NK‑Sonar‑Image‑Dataset (NKSID) Overview

Basic Information

• Name: NanKai Sonar Image Dataset (NKSID)
• Number of Classes: 8
• Number of Images: 2,617
• Collection Location: Bohai Bay ($39^{\circ},\text{N},118^{\circ},\text{E}$)
• Collection Tool: ROV equipped with Oculus M750d multibeam forward‑looking sonar
• Image Features: Targets attached to buoys by ropes, suspended ~5–10 m below the surface, captured from different angles, distances (2–15 m) and frequencies (750 kHz, 1.2 MHz)
• Processing: Target selection, preprocessing, and annotation

Usage

• Download & Extract: Download all .zip files from the repository and extract them; each class is in a separate archive
• File Description:
  • train_abs.txt: Relative path and label for each image
  • kfold_train.txt and kfold_val.txt: Random 10‑fold cross‑validation train/validation splits, where $n$ denotes the sample index corresponding to line $n$ in train_abs.txt

Example Application

• Example Repository: Jorwnpay/Sonar‑OLTR (github.com) demonstrates open‑set long‑tail recognition using this dataset

Citation

• Paper Citation:

  @article{jiao2024open,
    title={Open‑set recognition with long‑tail sonar images},
    author={Jiao, Wenpei and Zhang, Jianlei and Zhang, Chunyan},
    journal={Expert Systems with Applications},
    pages={123495},
    year={2024},
    publisher={Elsevier}
  }