Article by Dr Paul Williams
Department of Food Science, Stellenbosch University
Photos provided
F
or many years scientists have relied on the
power of imaging techniques to make the invisible, visible; in so doing advancing scientific
knowledge and understanding. Imaging techniques are the future of rapid analytical analyses. Not
only do they offer accuracy and rapidity, but a number
of them also afford the ability to analyse entire samples
at once, non-destructively.
One such technique, near infrared (NIR) hyperspectral
imaging, utilises the NIR region (750-2500 nm) of the
electromagnetic spectrum. It is a combination of digital
imaging and conventional NIR spectroscopy that captures spectral and spatial information. Unlike colour
digital imaging that only captures three wavelength
channels, red, green and blue (RGB), NIR hyperspectral
imaging acquires many contiguous spectral bands of an
object.
Three instrument configurations are available i.e.
whiskbroom (point-by-point imaging), staring (focal
plane array) and push broom (line-scanning). The setup
best-suited for use in the agricultural industry, where
rapid non-destructive analyses are required, is the push
broom system. As its secondary name suggests, the
line-scanning system acquires an image line-by-line. A
typical push broom system is illustrated in Figure 1. The
samples (in this case maize kernels) are moved on a
conveyor belt into the field of view of the camera,
where it is illuminated. From there light is reflected towards the detector where a hyperspectral imaging is
constructed line-by-line. Two spatial dimensions (x and
y pixel coordinates) and a spectral dimension ( z wavelength) is acquired, forming data cubes commonly
known as hypercubes. With mathematical processing,
scientists are able to explore the nature of the samples
thus determining which chemicals are present, how
much of the chemicals are present and, importantly,
where in the sample it is located. Thus, hyperspectral
images can be considered as chemical maps, allowing
us to visualise the distribution of chemical compounds
in samples. Figure 2 illustrates how one would interpret
such a map of a maize kernel. Figure 2a is the classified
chemical image, Fig 2b a digital image and Fig 2c a diagram. Using the diagram and cross section of a maize
kernel Fig. 2c, it is possible to categorize the hyperspectral image to obtain Fig. 2a. This classification of the
different components would not have been possible if
there were no chemical differences between the structures.
Figure 1. Push broom NIR hyperspectral imaging
system. Samples are moved on a conveyor belt
into the field of view of the camera. An image is
acquired line-by-line until the entire sample has
been captured. This setup is ideal for industry as
samples are scanned in approximately 10 s.