1. Introduction to Multi- and Hyperspectral Cameras

The next generation of machine vision is centered on spectral imaging technology.

Multi- and hyperspectral cameras capture wavelength information that conventional RGB cameras cannot, enabling precise analysis and advanced data interpretation. These technologies are rapidly expanding into diverse industries such as agriculture, healthcare, and semiconductor manufacturing.

In this article, we explain the principles, differences, and main application cases of the two technologies, and why spectral cameras are essential, along with key points for selecting the right one.

※ The wavelength ranges and specifications in this article are based on Avaldata camera models.

2. The Concept of Spectral Imaging in Machine Vision

Spectral imaging is a technology that analyzes the optical characteristics of objects by capturing light in different wavelength bands.
While a standard camera captures the entire visible spectrum, spectral imaging divides the spectrum into discrete bands, enabling the analysis and classification of material characteristics.

3. Principles and Structure of Hyperspectral Cameras

A hyperspectral camera detects hundreds of continuous wavelength bands using prism or grating-based spectrometer technology.
Light passes through a slit that spatially limits its incoming light, and is then dispersed by a grating. The dispersed light is projected onto the sensor.
In the case of the AHS-003VIR, spectral data can be acquired in up to 512 bands across a wavelength range of 450nm to 1700nm.

 

A hyperspectral camera typically captures only one spectrally dispersed line per scan, so to image a desired area, it must be captured using the push broom method.

The push broom method is a line-scan imaging technique commonly used in hyperspectral systems.
The camera or the object moves during the scan, capturing one line of data at a time.
As the camera or object continues to move, multiple lines are acquired to construct 2D images and 3D data.

The acquired data forms a cube, as shown in the image above, where two-dimensional images are stacked according to the number of spectral bands. This cube data is represented in a 3D cube format consisting of spatial axes (x, y) and the spectral axis (λ, wavelength).

When captured using our model AHS-003VIR (default setting), cube data is generated with a resolution of 640 (width) × 512 (height) × 512 (bands).

The graph in the upper left shows the intensity information obtained by capturing plastic (PS).

Other plastics such as nylon or PP have different spectral characteristics compared to PS, making it possible to distinguish them at specific wavelength bands.

For more detailed information on distinguishing plastic materials, please refer to the blog post titled Classification of Various Plastic Materials Using a Hyperspectral Camera.

If processing the acquired cube data is difficult for the user, various third-party spectral analysis tools (such as perClass Mira) can be used for data analysis and classification.

4. Principles and Structure of Multispectral Cameras

The structure of a multispectral camera varies depending on the manufacturer.
In the case of AVALDATA’s AMS-013VIRLF2, a filter is added on top of the sensor to allow only specific wavelength bands to pass through.

With a single capture, it can acquire data from four bands: 1200nm, 1300nm, 1450nm, and 1600nm.

Within each band, 1 to 32 lines can be captured, meaning that up to 128 lines can be acquired per scan. Similar to hyperspectral cameras, the imaging process must be performed in a push broom method.

Each frame captures four spectral bands simultaneously.

The multispectral camera viewer supports an NDVI (Normalized Difference Vegetation Index) option.

NDVI is a representative vegetation index used to quantitatively assess plant vitality (growth status or health).

NDVI enables rapid detection of pest damage, growth conditions, and water stress, and can be used to analyze crop maturity through NDVI changes.

NDVI is widely used in precision agriculture, drone-based forest monitoring, and urban green space management.

5. Differences Between Multispectral and Hyperspectral Cameras

Both multispectral and hyperspectral cameras are designed to analyze specific wavelengths of visible and non-visible light (such as infrared and ultraviolet). However, the key differences lie in the number of bands and their continuity.

  • A multispectral camera selectively detects a few (typically four) wide spectral bands. It extracts information efficiently by focusing on wavelength ranges optimized for specific applications.Since this method uses filters placed on top of the sensor to obtain wavelength information, the camera is compact and lightweight.
  • A hyperspectral camera can detect up to 1,680 continuous spectral bands (depending on the model), enabling much finer and more precise spectral analysis.One of its greatest advantages is the ability to identify subtle differences between materials that cannot be distinguished by the naked eye or standard cameras.It outputs data in a cube format and supports the ENVI file format, which can be used in compatible third-party applications.

Item RGB Camera Multispectral Camera (AMS-013VIRLF2) Hyperspectral Camera (AHS-052VIR)
Number of spectral bands 1 (R, G, B) 4 Up to 1680 (selectable)
Spectral resolution – Medium (broad bands) Very high (narrow bands)
Spectral range 400–700nm 1200, 1300, 1450, 1600nm 450–1700nm
Image acquisition Simultaneous capture Push broom Push broom
Sensor structure Bayer filter or 3-CMOS Band-pass filter Grating, prism
Data format 2D color image Limited band images 3D spectral data cube
Advantages Low cost, versatile, fast Selective spectral analysis, simpler setup High-precision analysis, fine material classification
Disadvantages No spectral data Limited spectral resolution High cost, large data size, complex processing

6. Key Applications

  • Agriculture: Crop health monitoring, moisture content analysis

  • Environmental monitoring: Marine and water quality analysis, forest surveillance

  • Medical and biomedical analysis: Skin diagnostics, blood flow monitoring

  • Food quality inspection: Fruit ripeness, meat freshness

  • Forensics: Digital evidence analysis

Note: Near-infrared (NIR) imaging is a non-destructive testing method, making it ideal for examining sensitive or high-value samples without causing damage.

7. Utilization and Future Prospects of Spectral Cameras

Spectral imaging technology is increasingly recognized in the field of machine vision, with hyperspectral cameras now actively used in automated inspection, quality control, recycling, and medical diagnostics.

The ability to detect subtle differences invisible to the naked eye or standard cameras makes them invaluable for precision analysis and automated processes.

With the integration of AI and deep learning, spectral data analysis has become more accurate and even real-time. This trend is expected to accelerate, leading to smarter and faster analysis technologies in the future.

💡 Summary
Spectral cameras allow the simultaneous acquisition of multiple wavelength bands, enabling analyses that are impossible with standard RGB cameras.
Multispectral cameras offer efficient use of selected bands, while hyperspectral cameras excel in ultra-precise full-spectrum analysis.
Choosing the right type depends on the application field and required level of detail.