China Offshore SSS-AI: A multi-region side-scan sonar image dataset for cross-regional seafloor target recognition

China Offshore SSS-AI is a public, standardized image-only side-scan sonar (SSS) dataset for seafloor target recognition. The dataset contains 3,255 cropped SSS image chips from representative sectors of the Chinese marginal seas, including the Bohai Sea, Yellow Sea, Taiwan Strait and northern South China Sea sectors.

The release contains standardized image pixels and non-sensitive derived file-level metadata only. It does not include raw side-scan sonar records, precise coordinates, survey-line identifiers, acquisition settings, original project folders or project-sensitive engineering records.

The package includes:1. images/: the complete image-only release;2. metadata/: image manifest, class mapping, benchmark split, quality-control records and metadata dictionary;3. README.md: package structure, sensitive-information policy and recommended use.

The recommended benchmark subset contains 3,241 images and is defined by metadata/train_val_test_split.csv. The complete release retains all 3,255 images, including 14 release-only images excluded from the main benchmark because of rare or under-populated source-region classes.

Associated code for dataset checking and baseline validation is publicly available at:https://github.com/Lorddu/China-Offshore-SSS-AI-code

Users should cite the Zenodo dataset record and the associated Scientific Data article when reusing the dataset.