Slow Scan Hyper Spectral Imaging of Storvola (77.5°N,16.2°E): preliminary results.
Fred Sigernes 1, Bjørn Sæther 2 , Ståle Johansen 3 , Karsten Heia 4, Arve Kylling 5 and Yngvar Gjessing 1
1 The University Courses on
Svalbard, N-9171 Longyearbyen, Norway;
2 Statoil R&T, N-7005 Trondheim, Norway;
3 Norwegian University of Science and Technology, N-7491 Trondheim, Norway;
4 Norwegian Institute of Fisheries and Aquaculture, 9005 Tromsø, Norway;
5 Norwegian Institute for Air Research, N-2027 Kjeller, Norway.
Slow scan spectral imaging in the wavelength range 380 - 970nm of the bed rock layers of Storvola (77.5°N,16.2°E) are presented. The main physical principle of the technique is outlined. The instrument is capable of identifying rock signatures of the mountain side with high spectral and spatial resolution, simultaneously. The spatial resolution is 2.4 x 0.8 m and the spectral bandpass is in the range 1 - 2 nm. Shale and sandstone directional reflection factors are calculated. The reflection factors are slowly increasing in the visible and exponentially in the near infrared region of the spectrum. Reflection ratios show that the instrument can distinguish the spectral difference between sandstone layers. It can also detect shale from sandstone. The experiment have provided basic instrumental parameters for the design of an airborne instrument.
A field experiment has been conducted to measure the spectral reflection factors of the mountain bed rock layers of Storvola (77.5°N,16.2°E) at Spitsbergen, Norway. The location is well known by geologist. Throughout the last decade, numerous campaigns have been launched at the site to study the geological history and to train students in field work. The exposures in Storvola represent part of a geological cross section through the Central Basin on Spitsbergen. This Basin was a relatively small foreland basin formed in front of the fold-and-thrust belt (e.g. Harland, 1965). The latest Paleocene-early Eocene (approximately 58-48 mill. Years before present) infilling of the basin progressed from west to east, and left a spectacular record of large scale shallow-to-deep-water clinoforms reflecting the overall progradation of varied sedimentary facies (Steel et al., 1985). These clinoforms are now exposed along the mountainside. The general aspect of the clinoforms are well described by Helland-Hansen (1992). The basin-infill face of interest here consists of the Gilsonryggen Formation (mainly marine shales) and the Battfjellet formation (mainly delta-front and shoreline sandstones). In Figure 5 the sandstones belong to the Battfjellet formation while the shales belong to the Gilsonryggen formation.
The aim and motivation for the experiment is to find a new method to map the bedrock layers of Spitsbergen by the use of remote sensing techniques. An airborne spectral imager with a view angle of 45° to the vertical should in theory be able to map rock layers of the mountain sides with both high speed and resolution. In order to construct and optimize a future airborne imager, a slow scan spectral imager was selected to probe Storvola. On the 11'th of August 2001 the instrument was transported by helicopter to the site and installed approximately 800 m from the mountain side.
The instrument is able to image static objects as a function of wavelength (Sigernes et al., 2000). The idea is that the multi-channel output of the instrument, should provide us with the spectral characteristics representing the different features of each layer. This type of data is ideal as input to image processing classification routines such as the Bayes method (Niblack, 1986). The technique is known as hyper spectral imaging and is used extensively from both air- and space born platforms as well as in industrial applications. A great deal of experience has been gained with these instruments and they have proven to be powerful tools in remote sensing (c.f. Vane, 1987; Wolf, 1997; Herrala et al., 1996; Hyvarinen et al., 1998).
The instrumental technique and data obtained by the instrument are presented. Finally, the results are summarized in order to prepare and meet the requirements for an airborne experiment.
2. INSTRUMENTAL TECHNIQUE
2.1 Main principle
The optical diagram of the instrument is shown in Figure 1. The main idea behind spectral imaging of objects may be described as follows: Firstly, light reflected off the target (2) must be focused by a lens (4) to form an image at the spectrographs entrance slit plane (5). The resulting spectrogram (10) is the intensity distribution as a function of wavelength and position along the slit. The diffracted slit image contains both spectral and spatial information along a thin track of the target.
Secondly, in order to obtain the target's full spatial extent, it is necessary to sample the whole target area. This requires the use of a high resolution rotary element. The whole idea is to record spectrograms for each track of the target as the image at the entrance plane is moved across the slit. A rotating front surface mirror (3) in front of the focusing element (4) may be used. Another approach would simply be to rotate/move the whole instrument itself or the target on a conveyor belt etc. In fact, the use of a mirror enables us to sample target objects that are static or moving relatively to the instrument by rotating or keeping the mirror fixed, respectively. The instrument uses a reflecting grating (7) as the dispersive element. The grating and front surface mirror are controlled / tilted by stepper motors. The spectral range is from UV to NIR depending on the grating angle. In our case the sun is the light source (1) and the fish (2) is exchanged with Storvola as target.
The image at any wavelength l
is constructed by processing the spectrograms (10)
recorded as a function of front surface mirror angle q.
The image is simply obtained by stacking side by side the intensity curves (rows
at wavelength l parallel with the slit (5))
for each spectrogram.
2.2 Experimental setup
A picture of the assembled system is shown in Figure 2. The systems optical diagram is identical to Figure 1. The camera head (1), the MX516 is made by the company StarXpress. It contains a 510x290 pixel CCD detector. The size of the CCD is 5 x 3.6 mm. The size of each pixel is 9.8x12.6 mm. The dynamic range is 16 bits per pixel and the spectral range is 400-1100 nm. The detector is made for the Astronomical community to image stellar objects using large telescopes. The CCD is cooled by a Peltier element down to -20C, which enables it to make long exposures in the order of minutes with low noise level. The exposure time and the read-out process is controlled through the printer port of the PC. The readout time of the CCD is 10 seconds.
A standard C-mount 50 mm focal length lens (2) is used to focus the diffracted and dispersed light from the grating onto the CCD. The lens has manual iris control used to detect the background level of the CCD. The front of the lens is connected to a square housing (3), containing the grating and the fixed front surface mirror. The reflective grating is machined ruled with 1200 grooves per mm. It is 30 x 30 mm square with a blaze angle of 17.46 degrees. The fixed front surface mirror is 35 x 35 mm square, tilted with an angle of 24.5 degrees to the optical axis.A variable length tube / barrel assembly (7) of 48 mm in diameter is connected to the square housing in front of the fixed mirror. It contains a manual iris (6), the collector lens, the adjustable entrance slit and a 35 mm camera bayonet base adapter. The collector lens is 20 mm in diameter with a focal length of 50 mm. The slit is 7.2 mm high and 150 mm wide. The spectral resolution or bandpass of the instrument is approximately 1 nm for the first order at wavelength 546.1 nm (Mercury green line).
The field lens (8) or the front lens of the instrument is connected to the 35 mm camera bayonet base adapter. The adapter has standard C-threads on the other end to fit the variable length barrel with the entrance slit positioned in the focal plane of the lens. This type of adapter enables us to select from a wide range of 35 mm camera lenses manufactured by Nikon. The lens in Figure 2 is a 50 mm focal length normal objective including variable focus and manual iris control. All mounts and adapters are found in the Mix and Match assemblies of the company Edmund Scientific (ES).
The above described spectrograph is named Spextube IV. It is mounted on square aluminum bars (12) which slide on 0.5 m long steel rods (13). The scanning front surface mirror (9) system is mounted in front of the field lens. Light from the target object is reflected by the front surface aluminized 70 mm square glass mirror into the field of view of the field lens (8). The mirror is attached to a 1:500 gear drive (10). A stepper motor (11) is used to control the mirror's angular position. The motor was disconnected from an old floppy disk drive. It has 200 steps per revolution. The same type of motor is connected directly to the grating shaft. The stepping sequence to the motors are generated by a microcomputer (5), the Stamp II made by the company Parallax Inc. Each motor uses an array of Darlington transistor pairs, the Motorola's ULN2003 chip. These chips switches and amplifies the current to the coils in the motors. The microcomputer communicate with the PC through a serial RS-232 port.
2.3 Calibration routines
Ideally each photon entering the imager will be transformed to an electronic count by the detector, but some loss will always occur. The throughput can be considered as an equivalent for the sensitivity or wavelength dependent calibration curve of the instrument, giving the ratio of electronic counts out to the number of photons passing through the instrument. It is necessary to calibrate the instrument against a known light source (Standard Tungsten Lamp) to obtain the spectral intensity in absolute units. The wavelength calibration is carried out using a Mercury gas discharge lamp.
The instrument was calibrated both before and after the transport to Storvola. Figure 3. shows the result of the absolute calibration. B0(l) is the known intensity of the calibration lamp (certificate), and C0(l) is the corresponding response in counts per unit time as measured by the spectrograph. A detailed description of the calibration routine is given by Sigernes et al. (2000). From these curves we conclude that the imager is most sensitive in the green/yellow/red region of the spectrum. The response falls rapidly when approaching the ultraviolet and the near infrared.
3. FIELD SETUP
The above described instrument was constructed for indoor laboratory use only. An instrument house and a heavy duty camera tripod were applied to protect the imager and to point / aim it at the target. Figure 4. shows the setup at the site. A holographic laser sight was mounted on top of the instrument house in order to align the imager at the target area. When the right position was found the imager was locked by ropes to the ground in order to minimize influence of wind gust.
The power to the instrument and the controlling PC was supplied by a car battery which was continuously charged by a petrol aggregate. The 12V from the battery was transformed to 220V AC by a DC/AC converter. This setup gives a stable power line and gives the operator time to refuel the aggregate without loosing power. In addition, a small tent was setup to screen the PC from sun light.
4. RESULTS AND DISCUSSION
The target area selected for the experiment consist of several layers of sandstone with layers of shale in-between. The photograph of Figure 5. shows how these layers are stretched in the horizontal direction. The layers are cut of by small ravines, formed by avalanches, leading melt water and sediments down to the Van Keulen fjord.
The original idea for the experiment was to use the instrument to form images throughout the spectral range of the instrument. This did not work, because the time it take to scan the target and store the data is approximately 30 minutes per spectral region. Each spectral region is up to 50 nm wide. A total of 12 sweeps would then be necessary to cover the whole range from 380 to 970 nm. During this time both the solar zenith angle and changes in cloud cover will make the source illumination of the mountain side non-uniform. Figure 5 shows an example of the effect. The 455 nm, 555 nm and the 622 nm images are generated from spectrograms that are obtained when the instrument recorded the wavelength ranges 432 - 494 nm, 524 - 581 nm and 613 - 667 nm, respectively. The change in light condition is seen as a vertical dark or bright stripes, occurring with change in cloud cover. Future outdoor use of this instrument requires faster readout time of the detector and a measure of the total flux variation with time.
A new strategy was formed. A cross-section of the scene was selected to cover the layers vertically. Panel (A) of Figure 5 shows the vertical slice marked. The rotating mirror was then moved and locked in position to cover this region only. The angle of the mirror was calculated by using the instruments own images. The next step was to rotate the grating instead of the mirror to obtain spectrograms covering the whole spectral region from 380 to 970 nm. The procedure took only about 3 minutes and enables us to extract a spectrum for each point along the selected line. Each point is calculated to be close to 2.4 x 0.8 m in size. The solar zenith angle was 74.4° and the cloud condition was stable. The azimuth angle of the sun was 269° from the South, which is almost straight westward.
As a first cut, 2 spectra were extracted from the line crossing sandstone layers of different height. The region between represent the shale spectrum. See the line marks in panel (A) of Figure 5. The spectra are shown in Figure 6. Note that the sandstone spectra (1) and (2) are the average response of 10 points along the selected line. The shale spectrum (3) is an average of 30 points. In order to compare with the light source, the global solar irradiance was calculated using the radiative transfer equation solver of (Stamnes et al., 1988). The solar zenith angle was set to 74.4°, the surface albedo to 0.5 and the ozone content to 300 DU. The arctic summer atmosphere model of Anderson et al. (1986) was used and the gaseous absorption properties taken from the SBDART model (Ricchiazzi et al., 1998). The principal Fraunhofer lines and the atmospheric water absorption bands are all clearly identified to coincide with the model. The shape of our spectra are more flat in the visible (VIS) compared to the source. The blue component of the global radiation (blue sky) is as expected not that strong present. As noted during the sensitivity calibration, the noise level in Near Infra red (NIR) is higher than in the VIS part. But on the other hand, the intensities are all clearly rising compared to the source in the NIR region. The bias or average intensity level for the sandstone (1) spectrum is in the order of 1.5 to 2 times stronger than compared to the shale (3) and the sandstone (2) spectra, respectively. It should be noted that this bias depends on the illumination conditions, mainly controlled by the location of the sun and angle between the observer and the target. The measured spectra are the portion of sunlight that is reflected off the target area and directed into the field of view of the instrument, making the measurements directional intensity fluxes.
In order to study the individual differences between the spectra, the bi-directional reflectance factors are calculated by dividing the measurements with the global solar irradiance as reference. The bandpass of the global spectrum is fixed at 1.0 nm, while the measured spectra have variable bandpass from 1 - 2 nm. A convolution with a triangular function of bandpass 25 nm was introduced to reduce noise and make the bandpasses compatible. One more aspect about the convolution is to make the spectra as broadband as possible without loosing to much information. This will put less constrains on speed and throughput when performing an airborne survey. The main obstacle in the design of an airborne imager is to maintain high throughput at as both the readout time and exposure time of the detector must mach the velocity of the carrier relative to the target (Sigernes et al., 2000).The reflection factor are plotted in Figure 7. Since the reference spectrum is based on a model, the reflectance curves must be viewed as relative and not absolute. If we for example reset the albedo to zero, then the reflectance curves will be raised.
From Figure 7 we see that the reflectance curves look parallel and slowly increasing from blue to red wavelengths in the VIS part of the spectrum. In the NIR region they increase exponentially with a peak close to 945 nm. Again as noted above, the sandstone (1) reflections factors are overall 1.5 - 2 times higher than the shale and the sandstone (2) factors, respectively. It is hard to see any difference in spectral shape at this stage. Nevertheless, the NIR part seem to be the region to use in order to obtain high reflection values from all layers. In order to get a more detailed view, the ratios between the reflection factors must be calculated. Taking the ratio between the sandstone layers and the shale should reveal whether the instrument is cable of distinguishing the spectral difference. The ratios showed in Figure 7 are normalized. The ratio between sandstone (1) and the shale (3) shows that the sandstone (1) layer has relatively higher blue to red reflection factors with peaks at 500 and 645 nm. In the NIR the opposite is true except for a 60 nm wide window centered close to 905 nm. The maximum ratio is at 645 nm and the minimum at 830 nm. The ratio between sandstone (2) and shale (3) shows that the sandstone (2) has relatively higher reflection factors up to 445 nm and in the region from 600 to 905 nm. The opposite is true from 445 to 600 nm and above 905 nm. Here the local minima and maxima are located at 500 and 730 nm, respectively. The next step is to consider the sandstone (1) to sandstone (2) ratio. Up to 445 nm the ratio is less than one. Above 445 nm the relative reflection factors are higher for sandstone (1) compared to sandstone (2). Especially the peak close to 500 nm. At 660 nm it turns. The deep red color in layer (2) is stronger than in layer (1). Up to 890 nm the ratio is less than one and above it is greater than one again. The local maximum is at 500 nm and the minimum at 730 nm. Note that the above discussion only considered relative ratios.
Based on the above result it is possible to both separate the sandstone layers from each other and the sandstone from the shale by the use of imaging spectroscopy. One simple test could be to obtain images at a bandpass of 25 nm at center wavelengths close to the above minimum and maximum wavelengths and then take the normalized ratio between them. A camera system based on broadband interference filters in front of the camera lenses should work. An airborne spectral imager would be an advantage compared to a camera system based on the simple fact that we only need to calibrate one detector, and that we obtain images throughout the whole spectrum. More advanced processing routines such as classification can then be applied with good statistics due to the larger pool of data describing the scene. For example, using an modified version of the above instrument with the rotating mirror removed, a slit width of 25 mm and a detector capable of real time imaging (25 frames / spectrograms per second), the spatial resolution is calculated to be 2.4 x 1.5 m. The velocity of the airplane is then typically 200 km / hr and the altitude is 3000 m.The calculated resolution should in theory be sufficient to map the mountain sides of Spitsbergen. In addition, the grating must be changed to cover a wider spectral range. The construction of an airborne imager is under way.
5. CONCLUDING REMARKS
The principal results obtained by this study may be summarized as follows.
Hyper spectral imaging of the mountain bed rock layers of Storvola (77.5°N,16.2°E) in Spitsbergen have been performed in the wavelength range 380 - 970 nm with a spectral resolution of 1-2 nm and a spatial resolution of 2.4 x 0.8 m.
Relative reflection factors as a function of wavelength were obtained for the shale and two different layers of sandstone.
All 3 layers show slowly increasing reflection as a function of wavelength from the blue to red in the visible region, followed up by an exponential increase as we approach the near infra red region of the spectrum.
The calculated broadband reflection ratios show that the slow scan spectral imager is capable of detecting the spectral difference between shale and sandstone. It can also identify the signature of different sandstone layers.
Design / instrumental parameters for an airborne spectral imager have been found.
Figure 1. A 3 dimensional optical diagram illustrating the main principle of the instrument. (1) is light source, (2) target, (3) scanning front surface mirror, (4) front lens, (5) entrance slit, (6) collector lens, (7) grating, (8) fixed front surface mirror, (9) camera lens, and (10) CCD imaging detector. The optical axis is parallel with the X-axis of the XYZ-coordinate system. The slit is located parallel to the Y-axis. q is the angle between the mirror's normal and the optical axis. w is the tilt angle of the fixed front surface mirror, equal to the grating blaze angle. a is the incident grating angle. The detector is located in the lS - plane which is parallel to the ZY- plane.
Figure 2. Picture of the assembled Spextube IV spectral imager. (1) is detector (CCD), (2) camera lens, (3) instrument house containing grating and fixed front surface mirror, (4) grating stepper motor, (5) Stamp II microcomputer, (6) adjustable iris, (7) variable length tube / barrel with collector lens, adjustable slit and camera bayonet adapter, (8) field lens, (9) front surface mirror, (10) gear box, (11) stepper motor, (12) aluminum mount bars, and (13) steel rods.
Figure 3. Results of absolute calibration of Spextube IV. The green function C0 shows instrument's response to the diffuse re-emitting screen illuminated by the absolute calibration lamp (Tungsten). The units are given in raw counts ([Cts/s]). The exposure time was 200 msec. The color coded curve represents the absolute certificate spectrum B0 for the 200W Tungsten lamp in units [mW/m2 Å].
Figure 4. Experimental setup of the Spextube IVspectral imager in front of the mountain Storvola (77.5°N,16.2°E) at Spitsbergen, Norway, 11.08.2001.
Figure 5. Panel (A): Image shows a section of the mountain Storvola (77.5°N,16.2°E) at 17:45 LT 11.08.2001. The green line marks the position the spectral imager was moved to in order to obtain a vertical cross section of spectrograms in the range 380-970 nm. The selected points (1) and (2) are bed rock layers of sandstone, and (3) is the layer in between (shale). Panel (B): Images generated by the Spextube IV spectral imager with wavelength 453, 555 and 622 nm. The bandpass is 4 nm, the size is approximately 115 x 118 m, and the spatial resolution is close to 2.4 x 1.6 m.
Figure 6. Spectra from Storvola (77.5°N,16.2°E) obtained with the Spextube IV spectral imager at 17:45 LT 11.08.2001. The spaces between the vertical dotted lines mark the regions that the instrument covers on wavelength as a function of grating angle. The curves (1) and (2) represent spectra of the sandstone layers, while (3) is the shale spectrum. The spectrum plotted in black is the modeled global solar irradiance at the site. The azimuthal angle of the sun was 269° from South with a zenith angle of 74.4°. The horizontal tagged line at the bottom marks the principal solar Fraunhofer lines. The absorption bands in the range 650 - 970 nm are due to atmospheric water vapor. An ozone content of 300 DU and an albedo of 0.5 are used as input to the calculations of the global solar irradiance.
Figure 7. Relative directional reflectance factors from the bedrock layers of Spectra from Storvola (77.5°N,16.2°E) obtained with the Spextube IV spectral imager at 17:45 LT, 11.08.2001. The solid lines (1) and (2) are the relative reflection factors for the sandstone layers, respectively. (3) is the reflection factors of the shale. The dotted red line represents the normalized ratio between the sandstone (1) and the shale (3) reflection factors. The dotted green is the corresponding sandstone (2) by shale (3) ratio. The dotted blue line is the normalized ratio between sandstone (1) and (2).
We would like to thank Dr. Ole Humlum at UNIS for his kind support of digital images to our project. We also wish to thank Asle Strøm, Statoil, for photos and Sindre Flatås, Statoil for organizing security and logistics during the field work.
Anderson, G., S. Clough, F. Kneizys, J. Chetwynd, and Shettle, E., AFGL atmospheric constituent profiles (0-120 km)}, Tech. Rep. AFGL-TR-86-0110}, Air Force Geophys. Lab., Hanscom Air Force Base, Bedford, Mass., 1986.
Harland, W. B., The tectonic evolution of the Arctic-North Atlantic Region: Philosophical Transactions of the Royal Society of London, Series B., v. 258, p. 59-75, 1965.
Helland-Hansen, W., Geometry and facies of Tertiary clinothems, Spitsbergen: Sedimentology, v. 39, p. 1013-1029, 1992.
Herrala, E., and Okkonen J., Imaging spectrograph and camera solutions for industrial applications,
International Journal of Pattern Recognition and Artificial Intelligence, 10, 43, 1996.
Hyvarinen, T., Herrala, E., and Dall'Ava, A., Direct sight imaging spectrograph: a unique add-on component brings spectral imaging to industrial applications, SPIE-The International Society for Optical Engineering: Symposium on Electronic Imaging, Paper 3302-21, San Jose, California, January 25-30, 1998.
Niblack, W., An Introduction to digital image processing, Prentice/Hall international (UK) Ltd, 1, 168-181, 1986.
Ricchiazzi, P., S. Yang, C. Gautier and Sowle, D., SBDART: A research and teaching software tool for plane-parallel radiative transfer in the Earth's atmosphere, Bull. Amer. Met. Soc., 79, 2101-2114, 1998.
Sigernes, F., Lorentzen, D.A., Heia, K., and Svenøe, T., A multi-purpose
spectral imager, 39, N0. 18, 3143, (Front cover issue), Applied Optics,
Stamnes, K., Tsay, S.-C., Wiscombe, W., and Jayaweera, K., Numerical stable algorithm for discrete-ordinate-method radiative transfer in multiple scattering and emitting layered media, Applied Optics, 27, 2502, 1988.
Steel, R.J., Gjelberg, J, Helland-Hansen, W., Kleinspehn, K., Nøttvedt, A. and Larsen, M. R., The Tertiary strike-slip basins and orogenic belt of Spitsbergen, in
Biddle, K.T. and Christie-Blick, N. (eds.) strike –slip deformation, basins formation and sedimentation: Society of Economic Paleontologists and Mineralogists, Special Publication,, v.37, p. 339-358,1985.
Vane, G., Imaging Spectroscopy II, SPIE-The International Society for Optical
Engineering, 834, 1-238, 1988.
Wolfe, W.L., Introduction to Imaging Spectrometers, SPIE-The International Society for Optical Engineering, TT25, 1-147, 1997.