Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Automatic Spectral Calibration of Hyperspectral Images:Method, Dataset and Benchmark

Zhuoran Du, Shaodi You, Cheng Cheng, Shikui Wei

TL;DR

This work tackles automatic spectral calibration of hyperspectral images under diverse natural illumination by proposing a learning-based method, the Spectral Illumination Transformer (SIT). It introduces the BJTU-UVA dataset and its 10-illumination expansion BJTU-UVA-E to enable large-scale, real-world calibration benchmarks, and demonstrates state-of-the-art performance across full-spectrum and 31-channel HSIs. Central to SIT is the Illumination Attention and Spectral Illumination Attention that jointly model scene radiance and global illumination, yielding robust calibration even in challenging low-light and infrared regions. The dataset and method together push forward practical, non-occluding HSI calibration with broad implications for remote sensing and laboratory imaging.

Abstract

Hyperspectral image (HSI) densely samples the world in both the space and frequency domain and therefore is more distinctive than RGB images. Usually, HSI needs to be calibrated to minimize the impact of various illumination conditions. The traditional way to calibrate HSI utilizes a physical reference, which involves manual operations, occlusions, and/or limits camera mobility. These limitations inspire this paper to automatically calibrate HSIs using a learning-based method. Towards this goal, a large-scale HSI calibration dataset is created, which has 765 high-quality HSI pairs covering diversified natural scenes and illuminations. The dataset is further expanded to 7650 pairs by combining with 10 different physically measured illuminations. A spectral illumination transformer (SIT) together with an illumination attention module is proposed. Extensive benchmarks demonstrate the SoTA performance of the proposed SIT. The benchmarks also indicate that low-light conditions are more challenging than normal conditions. The dataset and codes are available online:https://github.com/duranze/Automatic-spectral-calibration-of-HSI

Automatic Spectral Calibration of Hyperspectral Images:Method, Dataset and Benchmark

TL;DR

This work tackles automatic spectral calibration of hyperspectral images under diverse natural illumination by proposing a learning-based method, the Spectral Illumination Transformer (SIT). It introduces the BJTU-UVA dataset and its 10-illumination expansion BJTU-UVA-E to enable large-scale, real-world calibration benchmarks, and demonstrates state-of-the-art performance across full-spectrum and 31-channel HSIs. Central to SIT is the Illumination Attention and Spectral Illumination Attention that jointly model scene radiance and global illumination, yielding robust calibration even in challenging low-light and infrared regions. The dataset and method together push forward practical, non-occluding HSI calibration with broad implications for remote sensing and laboratory imaging.

Abstract

Hyperspectral image (HSI) densely samples the world in both the space and frequency domain and therefore is more distinctive than RGB images. Usually, HSI needs to be calibrated to minimize the impact of various illumination conditions. The traditional way to calibrate HSI utilizes a physical reference, which involves manual operations, occlusions, and/or limits camera mobility. These limitations inspire this paper to automatically calibrate HSIs using a learning-based method. Towards this goal, a large-scale HSI calibration dataset is created, which has 765 high-quality HSI pairs covering diversified natural scenes and illuminations. The dataset is further expanded to 7650 pairs by combining with 10 different physically measured illuminations. A spectral illumination transformer (SIT) together with an illumination attention module is proposed. Extensive benchmarks demonstrate the SoTA performance of the proposed SIT. The benchmarks also indicate that low-light conditions are more challenging than normal conditions. The dataset and codes are available online:https://github.com/duranze/Automatic-spectral-calibration-of-HSI

Paper Structure

This paper contains 38 sections, 20 equations, 10 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Hyperspectral Image Calibration
  4. RGB White Balancing
  5. HSI Restoration and Reconstruction
  6. HSI Calibration
  7. HSI Illumination Calibration
  8. Synchronous method
  9. Asynchronous method
  10. Learning-based method (proposed)
  11. BJTU-UVA Dataset
  12. BJTU-UVA Dataset
  13. BJTU-UVA-E Dataset
  14. 31 Channels Subset
  15. Specifications
  16. ...and 23 more sections

Figures (10)

  • Figure 1: Comparison of HSI Calibration Methods: (a) Synchronous, (b) Asynchronous, and (c) Automatic.
  • Figure 2: Overview of the BJTU-UVA Dataset: (a) Dataset Recording Setup, (b) Variety of Scenes, (c) Illumination variety, and (d) HSI Camera Specifications and Dataset Statistics
  • Figure 3: Measurement of Ten Different Illuminations Using a Whiteboard: Five Natural Conditions (Sunny, Cloudy, Rainy, Evening, and Shadow) and Five Color-Filtered Conditions (Red, Blue, Yellow, Purple, and Green)
  • Figure 4: Structure of Spectral Illumination Transformer (SIT) framework and its unit (SIT-U) with Spectral and Illumination Attention branches for improved HSI calibration. The Illumination Attention branch mimics the Gray-World method to capture illumination features.
  • Figure 5: Visual Comparison of Absolute Error Using Heat Maps: The first two columns show the RGB rendering of the input and ground truth (GT) for visualization. Rows 1-3 display samples from BJTU-UVA, and Rows 4-6 show samples from BJTU-UVA-E.
  • ...and 5 more figures