How to generalize a spectral spot measurement.

Spectrum of tiles, Photon count as a function of wavelength.

Spectrum of tiles, Photon count as a function of wavelength.

The spectrometer part of the SpecCam takes 2048 samples over an interval from wavelength  250 – 1050 nm. The FOV [Field Of View] is 100 mrad high by 5 mrad wide. This “spot” sample represents one sample in 2 dim image space [ Azimuth, Elevation].

How can this spot sample be generalised over 2 dim. image space ?

Tile image with spectrometer FOV overlay.

Tile image with spectrometer FOV overlay.

Test the hypothesis that The reflection spectra in the RGB image slit subset extend beyond the slit. We test this hypothesis independently on the Red, Green, Blue channels of the camera.

Factor analysis : we measure photon_flux * integration_time = Photon_Count , with a scaling factor. Reflected_photon_flux(band) = Sun_flux(band) x cos(angle) x Reflectance(band)

 Reflectance(band)  is a material property. The factor  Sun_flux(band) is normally removed by dividing by a white reference image. The cos(angle) factor is common to all bands, it is removed by the application of the sum_norm(b):

The next image shows the sum_norm with histogram equalisation per band.

Sum_Norm over RGB, histigram equalisation per band.

Sum_Norm over RGB, histogram equalisation per band.

reflectance(band) =  Reflected_photon_flux(band)/ sum( Reflected_photon_flux(band))

Next per band thresholds are found for the [0.05 , 0.95] confidence interval from the normalised cumulative frequencies of Slit subset of RGB photon counts.

The result is a Boolean ‘Image’ for each of the bands.

In the next figure the 3 Boolean images are combined in one colour representation.

Hypothesis : same cluster in Red, Green, Blue as in Slit rectangle.

Hypothesis : same cluster in Red, Green, Blue as in Slit rectangle.

If all bands R,G,B give evidence for the Hypothesis in the [0.05- 0.95] confidence interval then the colour is white. The numerical equivalent is the and function over the 3 Boolean layers.

Connected_ness to the slit subset is evaluated by segmenting the BW_and_RGB Boolean data and selecting the segment that contains the connected subset.

In the next figure the BW_and_RGB members are shown in black.. The connected subset is outlined by its bounding box.

Black if Red,&Green&Blue   give evidence for 'same spectrum as in Slit"

Black if Red,&Green&Blue give evidence for ‘same spectrum as in Slit”

This display help the researcher in the field to select a next sample from the SpecCam.

It is to be foreseen that we will use this procedure for choosing samples during flights with a UAV.

Todo:

Radiometric and geometric calibration of the RGB camera.

Optimize the SpecCam for use in UAV, add thermal IR spot meter, add microwave reflectance sensor. Link GPS and INS (6 DOF) to database of SpecCam Images and spectra.

 

Nanno J.Mulder, M3X.

Enschede 2013-09-28