A scanning polarized lidar was used to detect flying honey bees trained to locate buried land mines through odor detection. A lidar map of bee density shows good correlation with maps of chemical plume strength and bee density determined by visual and video counts. The co-polarized lidar backscatter signal was found to be more effective than the cross-polarized signal for detecting honey bees in flight. Laboratory measurements show that the depolarization ratio of scattered light is near zero for bee wings and up to 30% for bee bodies.
©2005 Optical Society of America
At the current removal rate it would take about 450 years to rid the world of anti-personnel land mines, which kill approximately 15,000–20,000 people each year in roughly 90 countries (even if no new mines are installed) . Actual and suspected land mines also inhibit economic recovery of afflicted regions by removing land from productive use. There is a tremendous need for land-mine detection methods with improved accuracy, speed, and safety compared with operational techniques based primarily on metal detection devices swept by hand over a test area. Research is being conducted on technologies including but not limited to electromagnetic induction, ground-penetrating radar, infrared and hyperspectral imaging, acoustic and seismic methods, electrical impedance tomography, nuclear quadrupole resonance, X-ray scattering, neutron technologies, electrochemical methods, contact methods, and biological techniques . Perhaps the most common biological technique is the use of dogs who locate buried land mines through smell. Dogs, however, usually are kept on a leash and are heavy enough to explode mines, thereby placing both the dog and handler in jeopardy. Similarly, artificial biological systems that locate explosives through a chemical signature usually must be carried somehow into the test region, endangering the sensor and its platform.
One of the most recently proposed biological detection techniques is the use of honey bees to locate buried land mines through odor detection. This technique is particularly useful in humanitarian de-mining where time can be taken to locate buried mines that emit plumes of chemical into the soil and air. Bees do not cause mines to explode, do not require a handler, and can be trained more rapidly than dogs. This technique makes use of the natural foraging behavior of bees, which frequently cover ranges up to several km around a hive. The bees identify the sample location by their increased dwell time while flying in its vicinity.
Bromenshenk et al. [2,3] have developed techniques to train bees by injecting trace amounts of a target chemical into feeders, which encourages the foraging bees to seek sources of food with the same smell. When the bees locate a vapor plume of the same odor, they tend to fly along the plume to its source, where they pause before continuing. Bees can be trained in one or two days to seek out buried explosives because of their high odor sensitivity, in the low parts-per-trillion (pptr), comparable to that of dogs . Bromenshenk et al. [2,3] demonstrated the ability of bees to locate, within one hour, 2,4-dinitrotoluene (DNT) with pptr vapor concentrations, at a distance of approximately 100 m from the hive. (2,4-DNT is a TNT synthesis byproduct that emits a stronger vapor plume than TNT. TNT is found in approximately 90% of all mines, making it a useful marker ). The vapor concentration in mine fields is typically 0.01–100 pptr, well within the detection capabilities of honey bees . In controlled field experiments, bees have located ppb-to-pptr targets with 97–99% detection probability, 1.0–2.5% false-positive probability, and less than 1.0% false-negative probability.
Practical use of bees in mine detection requires a method of monitoring bee locations and dwell time throughout a test region, with significant operator stand-off distance. Bromenshenk et al.  recognized that light detection and ranging (lidar) offered a potential solution to this problem. However, although it is well known that insects are a dominant source of clear-air signal for many radars [5,6], insects do not often enter into discussions of lidar signals. In fact, the lack of lidar returns is used to decide that cloud radar returns are from insects rather than from boundary layer clouds  (perhaps this algorithm should be revisited).
In 2002 a team from Sandia National Laboratory demonstrated that a lidar transmitting 30 pulses per second of 355-nm light (1–40 mJ per pulse) was able to detect bees from a distance of 1.33 km by pointing the lidar beam over the top of a feeder . The Sandia team successfully detected clusters of bees over the feeder, but did not observe bees in transit between the hive and the feeder. The primary limitation was the fixed height of the laser beam at approximately 60 cm above the ground, while the bees flew in transit closer to the ground.
In the summer of 2003 we conducted a blind field trial at an active mine field with a horizontally scanning lidar to measure bee density as a function of time and space over the mine field and an adjacent mine-free control region . This experiment showed detection of individual bees in flight and demonstrated that the lidar measured the highest bee density near the maxima found by visual observations and nearby cameras, which also correlated well with regions of high vapor plume density measured by chemical sampling. We also investigated depolarization of the scattered light as a potential mechanism for obtaining enhanced scattering information to identify bees from clutter. The remainder of this paper describes the mine field bee lidar experiment, a comparison of lidar-measured bee density with in-situ bee counts and chemical sampling, and laboratory measurements of laser scattering from bees.
2. Bee lidar measurement technique
From 25 July to 5 August, 2003, we conducted an experiment at Fort Leonard Wood, Missouri, to demonstrate bee detection of buried land mines and remote measurement of bee density with a scanning lidar system. The lidar we used was developed originally for nonscanning airborne measurements of fish . Figure 1 illustrates the lidar system configured in a horizontally scanning mode for this experiment. A frequency-doubled Nd:Yag laser emits 532-nm green light with 100-mJ pulse energy at a repetition rate of 30 pulses per second, and backscattered light is received through a linear polarizer and collected with a 17.5-cm-diameter refractive telescope through a field stop (FS), lens (L), and 1-nm bandpass interference filter (F), onto a photomultiplier tube (PMT). The PMT output current is converted to a voltage, amplified with a logarithmic amplifier to increase dynamic range, and digitized with 8-bit resolution at a rate of 1 Gsample/s. The lidar is operated such that the PMT does not saturate (except for brief instances with unwanted hard targets). In this experiment the lidar was mounted on a horizontal plate that pivoted at a point near the telescope aperture to allow horizontal scanning with an adjustable elevation angle. The scanner was driven with a dc motor and the angle encoded via the resistance of a potentiometer attached to the scanning mechanism. The full scanning lidar system was mounted inside a small utility trailer, looking through an opening in the side wall.
The lidar scanned over an 83-m-long mine-free control area to the 44-m × 24-m active test area (Fig. 2). The horizontal scan rate was approximately 25.5 mrad (1.46 degrees) per second with a full-angle beam divergence of 3.5 mrad and a pulse repetition rate of 30 pulses per second. This resulted in approximately 4.4 independent laser pulses within the transmitted beamwidth at the front of the mine field. The laser spot diameter varied approximately from 30 cm at the front of the mine field to 45 cm at the back of the mine field. Each 10-ns laser pulse was sampled 109 times per second, producing 15-cm sample resolution in range with 3-m-long pulsewidth-limited range bins (range is determined with threshold crossings). This combination of transverse and range resolution generated sample volumes measuring approximately 15 cm long (range) × 30 cm wide × 30 cm high at the front of the mine field.
The polarization of the lidar receiver is set by a manually rotated polarizer sheet that covers the full telescope aperture, providing linear polarization purity better than 1%. We alternated polarization states during data collection, especially early in the experiment, to see if any advantage existed in receiving co-polarized or cross- polarized scattered light (i.e., light polarized parallel to or perpendicular to the transmitted polarization axis, respectively). This lidar often is used in cross-polarization mode for airborne fish studies to take advantage of the high contrast between scattering from fish and from within the water ; however, we found that bees were most effectively detected with the lidar in co-polarized mode.
The most significant operational limit in this experiment was the need to have a clear line of sight from the lidar to every point in the mine field. The field was relatively flat, but had a high point near the middle and rolled off gently to the back and front and to each side. The beam had to be swept over the ground at a high enough level to avoid the high spot, thereby causing the height of the bottom edge of the beam to vary over a range of approximately 18–60 cm. This is significant because bees tend to fly quite near the surface. Furthermore, because a direct-detection lidar cannot distinguish between scattered signals from bees and vegetation, we had to mow the grass throughout the field at the start of the experiment (the mines were live, but not fused). Clearly this is a massive impediment to operational deployment of direct-detection lidars in de-mining operations, but it allowed us to demonstrate the potential of optical detection of bees for locating mines. Research is presently under way to develop much more robust, bee-specific laser-based detection systems.
3. Lidar measurements of bee density
Our first bee lidar measurements were made with the beam pointed above feeders that attracted a large number of bees (the feeders use syrup with 2,4-DNT to condition the bees to hunt for land mines [2,3]). With the bottom of the lidar beam pointed 30 cm above the feeders at a range of approximately 90 m, we alternately collected data in co-polarized and cross-polarized modes to determine if there is an advantage of one mode over the other for detecting bees. The results, shown in Fig. 3 as a graphical time series of bee detections, show that more than twice as many bees are detected with co-polarized light than with cross-polarized light (repeated several times to confirm consistency). Note that Fig. 3 shows a higher bee density than was observed in the open mine field. This is definitely not a process of locating a swarm of bees, but rather one of accumulating bee-density statistics over a time period ranging from tens of minutes to hours.
With the lidar beam pointed over the feeder (as in Fig. 3), we collected co-polarized and cross-polarized data from which we investigated the relative signal strengths. Unfortunately, because this lidar does not collect the co- and cross-polarized signals simultaneously or in subsequent laser shots, we could not determine actual depolarization ratios for individual bees (ratio of cross-pol to co-pol signals). However, we did determine that a typical mean depolarization ratio was approximately 0.39 ± 0.1, calculated as the ratio of the mean cross-polarized bee signal to the mean co-polarized bee signal in several minutes of data (the signal was the photomultiplier tube current, corrected for the logarithmic amplifier response). There was also evidence of bees in the cross-polarized signal that did not exceed the background signal sufficiently to pass the threshold filter (note Fig. 3), so including these weaker cross-polarized signals could reduce the mean depolarization ratio to approximately 0.2.
We also calibrated the lidar signal radiometrically to determine bee reflectance for use in bee lidar simulations. The calibrations were obtained from measurements taken with the lidar pointing at photographic gray cards. Lidar measurements of bees at approximately the same range were used to determine the equivalent gray-Lambertian bee scattering cross section. The resulting mean value was 0.093 cm2, with a minimum of 0.070 cm2 and a maximum of 0.15 cm2. This quantity is equal to an equivalent Lambertian reflectance multiplied by the physical area presented by the bee. Thus, dividing by an assumed physical area of 0.375 cm2 (an average of 0.5 at the sides and 0.25 at the front and back) yields an equivalent mean Lambertian reflectance of 0.25 (minimum = 0.19 and maximum = 0.41).
Bee density was measured over the full mine field, with the beam sweeping over both clear air and obstructions. Once the field was mowed, these obstructions included wooden posts marking the corners of the field and video cameras on tripods that provided independent bee-density counts. Although the beam encounters the same posts and tripods repeatedly, the locations of these obstructions vary in the data because of inconsistent scan speed, resulting in smearing by one sample width, occasionally smearing up to three samples (the scanner was constructed quickly at low expense, and stability was further impaired by a fluctuating electrical generator used to power the lidar in the field).
Scanning data were processed according to the following procedure: 1) compensate for range-dependent variation of the signal (caused primarily by ringing in the log amp) by subtracting from each point the median signal computed at a given range over 2000 laser shots; 2) divide each point by the median for normalization; 4) split each 2000-shot data file into individual scans of the field (~390 shots) and mask out known post and tripod locations.
Figure 4 shows median-corrected data for one scan across the field (13 s one way). The result is a data surface that is much smoother than that of the raw data, above which the obstructions and bees appear as large and small spikes, respectively. Bees, which appear as small spikes above the remaining background signal, are detected by applying a threshold to each scan. One bee is counted for every signal spike that is above the threshold, although it is conceivable that occasionally two or more bees are within the laser beam simultaneously. The bee density, however, was visually observed to be sufficiently low that the lidar beam usually encountered single bees.
Figure 5 shows maps of (a) scanning lidar measurements of relative bee density averaged in 2-m bins, (b) chemical plume measurements measured by a team from Sandia National Laboratory , and (c) visual and video camera bee counts . The lidar bee data shown here are averaged over 12 hours on two subsequent days (similar patterns exist in one-hour data files and in data from other days). The lidar data exhibit peaks in most of the same locations as the chemical sampling, indicating that over time the lidar successfully detected bees that did indeed tend to cluster preferentially in regions of strong explosive plumes.
A particularly encouraging result occurred when real-time observations of lidar data on the computer screen led to the discovery of an unsuspected chemical hotspot inside the supposedly mine-free control area. Chemical sampling in this area, approximately 20–25 m from the mine field (to the left of Fig. 5), initiated because of the lidar data, found significant concentrations of TNT, 2.4-DNT, and 4-amino DNT . This successful blind location of unknown explosives through lidar bee detection promotes confidence in the technique.
There are also important differences between the lidar bee density and chemical plume maps. For example, neither the lidar nor the visual/video bee counts identify clearly plume #5, and the visual/video counts do not identify plume #1. However, plume #5 was detected only briefly, whereas the other plumes were detected consistently; therefore, plume #5 is believed to be contamination brought in during later phases of the experiment after the lidar data collection was complete. A factor that may have contributed to the weak lidar signal at plume #1 is that this is approximately the lowest point in the field, where the lidar beam swept high above where the bees tend to fly most often. Plume #7 is located in the vicinity of a cluster of video camera tripods, a region around which had to be masked out of the lidar data. A lidar peak near the location of plume #3 is displaced by several meters (up in Fig. 5), possibly resulting from the lidar detecting bees as they fly up the plume toward the source (the bees would be below the lidar beam at the source).
Noting these conditions, Fig. 5 indicates that the lidar successfully located the highest bee density in nearly all of the same regions as the visual and video counts (but did so from a safe distance), and the regions of higher lidar bee density correlate reasonably well with regions of high chemical plume strength (it is important to recognize that neither the chemical sampling nor the visual/video counts covered the full area with the spatial resolution or temporal continuity of the lidar). The correlation coefficient is 0.29 between the Fig. 5a lidar bee map (averaged into 4-m bins) and chemical plume map, and increases to 0.38 when plumes # 5 and 7 are removed from the chemical map (recall that plume #5 is questionable and plume #7 was masked out from the lidar data). For reference, the correlation coefficient between the chemical map and the visual/video bee-count map is 0.52. This analysis is encouraging, but also clearly indicates the need for an optical sensor with a more unique bee-detection scheme that will allow a laser beam to be swept closer to the ground where the majority of bees fly.
Throughout the experiment we observed the bees carefully, looking for evidence that the laser beam either harmed them or caused avoidance. The only time we saw any reaction from the bees was when we shined the pulsed laser beam directly on the hive, presumably causing a mechanical vibration that disturbed the bees. Whenever we pointed the beam directly at the hive, agitated bees emerged rapidly. The bees returned to their routine behavior almost immediately after the beam was removed. When we shined the beam directly over a feeder, the bees continued feeding and flying through the beam with no apparent concern. Similarly, we never observed any reaction that indicated that the laser beam harmed or blinded bees.
4. Laboratory measurements of bee wing and body depolarization
Having observed that the lidar preferentially detects flying bees with co-polarized radiation, we returned to the laboratory (at Montana State University) and conducted scattering measurements to determine if there was a significant difference in depolarization for scattering from bee bodies and bee wings. We illuminated individual bee bodies and wings with a continuous-wave Nd:Yag laser operating at 532-nm wavelength and measured the scattered light in the near-backward direction (collected through a 15-cm-diameter lens) with an optical power meter. The transmitted light was expanded to a diameter of approximately 1.5 cm (large enough to illuminate a large fraction of a bee body or wing without significantly illuminating the mount or background) and passed through a linear polarizer with polarization purity of 105:1 to provide an illuminating beam with essentially perfect polarization. The backscattered light was passed through an analyzer with linear-polarization purity of 104:1.
Figure 6 is a top-view diagram of the orientation of bee bodies and wings showing the definition of the scattering angle in these experiments. Figure 7 shows photographs of a bee body viewed at θ = 90° and a bee wing viewed at 0° (but the wings were mounted with their top edge more horizontal than shown in Fig. 7).
In the scattering measurements, bee bodies were illuminated at angles of 0–360°, with 0° defined as where the head faces the laser. The bee bodies were rotated about a vertical axis, allowing the laser to see the bee head, followed by the side, the tail, and the other side (the bees were not viewed from below or from above). Wings were illuminated over an angular range of ±70°, where 0° is defined as the angle where the surface normal at the center of the broadest wing surface faces the laser. Wings were oriented with their long axis horizontal and their short axis vertical, and were rotated about the short vertical axis. The mean background signal, measured with the body or wing removed from the setup, was subtracted from each measurement. A depolarization ratio was calculated as the ratio of cross-polarized to co-polarized signals after background subtraction.
Figure 8 shows the measured depolarization ratio, plotted versus angle (defined above), for (a) a bee body and (b) a bee wing. These results are fairly typical of those obtained with different wings and bodies (although recently we repeated these experiments with fresher bee bodies and an improved procedure, consistently finding higher body depolarization, near 0.3 peak, but with the same angular pattern). Generally, we observed in the laboratory measurements approximately 15–30% depolarization for bee bodies and 1–10% for bee wings. for bodies, the highest depolarization occurred when illuminating the side of the body and the lowest occurred when illuminating the front and back.
For wings, the highest depolarization occurred when illuminating the surface at large grazing angles and the lowest occurred at normal incidence. The wing depolarization is nearly zero over the majority of the illumination angle range, rising to 10% only at the extreme angles near ± 70°. As shown in Fig. 7, the wing is a rather smooth surface that exhibits a somewhat specular reflection. The wing also curves in two dimensions, providing near-normal incidence over a wide range of wing-orientation angles. This leads to a slower rise of depolarization for positive orientation angles (rotating to illuminate the curved end) than occurs for negative angles. It is not surprising that the fuzzy bodies generate higher depolarization than the rather smooth wing surface, which has a quasi-specular nature that is apparent in Fig. 7.
5. Discussion and conclusions
The field measurements described here show that the scanning lidar consistently detects higher bee density near most of the significant chemical plumes, although the actual correlation coefficients are fairly low between the lidar bee-density image and the chemical plume map (0.29). In visual observations it was clear that bees flew below the beam fairly often, so that the ability to sweep a beam closer to the ground will be necessary to increase the number of bee detections and to improve the spatial accuracy of the detected hot spots. In this experiment we swept the beam as low over the ground as possible without encountering scattering signals from the ground or low vegetation. However, even after the vegetation was mowed, this prevented the brightest part of the laser beam from sweeping closer than about 20 cm above the surface.
These early field measurements show clearly that a bee-specific detection mechanism is needed so that the laser beam can be swept closer to the ground and so that bees can be detected as they fly through vegetation. Bender et al.  at Sandia National Laboratory also recognized this limitation and suggested dusting the bees with a fluorescent dye and detecting the bees with a fluorescence lidar.
The measurements discussed here suggest that polarization alone is insufficient as a bee-identification mechanism. While vegetation can have somewhat higher depolarization, there does not appear to be adequate polarization contrast to allow polarization discrimination of bees from vegetation. However, the co-polarization preference observed in the bee lidar field measurements led us to wonder if perhaps wing reflections might be the dominant scattering mechanism of a bee lidar signature. If so, then perhaps a wing-beat modulation scheme would provide a more bee-specific detection mechanism. A primary focus of the bee-detection research at Montana State University is now on developing sensors using this sort of technique, and preliminary results suggest that this will be a much more useful detection mechanism than direct-detection lidar (3-D imaging lidar could be used to overcome some of the limitations of a 2-D scanning lidar, but still does not provide a bee-specific signal).
The combination of laboratory depolarization measurements and the relatively high value of depolarization found in the field bee lidar data do not provide a clear answer to the question of whether body or wing scattering dominates. The lidar depolarization ratio measured in the field data was high enough (even if reduced by additional weak cross-polarization signatures as discussed in section 3) to suggest that the dominant scattering was from bee bodies in the mine field lidar measurements. (This is still consistent with the co-polarized preference because even the bodies still scatter mostly co-polarized light.) Nevertheless, the possibility of significant scattering from wings is not ruled out by this result. Given the reasonably specular nature of the wing surface, it seems likely that the dominant wing scattering may occur in the form of glints (bright specular reflections). For wings oscillating at 200 Hz, a glint should occur once every 5 ms; the pulsed laser used in the field experiments generated 30 pulses/s, each lasting only 10 ns. Therefore, with the pulsed laser on for only 0.3 μs each second and off for the vast majority of the time, there is only about one chance in 104 that a wing will be oriented at the glint angle when the laser is on (neglecting the further complication of angular bee orientation). Nevertheless, current research (to be reported elsewhere) shows that wing-beat modulation is indeed a very viable detection mechanism if the laser duty cycle is sufficiently high. Future research will be directed at developing this and other bee-specific detection mechanisms, while continuing measuring bee scattering to further elucidate body and wing reflectance and scattering patterns.
We gratefully acknowledge the Montana State University Office of Research and Creative Activities, the University of Montana, and S & K Electronics for providing financial support for the bee lidar experiment, the Montana State University Undergraduate Scholars Program for funding the participation of undergraduate students in bee lidar data processing and laboratory depolarization studies, Dr. Philip Rodacy at Sandia National Laboratory (Albuquerque, New Mexico) for providing the chemical sampling data and valuable guidance about the explosives chemistry, Hector Bravo at the NOAA Environmental Technology Laboratory (Boulder, Colorado) for building the lidar scanner used in the field measurements, Jesse Way (MSU-Bozeman) for repeating the bee depolarization measurements with fresh bees for validation of prior results, and the support staff at Ft. Leonard Wood who made the field experiments possible by providing help that included mowing the mine field. The wing-beat-modulation laser sensor mentioned in this manuscript is under development at Montana State University by J. A. Shaw, K. S. Repasky, J. A. Carlsten, and L. H. Spangler.
References and links
1. J. MacDonald, J. R. Lockwood, J. McFee, T. Altshuler, T. Broach, L. Carin, R. Harmon, C. Rappaport, W. Scott, and R. Weaver, “Alternatives for landmine detection” (RAND Corp., 2003), http://www.rand.org/publications/MR/MR1608/MR1608.appg.pdf.
2. J. J. Bromenshenk, C. B. Henderson, and G. C. Smith,,”Biological Systems, paper II,” in Alternatives for landmine detection (Rand Corp., 2003), http://www.rand.org/publications/MR/MR1608/MR1608.apps.pdf.
3. J. J. Bromenshenk, C. B. Henderson, R. A. Seccomb, S. D. Rice, R. T. Etter, S. F. A. Bender, P. J. Rodacy, J. A. Shaw, N. L. Seldomridge, L. H. Spangler, and J. J. Wilson, “Can honey bees assist in area reduction and landmine detection?,” J. Mine Action 7.3 (2003), http://www.maic.jmu.edu/journal/7.3/focus/bromenshenk/bromenshenk.htm.
4. Philip J. Rodacy, Explosives Technologies Group, Sandia National Laboratory, P.O. Box 5800, MS 1452, Albuquerque, NM 87185–1452 (personal communication, 2005).
5. J. W. Chapman, D. R. Reynolds, and A. D. Smith, “Migratory and foraging movements in beneficial insects: a review of radar monitoring and tracking methods,” Int. J. Pest Mgmt 50, 225–232 (2004). [CrossRef]
6. B. E. Martner and K. P. Moran, “Using cloud radar polarization measurements to evaluate stratus cloud and insect echoes,” J. Geophys. Res. 106, 4891–4898 (2001). [CrossRef]
7. S. F. A. Bender, P. J. Rodacy, R. L. Schmitt, P. J. Hargis, M. S. Johnson Jr., J. R. Klarkowski, G. I. Magee, and G. L. Bender, “Tracking Honey Bees Using Lidar (Light Detection and Ranging) Technology,” Sandia Report SAND2003-0184 (2003), Sandia National Laboratory, Albuquerque, NM 87185.
8. J. H. Churnside, J. J. Wilson, and V. V. Tatarskii, “Airborne lidar for fisheries applications,” Opt. Eng. 40, 406–414 (2001). [CrossRef]