Computational Imaging for Scene Understanding

4Hyperspectral Imaging

Nathan HAGEN

Department of Optical Engineering, Utsunomiya University, Japan

4.1. Introduction

Spectral imaging systems measure the spectral irradiance I(x, y, λ) of a scene, collecting a 3D dataset typically called a datacube (see Figure 4.1). Since datacubes are of higher dimensionality than the 2D detector arrays available, system designers must either measure time-sequential 2D slices of the cube, or divide the datacube into multiple 2D elements that can be recombined into a cube in post-processing. These are typically described here as scanning and snapshot approaches.

In this chapter, we give a broad survey of scanning and snapshot instrumentation for doing spectral imaging and explain the advantages and disadvantages of various techniques, in order to explain why system designers choose different techniques depending on the needs of each application.

The terms spectral imaging, imaging spectrometry (or imaging spectroscopy), hyperspectral imaging and multispectral imaging used to describe the field are often used interchangeably. However, we generally find that in applications where the imaging aspect of the measurement is the focus of attention, authors generally prefer (multi-, hyper-)spectral imaging, whereas in applications where the spectrum is the focus of attention, imaging spectrometry/spectroscopy is more common.

Figure 4.1. The portions ...