Open Access
Add to BibSonomyAbstract
We present the design of a full waveform hyperspectral light detection and ranging (LiDAR) and the first demonstrations of its applications in remote sensing. The novel instrument produces a 3D point cloud with spectral backscattered reflectance data. This concept has a significant impact on remote sensing and other fields where target 3D detection and identification is crucial, such as civil engineering, cultural heritage, material processing, or geomorphological studies. As both the geometry and spectral information on the target are available from a single measurement, this technology will extend the scope of imaging spectroscopy into spectral 3D sensing. To demonstrate the potential of the instrument in the remote sensing of vegetation, 3D point clouds with backscattered reflectance and spectral indices are presented for a specimen of Norway spruce.
©2012 Optical Society of America
1. Introduction
Supercontinuum laser sources produce directional broadband light by making use of cascaded nonlinear optical interactions in an optical fiber (see [1] for a review). The frequency mixing processes are excited by high peak power laser pulses (with pulse peak power of about 2kW - 20kW) passing through a nonlinear optical fiber, producing a broadband laser pulse. The commercial availability of supercontinuum laser technology has lead into a number of applications in the recent years, such as those in biomedical optics (e.g., [2]). Combined with a hyperspectral time-of-flight sensor, the supercontinuum laser sources can be used for simultaneous measurement of distance and reflectance spectrum, which has been the basis for our recent efforts in development of the hyperspectral LiDAR.
There is a great need for improved automation in the classification and interpretation of laser scanning (also called scanning LiDAR) data. This has recently been facilitated, e.g., by radiometric calibration methods that have been developed for monochromatic LiDAR intensity (see [3] for a review and references). A hyperspectral LiDAR instrument is capable of producing one-shot topographic and spectral intensity information, which will enable a simultaneous study of, e.g., structural and biochemical vegetation parameters without registration problems between data sets. Additionally, spectra acquired with an active hyperspectral instrument are not affected by illumination conditions or shadows, thus significantly simplifying post processing.
Combining topographic and spectral intensity information has thus far been based on passive spectroscopic sensing. Laser scanner point clouds fused with images or passive imaging spectrometry have been applied in, e.g., investigating forest canopy structure, leaf physiology, or tree species classification [4–6]. Simultaneous geometric and spectral information can also be retrieved by the means of a novel approach for photogrammetric point cloud creation from automatic multispectral image matching [7]. Active hyperspectral imaging applications are also available, but mostly without the range information [8–10]. These systems acquire a spectral signature for every pixel in the image captured by an imaging detector. Range information is included in multi-wavelength laser scanners that use separate monochromatic lasers, or laser with frequency doubling as light sources for multiple wavelengths [11–13]. For these, the wavelength channels are determined by the light sources. Some satellite based LiDAR systems have also been deployed for atmospheric and land surface studies [14, 15]. Spectrally resolved ranging has recently been achieved by using frequency combs [16].
The concept of hyperspectral scanning LiDAR combines active hyperspectral imaging and laser scanning with the same instrument, enabling a simultaneous survey at the same weather and illumination conditions. This reduces the measurement costs and there will be no registration problems between data sets. The extended final product of hyperspectral laser scanning includes the point cloud and hyperspectral reflectance: (x,y,z,R(λ)), where R(λ) is the backscattered reflectance R as a function of the wavelength λ. The information content of the new type of data is vast and creates new prospects for automatic data processing and target characterization.
This article presents the design of a full waveform hyperspectral LiDAR and its first demonstrations in the remote sensing of vegetation. The feasibility of the concept was studied with a scanning active hyperspectral measurement system developed by the same team [17]. The active hyperspectral intensity data were fused with simultaneous terrestrial laser scanner measurement. This enabled us to study the usage of hyperspectral 3D point clouds in target classification [18]. A 2-channel spectral range finding system was also tested [19] to continue the development and design towards the LiDAR instrument presented in this article. To our knowledge, this is the first full waveform hyperspectral LiDAR producing spectral 3D point clouds and exploiting the supercontinuum laser technology, and one of the first environmental applications of supercontinuum lasers.
2. The instrument
We have assembled an optical setup (Fig. 1 ) for measuring the time-of-flight and return intensity of a hyperspectral laser pulse. The supercontinuum laser (NKT Photonics, SuperK) produces 1 ns pulses at repetition rate of 24 kHz and average power of 100 mW. The power is spread over a spectral range of 480–2200 nm (Fig. 2 ). The broadband output laser is collimated using a refracting collimator (Thorlabs, CFC-5-A). The collimated beam passes through a beam sampler, which takes a part of the beam for triggering the time-of-flight measurement. An off-axis parabolic mirror (50.8 mm diameter, 152.4 mm effective focal length and 90° off-axis angle) is used as the primary collecting optic. Two 4-mm holes are drilled through the mirror, one parallel to the optical centerline and one towards the focus point. This configuration allows the main beam to be passed through one hole to the target, while the light for triggering the waveform collection is guided directly to the spectrograph through the other hole.
Fig. 1 The optical setup: A laser pulse from a photonic crystal fiber (A) is collimated and sent to a 2D scanner (B). An off-axis parabolic mirror (C) is used as a primary light collecting optic. A spectrograph (D) disperses the colors of the trigger (E) and echo pulses to an APD array, which converts the light to analog voltage waveforms.
Fig. 2 The broadband output spectrum of the SuperK. The spectral range is approximately 480-2200 nm.
The off-axis parabolic mirror is focused to a spectrograph (Specim, ImSpector V10), which has a 50 µm wide slit, spectral range of 400–1000 nm, and f-number of f/2.8. A 16-element avalanche photo diode (APD) array module (Pacific Silicon Sensor) is used to convert the spectrally separated light to analog voltages. The spectral response range of the APD is 450–1050 nm, with a maximum at 880 nm. The APD module has built-in transimpedance amplifiers (Analog Devices, AD8015) with a bandwidth of 240 MHz producing an unambiguous resolution of approximately 4 ns. 12-bit analog to digital converters (SP Devices, ADQ412), with 1 GHz sampling rate, are used to digitize 8 of the 16 available spectral channels. An average of 10 pulses is saved to improve signal to noise ratio and to reduce the amount of data.
The scanning geometry is defined by the two rotators (Newport URS75BCC and URS100BCC) with an absolute accuracy of ±0.0115°. The rotators are attached to each other, with one performing the azimuth rotation and the other sweeping the laser over the target area in vertical passes. Each recorded waveform is associated with the azimuth and elevation angles using the rotator drive parameters and time stamps. Due to uncertainty in timing, the accuracy of the elevation angles is 0.1°.
3. Calibration and data processing
A monochromator (Oriel, Cornerstone 74125) was used to calibrate the spectral responses of the APD elements. The current configuration produces spectral Full Width at Half Maximum (FWHM) of about 19 nm for each element and spectral sampling interval of 34 nm. With 16 element sensor spectral range is 470-990 nm. However, the sensitivity of the APD array and the laser intensity below 550 nm are low, and therefore the first channel is selected to be at 542 nm followed by 606, 672, 707, 740, 775, 878 and 981 nm. This channel selection has been optimized for the measurement of vegetation (Fig. 3 ).
Fig. 3 Spectral channels of the hyperspectral LiDAR and passive spectrometer measurement of Norway spruce. Current channel selection (solid curves) is optimized for the measurement of vegetation.
The waveforms produced by the APD electronics cannot be directly converted to radiant power as the waveforms suffer from a negative overshoot after echoes (Fig. 4 ). With high intensity echoes, the overshoot is also accompanied by ringing, where the signal oscillates with decreasing amplitude. This can be seen in the data as echoes behind solid objects, e.g. behind a wall. Most of the overshoot and ringing can be mathematically removed from the waveforms. Waveforms containing the overshoot shapes are collected during the system calibration. An ideal waveform shape is then calculated from the overshooting one by replacing all values outside the main echo (a local maximum in the waveform) with null values. The overshoot effects are removed from the waveforms with an iterative algorithm, which replaces the overshooting waveform shapes with the ideal ones (Fig. 5 ). This algorithm is able to remove practically all of the overshoot effects from the signal of a single APD channel, but crosstalk between adjacent channels causes still some minor overshoot effects that cannot be removed. These crosstalk effects may slightly reduce the radiometric accuracy, especially in cases where the intensities received by adjacent channels differ significantly.
Fig. 4 Hyperspectral LiDAR waveforms at various stages of processing. The top left plot depicts the raw waveforms of each hyperspectral channel as recorded by the analog to digital converters. Thick black line is the mean waveform of all channels. Trigger and target parts of the waveforms are normalized in different scale. The negative overshoot is visible, e.g., after the trigger pulse. The bottom left plot depicts the same waveforms with overshoot effects removed. The 3D plot on right shows the backscattered reflectance waveforms of the targets, produced using Gaussian function fitting and instrument calibration.
Fig. 5 Flowchart of an algorithm removing overshoot effects from waveforms. The overshooting waveshapes are replaced with ideal ones, until the source waveform is diminished below a threshold level. To avoid errors caused by excess fitting, the residual of the source wave is added to the output waveform containing only ideal shapes.
The transmitted pulse energy of the SuperK laser source may vary slightly. To take this into account waveforms are normalized with the transmit pulse intensity. An average waveform of all spectral channels is calculated and a Gaussian peak function is fitted to the trigger part of the waveform. Similarly, the return echo positions are detected from the mean waveform. Once the return echo positions and widths are determined from the mean waveform, the hyperspectral intensities are extracted by fitting Gaussian peak heights to the spectral waveforms.
The intensity is converted into reflectance by applying the distance and spectral calibration. During the calibration measurements, waveforms are collected using a 99% Spectralon (Labsphere Inc) as a reference target at various distances. The echo intensities are normalized with the intensity of the Spectralon echo at the same distance, producing “backscattered reflectance”. As the backscattered reflectance spectra are combined with the corresponding time-of-flight and concurrent scanner orientation, a hyperspectral point cloud (x, y, z, R(λ)) is produced.
The accuracy of the measurement and the Gaussian fitting was tested by acquiring waveforms of 100 pulses reflected from a Spectralon panel at a 6-meter distance. The distance and the backscattered reflectance spectrum were retrieved individually for each pulse. The precision of ranging (standard deviation) was found to be 11.5 mm. The precision of backscattered reflectance of a single waveform was found to be better than 2% for spectral channels within the range of 600–800 nm and better than 5.5% for all channels. The quality of the fit is affected by the echo intensity, and thus lower precision is expected for targets that are darker or further away from instrument. Higher precision can be reached by averaging over a number of measurement points. For both range and reflectance, the absolute accuracy is expected to be lower than the precision due to uncertainty in calibration.
The instrument does not have a strictly defined maximum range of measurement, as the performance of the waveform echo detection decreases slowly with the fading echo intensity. The current configuration is focused to approximately 12-meter distance. Measurements have shown that high quality point clouds can be measured from targets within 10-meter range and bright targets can be detected even from over 20 meters.
4. Results and discussion
A Norway spruce (Picea abies) was measured in laboratory using the full waveform hyperspectral LiDAR. The bottom branches of the 2-meter spruce had suffered from lack of light and were in various stages of drying and dying, while the top branches had healthy new growth. The spruce was cut down at Nov. 7th and left inside overnight, with the end of the trunk submerged in water. The measurements were performed on the next day producing 107 000 hyperspectral points within the spruce. In addition to the LiDAR measurement, reference spectra were acquired using a passive spectrometer (Avantes, AvaSpec 3648) and a quartz-tungsten-halogen light source.
Both measurements were carried out in dark laboratory space with visually black background canvas. As the laser pulse is orders of magnitude brighter compared to even direct sunlight, there was no effect of ambient lighting to the LiDAR measurements. With passive spectrometer only the part of the tree being measured was illuminated. Spectralon was used as a white reference target both with the LiDAR and passive spectrometer. As the cross section of the spectrometer footprint and the illuminated tree area was unknown, the reflectance factor values could not be retrieved accurately and the passive spectrometer data had to be scaled to be comparable to the LiDAR data.
Spectra of the passive spectrometer and LiDAR measurements of selected regions of interest are presented in Fig. 6 . A clear distinction between the tree trunk and the top can be observed in the shape of the spectra. The LiDAR and passive spectrometer spectral shapes are clearly similar. In case of the tree top, the LiDAR observes less light than the passive measurement in near-infrared. This difference is caused by multiple scattering in a medium with a low optical density and a high single scattering albedo. In an active LiDAR measurement, only a small spot on the target is illuminated and observed. A significant part of the pulse energy is lost outside the sensor field of view, if multiple scattering plays a major role in reflectance and the scattering mean free path is long in the medium. This is not experienced in passive measurement as the same amount of light is scattered both in and out of the sensor field of view. The backscattered reflectance from Spectralon is not significantly affected by this effect, as Spectralon has a high single scattering albedo but only a short mean free path. As the LiDAR backscattered reflectance is calibrated with that of the Spectralon panel, the backscattered reflectance values are decreased for bright and low optical density targets such as needles.
Fig. 6 Comparison of spectra collected from the Norway spruce using the hyperspectral LiDAR and a passive spectrometer. To improve the comparability, the spectrometer spectra have been scaled to same level as the LiDAR spectra and smoothed with a 19-nm Gaussian filter.
The backscattered reflectance values produced by the LiDAR do not strictly follow the definition of reflectance factor for three reasons: First, due to hot spot effect [20], the 99% Spectralon is not a Lambertian surface in backscattering direction [21] causing systematic error in the reflectance values. Second, the illuminated surface area and the orientation of the target are not constant (as in the definition of reflectance factor) and this results in uncertainty in the returned intensity. Third, part of the transmitted light is lost outside the sensor field of view due to multiple scattering, as described above. Despite these limitations, the backscattered reflectance is a practical quantity providing intensity readings independent of measurement distance. For most applications, the backscattered reflectance spectra can be exploited similarly to traditional reflectance factors (e.g., in the computation and comparison of spectral indices), but caution should be used when accurate absolute values are needed.
Different vegetation indices can be obtained from the measured data set. For this study we selected Normalized Difference Vegetation Index (NDVI) [22], water concentration index [23] and Modified Chlorophyll Absorption Ratio Index (MCARI1) [24]. In Fig. 7 , these indices have been applied to the measured data set of the spruce.
Fig. 7 A photograph of the Norway spruce and 3D point clouds demonstrating various data products that were extracted from the backscattered reflectance spectra. To reduce noise the spectra have been averaged in 5-cm voxels. The spectral indices have been calculated for each voxel, and the results are displayed on the full 3D point cloud as colors.
The current setup of the instrument has been developed for laboratory measurements and the range is limited to 20 meters. However, we have also tested the instrument outdoors and concluded that it can be used under field conditions. With better optics, detector and more lightweight construction to be implemented in future versions, we are aiming to carry out terrestrial field measurements.
In the measurements presented in this article we have used eight discontiguous channels. Therefore it can be discussed whether the instrument is currently multispectral or hyperspectral. The measurement principle itself is hyperspectral; the spectrum of the supercontinuum laser is continuous as is the spectrum of the spectrograph output. In the instrument presented here we used a 16 element APD detector. With more digitizer channels all 16 APD outputs could be digitized and with an APD detector with more elements even larger number of spectral bands could be detected.
5. Conclusions
We present the concept and the first prototype of a full waveform hyperspectral terrestrial laser scanner. This instrument provides a novel approach for spectral imaging and laser scanning by producing one shot topography and spectroscopy. We have shown that the instrument is capable of producing hyperspectral 3D point clouds. The spectra can be used in, e.g., visualization and automated classification of the point cloud and calculation of spectral indices for extraction of target physical properties. The information content of the new type of data provided by the instrument is considerable and will facilitate more efficient and automatic retrieval of distinctive target properties, leading to improved monitoring tools for different applications. As a new application, this instrument makes it possible to study efficiently, e.g., 3D-distribution of chlorophyll or water concentration in vegetation. The first results indicate a possibility for improved efficiency in classification and interpretation compared to the traditional monochromatic LiDAR data. We believe that, as technology matures, hyperspectral laser scanners with extended distance and spectral range will also become available from commercial manufacturers.
References and links
1. J. M. Dudley, G. Genty, and S. Coen, “Supercontinuum generation in photonic crystal fiber,” Rev. Mod. Phys. 78(4), 1135–1184 (2006). [CrossRef]
2. A. Kudlinski, M. Lelek, B. Barviau, L. Audry, and A. Mussot, “Efficient blue conversion from a 1064 nm microchip laser in long photonic crystal fiber tapers for fluorescence microscopy,” Opt. Express 18(16), 16640–16645 (2010). [CrossRef] [PubMed]
3. W. Wagner, “Radiometric calibration of small-footprint full-waveform airborne laser scanner measurements: Basic physical concepts,” ISPRS J. Photogramm. Remote Sens. 65(6), 505–513 (2010). [CrossRef]
4. V. Thomas, J. McCaughey, P. Treitz, D. Finch, T. Noland, and L. Rich, “Spatial modelling of photosynthesis for a boreal mixedwood forest by integrating micrometeorological, lidar and hyperspectral remote sensing data,” Agric. For. Meteorol. 149(3-4), 639–654 (2009). [CrossRef]
5. T. G. Jones, N. C. Coops, and T. Sharma, “Assessing the utility of airborne hyperspectral and LiDAR data for species distribution mapping in the coastal Pacific Northwest, Canada,” Remote Sens. Environ. 114(12), 2841–2852 (2010). [CrossRef]
6. E. Puttonen, A. Jaakkola, P. Litkey, and J. Hyyppä, “Tree classification with fused mobile laser scanning and hyperspectral data,” Sensors (Basel) 11(5), 5158–5182 (2011). [CrossRef] [PubMed]
7. E. Honkavaara, L. Markelin, T. Rosnell, and K. Nurminen, “Influence of solar elevation in radiometric and geometric performance of multispectral photogrammetry,” ISPRS J. Photogramm. Remote Sens. 67, 13–26 (2012). [CrossRef]
8. B. Johnson, R. Joseph, M. Nischan, A. Newbury, J. Kerekes, H. Barclay, B. Willard, and J. Zayhowski, “A compact, active hyperspectral imaging system for the detection of concealed targets,” Proc. SPIE 3710, 144–153 (1999). [CrossRef]
9. M. Nischan, R. Joseph, J. Libby, and J. Kerekes, “Active spectral imaging,” Lincoln Lab. J. 14, 131–144 (2003).
10. G. Bishop, I. V. Veiga, M. Watson, and L. Farr, “Active spectral imaging for target detection” in Proceedings of the 3rd EMRS DTC Technical Conference (2006).
11. S. Tan and R. M. Narayanan, “Design and performance of a multiwavelength airborne polarimetric lidar for vegetation remote sensing,” Appl. Opt. 43(11), 2360–2368 (2004). [CrossRef] [PubMed]
12. M. Pfennigbauer and A. Ullrich, “Multi-wavelength airborne laser scanning,” in Proceedings of the International Lidar Mapping Forum, ILMF, New Orleans (2011).
13. G. C. Guenther, P. E. LaRocque, and W. J. Lillycrop, “Multiple surface channels in Scanning Hydrographic Operational Airborne Lidar Survey (SHOALS) airborne lidar,” Proc. SPIE 2258, 422–430 (1994). [CrossRef]
14. D. M. Winker, J. R. Pelon, and M. P. McCormick, “The CALIPSO mission: Spaceborne lidar for observation of aerosols and clouds,” Proc. SPIE 4893, 1–11 (2003). [CrossRef]
15. J. B. Abshire, X. Sun, H. Riris, J. M. Sirota, J. F. McGarry, S. Palm, D. Yi, and P. Liiva, “Geoscience Laser Altimeter System (GLAS) on the ICESat mission: On-orbit measurement performance,” Geophys. Res. Lett. 32(21), L21S02 (2005). [CrossRef]
16. M. Godbout, J. D. Deschênes, and J. Genest, “Spectrally resolved laser ranging with frequency combs,” Opt. Express 18(15), 15981–15989 (2010). [CrossRef] [PubMed]
17. J. Suomalainen, T. Hakala, H. Kaartinen, E. Räikkönen, and S. Kaasalainen, “Demonstration of a virtual active hyperspectral LiDAR in automated point cloud classification,” ISPRS J. Photogramm. Remote Sens. 66(5), 637–641 (2011). [CrossRef]
18. E. Puttonen, J. Suomalainen, T. Hakala, E. Räikkönen, H. Kaartinen, S. Kaasalainen, and P. Litkey, “Tree species classification from fused active hyperspectral reflectance and LIDAR measurements,” For. Ecol. Manage. 260(10), 1843–1852 (2010). [CrossRef]
19. Y. Chen, E. Räikkönen, S. Kaasalainen, J. Suomalainen, T. Hakala, J. Hyyppä, and R. Chen, “Two-channel hyperspectral LiDAR with a supercontinuum laser source,” Sensors (Basel) 10(7), 7057–7066 (2010). [CrossRef] [PubMed]
20. B. Hapke, Theory of Reflectance and Emittance Spectroscopy (Cambridge University Press, 1993).
21. T. J. Papetti, W. E. Walker, C. E. Keffer, and B. E. Johnson, “Coherent backscatter: measurement of the retroreflective BRDF peak exhibited by several surfaces relevant to ladar applications,” Proc. SPIE 6682, 66820E, 66820E-13 (2007). [CrossRef]
22. C. J. Tucker, “Red and photographic infrared linear combinations for monitoring vegetation,” Remote Sens. Environ. 8(2), 127–150 (1979). [CrossRef]
23. J. Penuelas, I. Filella, C. Biel, L. Serrano, and R. Save, “The reflectance at the 950–970 nm region as an indicator of plant water status,” Int. J. Remote Sens. 14(10), 1887–1905 (1993). [CrossRef]
24. D. Haboudane, J. R. Miller, E. Pattey, P. J. Zarco-Tejada, and I. B. Strachan, “Hyperspectral vegetation indices and novel algorithms for predicting green LAI of crop canopies: Modeling and validation in the context of precision agriculture,” Remote Sens. Environ. 90(3), 337–352 (2004). [CrossRef]
