘船网, I was able to get more details about the ship itself, which contains f"[{formatted_time}],{ais_mmsi_elements},{ais_callsign_elements},{ais_heading_elements},{ais_course_elements},{ais_imo_elements},{ais_sog_elements} elements},{ais_course_elements},{ais_imo_elements},{ais_sog_elements},{ais_shipType_elements},{ais_lon_elements},{ais_lat_elements },{ais_length_elements},{ais_width_elements},{ais_draught_elements},{ais_dest_elements},{ais_eta_elements},{ais_lastTime_elements} {ais_draught_elements},{ais_dest_elements},{ais_eta_elements},{ais_lastTime_elements} \n". You can find these two files from this repo with a ".txt" suffix.

Acknowledge

Thanks for the perfect work of Deepship and Shipsear authors. Also, thank you to the Ocean Networks Canada - Oceans 3.0 database for supporting the UATR data. Finally, thank you, Lucas, for open-sourcing the code (onc_dataset) for crawling underwater audio data from ONC, our work is based on this foundation.