SpectraGeni: Real and Synthetic Hyperspectral Signatures for Precision Agriculture

A dataset of 943-band hyperspectral spectral signatures for precision agriculture, containing both real field-collected and model-generated synthetic data. It was created by Manoj Kaushik to support the SpectraGeni framework, a convolutional conditional variational autoencoder for data generation under class imbalance. The data is provided in Parquet format and was last updated on 2026-04-19.

Use Cases

• Training generative models to address class imbalance based on the synthetic data generation framework.
• Evaluating the quality of synthetic hyperspectral signatures against real field-collected spectra.
• Developing classification algorithms for agricultural targets using 943-band spectral data.
• Reproducing experiments for the SpectraGeni (Conv-CVAE) deep learning model.

Strengths

• Contains 943 spectral bands per sample, providing high-dimensional spectral detail.
• Includes both real field-collected and model-generated synthetic data, enabling comparative analysis.
• Data is provided in Parquet format, which is noted for accessibility and efficiency.
• Released under a permissive CC-BY-4.0 license, facilitating reuse and sharing.

Limitations

• Row count is unknown, which may limit suitability assessment for large-scale training.
• Column-level documentation is absent; field semantics must be inferred after download.
• The description does not specify the geographic or temporal coverage of the real field data.

Provenance

Source

Manoj Kaushik via figshare.

Collection Method

Real data was field-collected; synthetic data was generated by a trained Conv-CVAE model.

Freshness

Last updated 2026-04-19 17:01:36; freshness should be verified.