BJTU-UVA: The First Dataset for Automatic Spectral Calibration of Hyperspectral Images

We are proud to introduce <b>BJTU-UVA</b>, the <b>first dataset designed specifically for the task of automatic spectral calibration</b> of hyperspectral images (HSIs). This dataset addresses the critical challenge of minimizing illumination variability without relying on manual intervention or physical references.Key Highlights<b>Task Proposal</b>:<br>We propose the novel task of <b>automatic spectral calibration</b>, aiming to advance the robustness of hyperspectral imaging in diverse real-world scenarios.<b>Dataset Characteristics</b>:<b>Camera</b>: Specim IQ, featuring a spectral resolution of 3nm across the 400–1000nm range.<b>Recording Method</b>: Each scene is captured twice:<b>Without reference board</b>: Captures raw scene data.<b>With white reference board</b>: Records illumination conditions under the same settings.<br>This approach ensures asynchronous yet precise pairing of <b>uncalibrated</b> and <b>calibrated</b> HSIs, effectively minimizing illumination variability.<b>Dark Current Correction</b>: Dark current noise, intrinsic to the camera sensor, is carefully recorded and subtracted during post-processing, ensuring high data accuracy.<b>Scene Diversity</b>:<br>The dataset encompasses a wide range of <b>urban and natural scenes</b>, captured under various weather conditions, lighting scenarios, and times of day.<b>Benchmarking Standard</b>:<br>BJTU-UVA establishes a new standard for spectral calibration by combining real-world scene variability with rigorous illumination recording, offering a robust foundation for testing and advancing spectral calibration techniques.Citation@misc{du2024spectral,<br>title={Automatic Spectral Calibration of Hyperspectral Images: Method, Dataset and Benchmark},<br>author={Zhuoran Du and Shaodi You and Cheng Cheng and Shikui Wei},<br>year={2024},<br>eprint={2412.14925},<br>archivePrefix={arXiv},<br>primaryClass={cs.CV},<br>url={ https://arxiv.org/abs/2412.14925 },<br>}