RSyntHyperPDID: Remote Sensing Synthetic Hyperspectral Paired Dehazing Image Dataset

We created a synthetic dataset, RSyntHyperPDID, based on the AVIRIS database. The dataset is sufficiently large to serve as a training set for hyperspectral dehazing models.

Official Dataset for HyperHazeOff: Hyperspectral Remote Sensing Image Dehazing Benchmark. MDPI, Preprints.org.

Official Code: GitHub.

RGB/

For each split part (train/test) and clear hyperspectral image, two types of RGB visualizations are provided:

• train/test/clear/

  • CSNC/ — RGB images synthesized using narrowband spectral reconstruction (.png format)
  • CSSO/ — RGB images synthesized via CIE XYZ → sRGB transformation using the standard observer (.png format)

Note: CSNC and CSSO contain the same RGB data with different color synthesis methods

Hyperspectral Images (HSI)

The RSyntHyperPDID dataset contains hyperspectral images of size 256×256 pixels with 182 spectral bands.

It is divided into three categories: clear, train, and test.

Each train and test sample includes hyperspectral images with synthetic haze, created based on a corresponding clear ground truth (GT) image.

The naming convention ensures the link between hazy and clear images:

• for example, clear/XXXX.npy serves as the GT for train/test samples named XXXX_YYYY.npy,
• where XXXX is the GT image identifier and YYYY represents haze synthesis parameters.

This naming scheme ensures straightforward pairing of hazy images with their clear counterparts for training and evaluation. It supports reproducible experiments in haze removal and spectral image restoration.

Additionally, wavelength information for the hyperspectral bands is provided in the file wavelengths_synthyper.npy, supporting spectral analysis across the dataset.

Dataset Structure

RSHPDID
|-- rgb
|   |-- clear
|   |   |-- CSNC
|   |   |   |-- 0001.png
|   |   |   |-- ........
|   |   |   `-- 0220.png
|   |   `-- CSSO
|   |       |-- 0001.png
|   |       |-- ........
|   |       `-- 0220.png
|   |-- test
|   |   |-- CSNC
|   |   |   |-- 0010_0030.png
|   |   |   |-- .............
|   |   |   `-- 0219_4040.png
|   |   `-- CSSO
|   |       |-- 0010_0030.png
|   |       |-- .............
|   |       `-- 0219_4040.png
|   `-- train
|       |-- CSNC
|       |   |-- 0001_0030.png
|       |   |-- .............
|       |   `-- 0220_4040.png
|       `-- CSSO
|           |-- 0001_0030.png
|           |-- .............
|           `-- 0220_4040.png
|-- clear
|   |-- 0001.npy
|   |-- ........
|   `-- 0220.npy
|-- test
|   |-- 0010_0030.npy
|   |-- .............
|   `-- 0219_4040.npy
|-- train
|   |-- 0001_0030.npy
|   |-- .............
|   `-- 0220_4040.npy
`-- wavelengths_synthyper.npy

Data Source

Data derived from AVIRIS imagery. Courtesy NASA/JPL-Caltech.

Citation

@Article{jimaging11120422,
AUTHOR = {Nikonorov, Artem and Sidorchuk, Dmitry and Odinets, Nikita and Volkov, Vladislav and Sarycheva, Anastasia and Dudenko, Ekaterina and Zhidkov, Mikhail and Nikolaev, Dmitry},
TITLE = {HyperHazeOff: Hyperspectral Remote Sensing Image Dehazing Benchmark},
JOURNAL = {Journal of Imaging},
VOLUME = {11},
YEAR = {2025},
NUMBER = {12},
ARTICLE-NUMBER = {422},
URL = {https://www.mdpi.com/2313-433X/11/12/422},
ISSN = {2313-433X},
ABSTRACT = {Hyperspectral remote sensing images (HSIs) provide invaluable information for environmental and agricultural monitoring, yet they are often degraded by atmospheric haze, which distorts spatial and spectral content and hinders downstream analysis. Progress in hyperspectral dehazing has been limited by the absence of paired real-haze benchmarks; most prior studies rely on synthetic haze or unpaired data, restricting fair evaluation and generalization. We present HyperHazeOff, the first comprehensive benchmark for hyperspectral dehazing that unifies data, tasks, and evaluation protocols. It comprises (i) RRealHyperPDID, 110 scenes with paired real-haze and haze-free HSIs (plus RGB images), and (ii) RSyntHyperPDID, 2616 paired samples generated using a physically grounded haze formation model. The benchmark also provides agricultural field delineation and land classification annotations for downstream task quality assessment, standardized train/validation/test splits, preprocessing pipelines, baseline implementations, pretrained weights, and evaluation tools. Across six state-of-the-art methods (three RGB-based and three HSI-specific), we find that hyperspectral models trained on the widely used HyperDehazing dataset fail to generalize to real haze, while training on RSyntHyperPDID enables significant real-haze restoration by AACNet. HyperHazeOff establishes reproducible baselines and is openly available to advance research in hyperspectral dehazing.},
DOI = {10.3390/jimaging11120422}
}

License

This dataset is released under license: cc-by-4.0