Hannu Holma, Antti-Jussi Mattila, Aappo Roos, Timo Hyvärinen, Specim, Oulu, Finland

Oliver Weatherbee, SectTIR LLC, Fairfax, Virginia, USA

Translated from "Photonik International.2011"

Hyperspectral sensing is of great importance in biomedical, chemical, industrial, aerospace and military applications. In this article, we focus on hyperspectral sensing in the long-wave infrared (LWIR) region, its development and its potential use in advanced applications.

A very important element that determines the achievable results of a thermal hyperspectral camera and its possible applications is the type of detector we use. In this paper, we consider a specific measurement of a hyperspectral camera in the long-wave infrared (LWIR, 7 to 14 µm) region that uses a cooled mercury-cadmium-tellurium (MCT) or FPA sensor.

At present, there are few instruments that are capable of achieving good quality results while being easy to use. For example, in the field of airborne sensing in LWIR, the Sabass instrument from The Aerospace Corporation [1,2] can be considered as a reference sensor, even though it is quite bulky and requires special handling and maintenance. However, with the growing interest in thermal spectral sensing by the military and commercial sectors, the development of new sensors is increasing. These then use both Fourier transform and pushbroom spectrometers and are applicable for example in agriculture, laboratory and aerospace measurements [3].

Hyperspectral measurements

The goal of hyperspectral imaging is to obtain an image of the imaged area such that each pixel contains the entire spectrum. At the same time, in hyperspectral sensing we can use the knowledge of the spatial relationship between different spectra to create a more detailed spectral-spatial model for more accurate segmentation and classification of the acquired image. Since this is a new analytical method, its possibilities have not yet been fully exploited.

There are several approaches to obtain a hyperspectral image. One of them is the use of a pushbroom sensor that captures only one line of the image. In order to obtain a hyperspectral image, it is then necessary to move either the sample or the camera to capture the entire area to be imaged.

Figure 1: The principle of a pushbroom hyperspectral sensor. The sensing optics include collimating, prismatic and focusing elements.

This device consists of three main parts: the input optics, the spectrograph and the camera sensor (Figure 1). The input optics contains a lens that is designed for the spectral range to be recorded. The aperture number, the field of view of the lens and the size of the camera sensor are then adjusted to achieve minimal aberration and over-illumination. The optical system then projects the recorded image through a narrow slit onto the input of the spectrograph. The slit thus ensures that only a narrow band of the image is passed through. The narrower the slit, the higher the spatial resolution achievable.

The light that passes through the slit is then decomposed by the spectrograph and projected onto the two-dimensional array of the detector (camera sensor). The image is then captured in one direction of the detector array and the spectrum of each pixel is recorded in the other direction of the detector array. The image thus formed is then a three-dimensional "data cube".

Pushbroom recording still represents the only hyperspectral sensing technology that is applicable in laboratories, agriculture, aerial sensing and online measurement. The advantage of this sensing is the high image quality of moving objects and a much higher signal-to-noise ratio than whiskbroom sensing.

The sensor signal depends not only on the spectral emission of the sample to be measured, but also on the intrinsic emission of the instrument, which is related to its temperature. Therefore, hyperspectral sensing requires spectral and radiometric calibration so that we are able to correctly match the wavelength and spatial emission of the sample to the sensor signal. In the past, the calibration required the use of blackbody radiation that was integrated into the sensor. Today, a new approach makes it possible to replace this calibration with on-chip background sensing (BMC). In this way, a background signal is simultaneously associated with each image when the instrument's radiation is continuously measured.

Examples of possible applications

Sensing in chemical industry and laboratory conditions in reflective mode

Until recently, hyperspectral sensing in the chemical industry was limited to the visible/near infrared (VNIR) and shortwave infrared (SWIR) regions. However, many materials have a strong characteristic response in the LWIR region. For example, geological regions can be rapidly mapped by combining SWIR and LWIR spectral sensing for almost all commercially interesting materials. LWIR is then dominant in the region for mineral detection in feldspar, silica, calcite, garnet and olivine soils.

Figure 2: Reflection measurements. Specim VNIR, SWIR and LWIR-HS cameras were used for mineral mapping. For quartz, garnet and feldspar characterization, a LWIR camera is then required.

Figure 2 shows the geological area imaged using VNIR, SWIR and LWIR spectral cameras in reflection mode. A halogen lamp was used as the light source for the VNIR and SWIR, and a quartz heating element was used for the LWIR camera. Although this was a prototype illumination source for the LWIR, good mineral identification was achieved. Thus, the data obtained by LWIR imaging contains information for reliable mineral detection.

Figure 3: Reflection measurements. The LWIR-HS and LWIR-C sensors were used to measure a set of stone samples. The LWIR-C sensor allows for better sampling and resolution, therefore providing much more accurate mineral identification. In the magnified LWIR-C image, the feldspar spectrum is easily recognizable. LWIR-C sensing provides noticeably better response, especially in the region from 10.5 µm to 12.4 µm.

Figure 3 shows the LWIR reflection data for material identification. A quartz heating element was again used as the illumination source. It can be seen from the figure that feldspar and quartz are easily distinguishable from each other. Based on preliminary studies, we know that sodic-calcic feldspars (plagioclase) can be separated based on their composition, which has important implications for planning and optimizing commercial mining.

LWIR outdoor emission measurements

Figure 4: Radiation measurements. Outdoor scan test of the LWIR -HS camera at an ambient temperature of 10°C (August 2010). Description of the relative spectra of the different objects in the figure: 1. sky (ozone radiation), 2. first person's shirt, 3. second person's shirt, 4. brick wall, 5. sidewalk, 6. grass.

Figure 4 shows the measurements scanned by an LWIR sensor that has been modified for reflectance measurements with artificial lighting. The data from the sensor was normalized using 60°C and 3°C blackbody radiation. The normalization is an adjustment: (data - cold reference )/(warm reference - cold reference). This image provides several visible details. The window frame has a different temperature than the window and wall itself. The light spot located on the tripod represents a hot fan, while the dark rectangle to the left of the tripod is an ordinary book.

Figure 5: Reflection measurements, LWIR-C camera adapted to winter conditions, ambient temperature -15°C. The person on the left is exhausting clean air towards the right, which contains a propellant gas (1,1,1-2-tetrafluoroethane) that has absorption bands of 7690, 8400, 9070 and 10290 nm in the LWIR region. Against a warm background (second person on the right) the absorption peaks are detected and shown as a red spectral line. Against a cold background (sky) the emission peaks are then detected as a green spectral line.

Figure 5 shows an outdoor image taken by LWIR with radiometric correction for winter conditions (-15 °C). Despite the temperature difference, which is close to 30 K, the measured data are of better quality compared to Figure 4. The spectra of the individual objects have clear characteristics. The objects can thus be easily mapped and classified according to their chemical surface. This method can also be used to measure and identify gases.

LWIR-C aerial experiments

For airborne thermal hyperspectral measurements, the AisaOWL system has been developed using the LWIR hyperspectral camera. The first test flight took place in March 2011 over the city of Reno and the Cuprite area of Nevada, USA. The urban area was selected for camera image verification and mutual calibration of the camera and GPS/INS sensor for accurate geological reference.

Figure 6: Radiance measurements, 27 February 2011, part of the AisaOWL aerial survey over the city of Reno, Nevada, USA. The measurement is shown as an RGB image in three wavelengths, represented as vertical lines in the spectral image. The green arrow refers to the spectrum of the sunlit portion of the Silver Legacy casino dome. Red arrow: spectrum of a circular very cold structure near the casino. Blue arrow: spectrum of a rectangular building. Yellow arrow: spectrum of a body of water.

Figure 6 shows a flight over the city of Reno. The gray scale is used to capture the radiation level in the LWIR waveband. The data was collected in daylight, meaning that it includes both thermal and reflective contributions from sunlight. The spectra of each object contain both spectral emission and temperature differences. During the measurements, the weather was sunny but cool with temperatures around 0°C. Temperatures were below freezing the night before data collection. The image shows snow lying on the roofs of houses and in the streets.

Figure 7: Radiation measurements. Mosaic of two adjacent AisaOWL aerial measurements over the Cuprite area, Nevada, USA on 27 February 2011. Sample spectra, arrows point to the location corresponding to the measured spectrum.

The Cuprite area in Nevada represents one of the most suitable reference areas for mineral mapping in the world. It has been mapped by several LIWR sensors, such as the Sebass instrument, and therefore the data from this area serves as a suitable comparison for new hyperspectral sensors. Figure 7 presents data that have been georeferenced and mapped using an elevation model with 1m resolution. The sensing was performed on a sunny day at 0°C. The night before the sensing then temperatures were below freezing. No snow is visible in the image, but some spectral curves are shown. A preliminary comparison of the measured results shows agreement with previous measurements. A more detailed analysis of the data is currently underway (2011) and the results will be published.

Conclusion

Thanks to new approaches, LWIR hyperspectral sensing is becoming increasingly applicable in many industrial, scientific and military applications that require high spectral resolution and signal-to-noise ratio.