VR4QC Composite Seismic PSD Dataset for Anomaly Detection and Classification

A composite seismic power spectral density (PSD) benchmark dataset supports the study 'From Virtual to Real: Generative Learning for Feature Signal Detection and Classification of Seismic Data.' It contains training, validation, and test subsets for 12 classes, including 1 background-noise and 11 anomaly classes. The dataset was created by XuChao Chai and last updated on 2026-04-28.

Use Cases

• Benchmarking PSD-based seismic signal classification workflows based on the described training, validation, and test subsets.
• Evaluating model generalization from synthetic/curated data to real operational seismic observations based on the dataset's composition.
• Anomaly detection in seismic signals based on the 11 anomaly classes included in the data.
• Training generative models for seismic data based on the inclusion of expert hand-drawn and model-generated PSDs.

Strengths

• The training set contains 8,000 generated samples and 800 hand-drawn samples per class.
• The validation set contains 2,000 generated samples and 200 hand-drawn samples per class.
• The test set contains more than 2 million real-world PSD samples.
• Data is organized for reproducible benchmarking across 12 distinct classes.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the full dataset is unknown, which may limit suitability assessment.
• Data may reflect temporal or source bias inherent to the specific seismic observations used.

Provenance

Source

figshare

Collection Method

Composite dataset including expert hand-drawn PSDs, model-generated PSDs, and large-scale real-world PSDs.

Freshness

Last updated 2026-04-28 07:42:11; freshness should be verified.