Mads Nibe Larsen

PhD student working with hyperspectral thermal imaging

Updated:

less than 1 minute read

Two additional lenses was placed in front of the Fabry-Pérot to “colimate” the light in order to reduce the effects of spectral bending. A hyperspectral data cube of a 60 °C hotplate has been acquired both with and without compensating optics. SNV has been applied to the data cubes and a single layer of each are presented in Fig. 1.

Bloom spot on bolometer after 5 minutes
Fig 1

Interferograms are extracted and presented in Fig. 2 showing the effect of spectral bending being reduced. The central interferogram is the mean of 40×40 pixels, while the “edge” interferogram is the mean of 20×20 pixels.

Bloom spot on bolometer after 5 minutes
Fig 2

However, the image quality is also dramatically reduced as seen in Fig. 3 which show the same layers of the data cube as in Fig. 1, but only mean centered.

Bloom spot on bolometer after 5 minutes
Fig 3

Categories:

Updated:

You may also enjoy

Flame experiments

Updated:

Hyperspectral images of flames have been acquired using the hyperspectral LWIR camera. The hope is to be able to distinguish between different kinds of fuel and identify different reaction regimes within the flames. The fuels are not premixed, meaning pure fuel is expelled from the nozzle where it then mixes with the air in order to maintain a sustaining flame (Fig. 1). Only simple and pure and fuels are used in the hope that the reactions (and hereby the spectral components) are simpler. Read more

Heated carbon composite samples

Updated:

The heating characteristics of six different carbon composite samples have been investigated by time series thermography. As depicted in Fig. 1, the samples are suspended in front of the thermal camera (not hyperspectral), and heated by a heat gun for 30 s after which the cooling is observed for 2 minutes. Initial experiments using a hot plate to heat one end of the samples were not suitable to detect the defect. This might be caused by the samples dissipate most of the heat energy before it can travel from the heated end to the defect. Read more

Summary of NMF-based reconstructions

Updated:

This is going to be a broader overview of my different efforts in the attempt to reconstruct the incident spectra based on the interferograms recorded by the hyperspectral camera. The setup has been described in further detail previously_ so here is just a very short introduction of the problem in terms of an $\mathbf{AX}= \mathbf{B}$ problem. The system matrix, $\mathbf{A}$ describes the transmission through the Fabry-Pérot interferometer at different combinations of wavelengths and mirror separations. A sensor response has been fitted an multiplied with each row of $\mathbf{A}$ (each row represent the transmission spectrum in terms of wavelengths/wavenumbers at a specific mirror separation). Each column of $\mathbf{X}$ contains the incident spectrum of the sample. This spectrum is a combination of the light source being transmitted through the material sample along with a reflection component coming from the environment. Both the transmission and reflection measurements are obtained by FTIR spectroscopy (not ATR, but two separate measurements). Finally, $\mathbf{B}$ contains the measured interferograms from the camera. However it is not the raw measurements, but a term including the reflection of imaging sensor off the backside of the scanning Fabry-Pérot interferometer (SFPI) has also been included. Read more

Estimating sensor response

Updated:

Using fast nonnegative least squares (FNNLS)

I have previously had some luck with with using fast nonnegative least squares (FNNLS) to estimate the sensor response based on empirical measurements. For this I need for formulate an $\mathbf{Ax} = \mathbf{b}$ problem for the solver. It has previously been described how to set up this problem, so I will not go into much detail here_. Just know that the $\mathbf{A}$ matrix expresses the known (estimated from FTIR) net flux (incident minus emitted), $\mathbf{b}$ is the measured interferogram (without offset - we’ll discuss that later), and $\mathbf{x}$ is the sensor response we need to solve for. Read more