Multispectral imaging
Multispectral imaging is a remote sensing technique that acquires images in multiple discrete wavelength bands of the electromagnetic spectrum, typically fewer than 20 bands, to capture spectral data in and beyond the human visible spectrum and enable the identification of materials or features based on their unique spectral signatures.[1][2] This method combines the spatial resolution of traditional imaging with the spectral selectivity of spectroscopy, allowing for the differentiation of objects or substances that appear similar in standard color photography.[3] In practice, multispectral systems use filters or sensors to isolate specific bands, often including near-infrared, ultraviolet, or thermal wavelengths, and process the data through techniques like principal component analysis to enhance contrasts and reveal hidden details.[2] The origins of multispectral imaging trace back to mid-20th-century aerial photography using natural color and color infrared films for resource mapping, evolving into digital systems with the launch of NASA's Landsat 1 in 1972, which featured the first spaceborne Multispectral Scanner for Earth observation.[4][5] Since then, advancements have included satellite instruments like the European Space Agency's Sentinel-2 Multispectral Imager, which captures 13 bands for high-resolution environmental monitoring.[6] Key applications span diverse fields, including agriculture for crop health assessment via vegetation indices, environmental science for land cover classification and water quality analysis, and medicine for noninvasive tissue diagnosis in oncology.[7][8][3] In cultural heritage, it recovers erased texts from ancient manuscripts by exploiting differences in ink absorption across wavelengths, while in remote sensing, it supports disaster management and forestry inventory.[2][7]Fundamentals
Definition and overview
Multispectral imaging (MSI) is a remote sensing technique that captures and processes images across multiple discrete spectral bands of the electromagnetic spectrum, typically 3 to 15 discrete spectral bands, which often include the visible spectrum as well as wavelengths imperceptible to the human eye.[9][10] This method exploits the unique spectral signatures—distinct patterns of reflectance, absorption, and emission at specific wavelengths—of materials to identify and analyze their composition, such as distinguishing vegetation types or soil minerals that appear indistinguishable in standard RGB imaging.[9] The origins of multispectral imaging trace back to the mid-1960s, when the U.S. Department of Agriculture (USDA), in collaboration with NASA, Purdue University, and the University of Michigan, pioneered its development for aerial photography applications in agriculture and forestry.[11] Early experiments focused on scanning systems to monitor crop health and forest resources, laying the groundwork for satellite-based implementations like the Multispectral Scanner on Landsat 1 in 1972.[11] Key advantages of MSI include its non-contact capability to detect chemical properties, such as nutrient levels in plants, moisture content in soils and crops, and concealed features like subsurface anomalies or stress indicators invisible to the naked eye.[12][13] The basic workflow begins with sensor acquisition of raw data in individual bands, followed by preprocessing steps like radiometric correction and geometric alignment, culminating in band combination to generate composite images for enhanced visualization and interpretation.[14]Underlying principles
Multispectral imaging relies on the principles of electromagnetic radiation, where light is characterized as waves spanning the electromagnetic spectrum. In this context, the relevant portion is the optical domain, typically ranging from ultraviolet (UV) at approximately 0.4 μm to short-wave infrared (SWIR) at 2.5 μm, encompassing visible light (0.4–0.7 μm) and near-infrared (NIR) regions.[15] This range allows imaging systems to capture interactions beyond human vision, exploiting variations in how materials respond to different wavelengths.[16] Materials interact with incident light through reflection, absorption, or transmission, each process varying distinctly across wavelengths and producing unique spectral signatures that enable material identification. For instance, vegetation may reflect strongly in the NIR due to cell structure while absorbing in visible bands for photosynthesis, whereas water absorbs NIR efficiently.[17] These signatures arise from molecular and atomic properties, such as electronic transitions in UV-visible or vibrational modes in IR, allowing multispectral imaging to differentiate substances based on band-specific responses.[18] The core quantitative measure is spectral reflectance, defined as R(\lambda) = \frac{E_r(\lambda)}{E_i(\lambda)} where E_r(\lambda) is the reflected irradiance and E_i(\lambda) is the incident irradiance at wavelength \lambda. In multispectral systems, this is approximated by integrating or averaging over discrete spectral bands, yielding band reflectance values that quantify the material's response and form the basis for signature analysis.[19] Atmospheric effects significantly influence multispectral data, primarily through scattering and absorption that alter the observed signal. Rayleigh scattering predominates in shorter wavelengths (e.g., blue light), causing haze, while Mie scattering from aerosols affects longer wavelengths; absorption by gases like water vapor and CO2 creates spectral "windows" and attenuation, particularly in IR bands. Multispectral imaging incorporates corrections, such as radiative transfer models, to mitigate these path radiance and attenuation effects for accurate surface reflectance retrieval.[20][21]Spectral Characteristics
Common spectral bands
Multispectral imaging systems typically employ a limited number of discrete spectral bands, often 4 to 12, selected from the ultraviolet, visible, near-infrared (NIR), short-wave infrared (SWIR), and occasionally thermal infrared portions of the electromagnetic spectrum. The visible bands, spanning approximately 0.4 to 0.7 μm, provide color information similar to human perception by capturing blue (0.45–0.52 μm), green (0.52–0.60 μm), and red (0.63–0.69 μm) wavelengths.[22][23] The NIR band, generally 0.7 to 1.1 μm, is essential for vegetation analysis due to high reflectance from healthy plant canopies in this range, contrasting with lower visible reflectance.[22] The SWIR band, from 1.1 to 2.5 μm, facilitates identification of minerals, soil types, and moisture levels through distinct absorption features in this region.[22] Optional ultraviolet (UV, 0.3–0.4 μm) bands support aerosol and atmospheric studies, while thermal infrared (8–14 μm) bands enable surface temperature mapping.[22] Band selection criteria prioritize the unique reflectance or absorption signatures of target materials, ensuring discrimination between features of interest. For example, red and NIR bands are chosen to exploit chlorophyll's strong absorption in the red (~0.65 μm) for photosynthesis and high NIR reflectance from internal leaf scattering, allowing assessment of vegetation vigor.[24] Additionally, bands are positioned within atmospheric windows—transparent regions with minimal absorption by gases such as water vapor, carbon dioxide, and ozone—to reduce interference and maximize signal quality. Prominent windows include 0.3–1.1 μm for UV-visible-NIR imaging and 1.1–2.5 μm for SWIR, avoiding strong absorption lines around 1.4 μm and 1.9 μm due to water vapor.[25] A representative configuration is found in the Landsat satellite series, which uses seven core multispectral bands for Earth observation, as detailed in the table below:| Band Number | Name | Wavelength Range (μm) | Primary Use |
|---|---|---|---|
| 2 | Blue | 0.45–0.51 | Water quality, aerosols |
| 3 | Green | 0.53–0.59 | Vegetation, bathymetry |
| 4 | Red | 0.64–0.67 | Chlorophyll absorption |
| 5 | NIR | 0.85–0.88 | Vegetation health |
| 6 | SWIR 1 | 1.57–1.65 | Soil moisture, vegetation |
| 7 | SWIR 2 | 2.11–2.29 | Mineral mapping, snow/ice |
| 10/11 | Thermal IR | 10.60–12.51 | Surface temperature |