Aerial Detection of Surface Unexploded Explosive Ordnance (UXO) in the Mid- and Long-Wave Infrared

Transcription

1 Aerial Detection of Surface Unexploded Explosive Ordnance (UXO) in the Mid- and Long-Wave Infrared Josée Lévesque Defence Research and Development Canada Valcartier Québec City, QC, Canada Abstract The feasibility of using off-the-shelf broadband thermal IR cameras, (3-5 micron and 8-12 micron) for the aerial detection of unexploded explosive ordnance (UXO) was investigated. To do so, it was necessary to study the UXO/background thermal behavior to determine the best image acquisition parameters such time of day, altitude, and sky condition. The study was broken down in two phases: (1) target background characterization over two 24 hour periods (clear and cloudy sky) with constant pixel size, and (2) acquisition of airborne images at various altitudes over different background types and at different times of day. Results showed that the highest contrast between target and background was observed in late afternoon and early evening except for fully buried projectiles which showed a higher contrast near the highest sun elevation. A multi-altitude survey provides the best compromise between a large field of view and a high spatial resolution. The sky condition (clear or cloudy) appears to matter for the vegetation background but not for the mud background, although the 8-12 micron camera performs best under a clear sky. 1 Introduction Unexploded explosive ordnance (UXO) have been known as an environmental, health and safety hazard as far back as the First World War. It is an ongoing worldwide problem that continues to evolve as conflicts around the globe result in more UXO being deposited than recovered. Cleanup is difficult because of the large areas affected by UXO and the threat they pose to those who are involved in their removal. Aside from the fact that the shells can still explode, UXO can undergo corrosion and become a source of environmental toxicity as the explosives and propellants often leak and remain in the environment. UXO surveys are typically the first step in the removal process, and can involve visual detection from people walking on the ground, ground-based sensors such as ground penetrating radar (Andrews et al., 1999; Arcone et al., 2000) and electromagnetic sensors (Zang et al., 2003; O Neill et al., 2005), and/or airborne sensors such as ultra-wideband synthetic aperture radar (Sullivan A., 2002; Sullivan et al., 2000). No single sensor can detect UXO in all possible conditions (buried, wet, under water) with a zero false alarm rate. It is instead a combination of different sensors that provides the best approach (Shamatava et al., 2004; Marble et al., 2000). The Spatial and Geospatial Exploitation Section of Defence Research and Development Canada is supporting the (Department of National Defence) DND UXO Legacy Site Program by investigating the feasibility of using off-the-shelf broadband thermal infrared (TIR) cameras in the 3 to 12 micron spectral range for the aerial detection of surface and partially buried UXO. The Nicolet firing range on the shore of Lac Saint-Pierre was selected for the experiments because of its well-known UXO problem. Moreover, because the lake

2 water level fluctuates and vegetation quickly grows over barren soil after water retreats, surveys of the area are incomplete. Airborne detection could aid in solving this problem. In order to understand the thermal UXO/background behavior and determine the best image acquisition parameters (time of day, altitude, sky condition), the study was broken down into two phases: Phase I: This phase has been completed and consisted of determining the thermal characterization of UXO and potential false alarms in various background types (mud, litter, grass, water) and conditions (surface, partially covered). The sensing was performed over two 24 hour periods (clear and cloudy sky) using a set of two towermounted TIR cameras, one 3-5 micron and the other 8-12 micron, to record the thermal radiance images. Pixel size was kept constant at 0.5cm. Phase II: This consisted of the acquisition of airborne images over the different background types and at different times of the day using the two TIR cameras on board a helicopter. Altitude was varied to produce pixel sizes varying between 0.5 cm and 15 cm. This phase is still in progress and only preliminary results are available. A third phase is projected if results from Phase II are deemed satisfactory. It will consist of adding mapping and automatic detection capabilities to the system and evaluating its performance on an operational basis. This paper presents a summary of the results for the various acquisition conditions (time of day, weather), the camera setups (altitude, wavelength) and the target/background types and conditions. 2 Methodology 2.1 Test Site The Nicolet firing rage is located on the south shore of Lac Saint-Pierre near Nicolet, Quebec. (Figure 1). The Nicolet site was selected because of the large number of inert projectiles (IP) and UXO that remain to be recovered. It is estimated that of the projectiles fired toward the lake between 1952 and 2000 have yet to be found. Although most projectiles are inert, an estimated are believed to be UXO. Initial observations at the Nicolet range facility have helped to categorize three backgrounds in which inert projectiles (IP) and UXO can be found: lake sediments or mud, vegetation and water. These backgrounds are located in the active shore zone of Lac Saint-Pierre, where water level fluctuates and where UXO are most likely to found since the land side has been intensively surveyed over the years by people on the ground.

3 Figure 1. Location map of the Nicolet firing range. 2.2 Camera System and Setup Table 1 shows the specifications of the cameras used in Phase I and Phase II. The Mitsubishi IR-M700 camera records the radiance in the 3-5 micron range and has a field of view of 22 deg X 22 deg. The Nytech Micro Bolometer camera records the radiance in the 8-12 micron range and has a field of view of 20 deg X 15 deg. Both cameras produce a frame size of 640 X 480 pixels. In Phase II, a visible and Near-IR camera was also used as a supplementary validation dataset for the detection results. It was equipped with a 6 position filter wheel with an empty space and five 0.02 μm width filters centered at 0.56 μm, 0.6 μm, 0.68 μm, 0.75 μm, and 0.95 μm. Table 1. Cameras specifications. Camera Wavelength FOV* Frame size (micron) (degree) (pixel) Phase I & II Mitsubishi IR-M700 3 to 5 22 X X 480 Nytech Micro Bolometer 8 to X X 480 Phase II VNIR** Lumenera 0.56 to X X 1032 *Field of view ** Visible Near Infrared In Phase I, the two cameras were mounted on a metal plate that was hooked on the metal hand bar of the observation tower OP2 at the Nicolet firing range (Figure 2). The distance between the camera lens and the ground was 8 m and produced a pixel size of about 0.5 cm.

4 Figure 2. Camera and target setup for Phase I. In Phase II, the three cameras were fixed to a metal plate which was mounted on a Tiler mount (Figures 3a, 3b). The Tiler mount was then affixed to the back seat of the helicopter so that the operator was inside and the camera sensor outside enough to allow a nadir view of the ground (Figures 3c, 3d). The use of a helicopter enabled full control of the altitude. Table 2 provides a list the airborne acquisitions performed in Phase II and their specifications. Presented at the 33rd AMOP Technical Seminar on Environmental Contamination and Response, Halifax, Nova

5 (a) (b) 3-5 micron camera 8-12 micron camera 6 bands VNIR camera VNIR filter wheel (c) (d) Figure 3. Camera mount and airborne platform for Phase II Table 2. List of airborne acquisitions in Phase II. Acquisition Date Time Altitude - pixel size (cm) Sky condition 38 m 76 m 182 m 365 m 548 m (0.5cm) (2cm) (5cm) (10cm) (15cm) 1 July :30 Cloudy x x x 2 July :00 Mostly sunny x x x 3 July :00 Mostly sunny x x x x x 4 July :30 Mostly sunny x x x x x 5 July :30 Sunny breaks x x x x x 2.3 Targets and Backgrounds Layout In Phase I, the two cameras looked down at different targets (inert projectiles, other metal objects) in various backgrounds (vegetation, soil, litter, water) (Figure 2). Images were recorded every 15 minutes over a 24-hour period under a mainly clear sky and over a 16-hour period under a mainly cloudy sky. Table 3 lists the targets and

6 their backgrounds as well as the condition in which the targets were exposed (on surface, covered, partially buried, buried). Because of the limited space available, plastic containers were used to hold the targets and sample backgrounds. In Phase II, one test site of approximately 5m X 20m was established within each background type. The four test sites were divided into 1m X 1m grid cells, and within each cell a target was positioned. A total of 98 inert projectiles were positioned on the ground within the four background sites. Target type, size and orientation were varied. False alarm targets were also positioned within each test site. They were made of various material (metal, plastic, glass, rubber) and represent common objects found on the shore of Lac Saint-Pierre. A summary of the elements being varied in the experiment is provided in Table 4. Table 3. List of targets used in Phase I. Target Target Background Target Target Target Background Target # type type condition # type type condition 1 IP plant litter Partly covered 22 Al plate Green grass Exposed 2 IP Dry grass Exposed 23 IP part Green grass Exposed 3 IP Dry grass Partly covered 24 IP part Green grass Exposed 4 IP plant litter Exposed 25 IP part Green grass Exposed 5 IP Mud Exposed 26 IP part Green grass Exposed 6 IP Mud Covered 27 IP part Green grass Exposed 7 IP Mud Exposed 28 IP part Green grass Exposed 8 IP Mud Covered 29 IP part Green grass Exposed 9 IP Mud Buried 30 IP part Green grass Exposed 10 IP Mud Buried 31 IP Green grass Exposed 11a Al can Green grass Covered 32 IP part Mud Exposed 11b Al can Green grass Partly covered 33 IP part Mud Exposed 11c Al can Green grass Exposed 34 IP part Mud Exposed 12 IP plant litter Exposed 35 IP part Mud Exposed 13 IP Green grass Exposed 36 IP part Mud Exposed 14 IP plant litter Covered 37 IP part Mud Exposed 15 IP Green grass Covered 38 IP part Green grass Exposed 16a Al can Mud Buried 39 IP part Green grass Exposed 16b Al can Mud Covered 40 IP part Green grass Exposed 16c Al can Mud Exposed 41 IP fragment Green grass Exposed 17 IP Water Exposed 42 IP fragment Green grass Exposed 18 IP Water Exposed 43 IP Water Covered 19 IP Water Exposed 44 IP Water Covered 20a Al can plant litter Exposed 45 IP fragment Green grass Exposed 20b Al can plant litter Partly covered 46 IP fragment Green grass Exposed 20c Al can plant litter Covered 47 IP fragment Wood Exposed 21 Al plate Green grass Exposed 48 IP fragment Wood Exposed

7 Table 4. Elements that were varied in Phase II. Background type Background condition Projectile size Projectile condition Projectile orientation False alarms Mud, green grass, dry grass, plant litter Dry, wet, moist 155mm, 105mm, 3''/50, fragments Surface, covered, buried 0 deg, 45 deg, 90 deg Various metal, rubber, wood, rocks, plastic, glass 2.4 Data Manipulation and Processing Each image frame from each camera collected every 15 minutes produces a separate TIF file. In Phase I, a total of 64 frames were collected with each camera during the 16-hour cloud free period and 96 frames were collected during the 24-hour cloudy period. Tests were performed to ensure that the gains were appropriate and no offsets were present. Because the cameras were motionless it was possible to overlay all the images of an acquisition period to form an image cube in which the third dimension represents time. Four image cubes were produced: one for each camera and each acquisition period. It was then possible to extract the target and background pixels at the exact same locations in all the frames. A total of 100 to 500 pixels were averaged from each target and each background. The absolute difference between the targets and their respective background was calculated to characterize the thermal contrast. The difference was then normalized as a percentage of the 8-bit grey level range of the images. The last two steps were also performed on the airborne images of Phase II to extract the target and background differences. 3 Results Results of Phase I are summarized in tables 5 and 6 for the 3-5 µm and the 8-12 µm cameras respectively. In black are the differences between the target and the background that are greater than 70%. The darker and lighter shades of grey represent the differences between 40% and 70% and 10% and 40% respectively. The differences below 10% are in white. Based on a Z-test all differences above 10% are statistically significant at the P <0.05 level. Sky conditions are also given for the 16- hour period covered by the two acquisition period (mainly clear and mainly cloudy). Clear sky appears in white, a thin layer of clouds in grey and a cloudy sky in black. 3.1 Effect of Imaging Time and Sky Conditions When using the 3-5 µm camera, except for IP in water, all exposed IP produce a greater than 40% average difference at some point in the day under cloudy or clear sky (Table 5). Most of the higher differences occur from late afternoon on. In a mud background, having a clear versus cloudy sky does not matter, while a mainly clear sky produces higher differences earlier and longer in the day in vegetation backgrounds. Although the 8-12 micron camera system does not produce as good results as the 3-5 micron overall, it was found useful to detect the IP covered by vegetation in

8 clear sky conditions throughout the day (Table 6). Moreover, still in clear sky conditions, IP surrounded by water constantly produce average differences in the 10-40% range throughout the day except between 10:00 and 15:00 in water and straw background. Figure 4 shows the effect of imaging time in the 3-5 micron images acquired during Phase II at 38 m over the mud background. At 10:00 the thermal contrast between the targets and backgrounds ranges between 5% and 10%, increases to 20% to 25% at 15:00 and reaches 29% to 33% at 19:30. A similar behavior is observed in the vegetation backgrounds (litter, grass) and with the 8-12 micron camera indicating that the natural backgrounds cool down faster than the metal targets after the sun stops illuminating the scene. Increased target/background contrast Warm 10:00 15:00 19:30 Cold Figure micron images acquired at 38 m over the mud background at three difference time of the day. 3.2 Effect of Altitude Table 7 shows the field of view (FOV) and the resulting pixel size for the various altitudes at which the camera system was flown in Phase II. As the pixel size increases so does the FOV. This is illustrated in Figure 5 with the multi-altitude images from the 8-12 micron camera taken over the mud background at 19:30. The 182 m image with 5 cm pixel size still shows the shape of the projectiles but not the 365 m and 548 m images in which only a higher contrast is perceived. This low level of spatial detail will contribute to an increase in the false alarm rate. Thus, for large area mapping, the best compromise will probably be a multi-altitude survey in which a first pass at a higher altitude will produce a base anomaly map that will be used to conduct the subsequent detailed surveys.

9 Table 5. Phase I results for the 3-5 micron camera Background Sky condition Projectile size & condition 6:00 7:00 8:00 9:00 10:00 11:00 12:00 13:00 14:00 15:00 16:00 17:00 18:00 19:00 20:00 21:00 22:00 Target / Background Difference MUD Mainly cloudy Mainly sunny Small exposed Large exposed Small covered Large covered Small burried 0.5 cm Small burried 2 cm Small exposed Large exposed Small covered Large covered Small burried 0.5 cm Small burried 2 cm > 70% 40-70% 10-40% < 10% Sky condition Overcast Thin clouds Clear VEGETATION WATER Mainly cloudy Mainly sunny Mainly cloudy Mainly sunny Small exposed on plant litter Large exposed on plant litter Small covered by plant litter Large partly covered by plant litter Small exposed on green grass Large exposed on dry grass Small covered by green grass Large partly covered by dry grass Small exposed on plant litter Large exposed on plant litter Small covered by plant litter Large partly covered by plant litter Small exposed on green grass Large exposed on dry grass Small covered by green grass Large partly covered by dry grass Small exposed on w ater & mud Small exposed on w ater & plant litter Small exposed on w ater & grass Small 1 mm under w ater Small 1cm under w ater Small exposed on w ater & mud Small exposed on w ater & plant litter Small exposed on w ater & grass Small 1 mm under w ater Small 1cm under w ater SKY CONDITION July 14 - mainly cloudy July 18 - mainly clear

10 Table 6. Phase I results for the 8-12 micron camera. Background Sky condition Projectile size & condition MUD Mainly cloudy Mainly sunny Small exposed Large exposed Small covered Large covered Small burried 0.5 cm Small burried 2 cm Small exposed Large exposed Small covered Large covered Small burried 0.5 cm Small burried 2 cm 6:00 7:00 8:00 9:00 10:00 11:00 12:00 13:00 14:00 15:00 16:00 17:00 18:00 19:00 20:00 21:00 22:00 Target / Background Difference > 70% 40-70% 10-40% < 10% NA Sky condition Overcast Thin clouds Clear VEGETATION Mainly cloudy Mainly sunny Small exposed on plant litter Large exposed on plant litter Small covered by plant litter Large partly covered by plant litter Small exposed on green grass Large exposed on dry grass Small covered by green grass Large partly covered by dry grass Small exposed on plant litter Large exposed on plant litter Small covered by plant litter Large partly covered by plant litter Small exposed on green grass Large exposed on dry grass Small covered by green grass Large partly covered by dry grass WATER Mainly cloudy Mainly sunny Small exposed on w ater & mud Small exposed on w ater & plant litter Small exposed on w ater & grass Small 1 mm under w ater Small 1cm under w ater Small exposed on w ater & mud Small exposed on w ater & plant litter Small exposed on w ater & grass Small 1 mm under w ater Small 1cm under w ater SKY CONDITION July 14 - mainly cloudy July 18 - mainly clear

11 Table 7. Altitudes at which the camera system was flown in Phase II and the resulting field of view and spatial resolution. Altitude Pixel size FOV (m) (cm) (m) x x x x x 72 Warm 30 m 76 m 182 m 365 m 548 m Figure 5. Images from the 8-12 micron camera taken at various altitudes over the mud background. Cold 3.3 Covered and Buried Targets Results from Phase I (Table 5 and 6) show that large IP covered with mud and IP partially covered with vegetation produce average differences in the 20% to 70% range while IP buried in mud (5cm) and totally covered by vegetation produce differences in the 0-28% range. It is the 3-5 micron camera under a clear sky that best discriminate between buried/covered targets and their background. When IP are in contact with water differences reach an average of 34% to 51% when using the 8-12 micron camera under a clear sky. Figure 6 shows the 3-5 micron images of a mud-covered 105 mm inert projectile acquired at 38 m at 10:00, 15:00 and 19:30. A grid location marker appears in the upper left corner of the images. The marker helps to locate where the projectile should be in the 10:00 image. In Figure 7 the projectiles identified with an arrow are covered with plant litter or straw (vegetation residues from previous years) to various degrees. The 13:00 acquisition shows how this background is affected by a high sun elevation while it contrasts uniformly with the metal targets throughout the day. Figure 8 illustrates the potential of detection for buried targets. In this example, two 105 mm IP were buried under 1 cm of soil. They were not visible at the surface and remained buried for 3 hours before they were imaged at mid-day with a hand held FLIR instrument. One of the targets was oriented at 45 deg (Figure 8a) and the other one at 0 deg (Figure 8b). An arrow points at their location. They are clearly visible and appear darker (cooler) than the surrounding soil. In Phase I, the buried target (0.5cm) detections above 10% occurred before 13:00 and after 19:00. More

12 measurements at different times, depths, and projectile size are required to determine the detection limit and the best time of day to detect buried targets. Warm Grid marker 10:00 15:00 19:30 Figure 6. Mud-covered 105 mm projectile from the 3-5 micron acquired at 38 m. Cold 70% 0% 50% 30% 10:00 13:00 15:00 19:30 Figure 7. Litter-covered projectiles (0%, 30%, 50%, 70%) in the 8-12 micron images acquired at various time at an altitude of 38 m. (a) 105mm, 45 deg (b) 105mm, 0 deg Figure 8. Hand held FLIR pictures of two 1-cm buried 105 mm projectiles oriented 45 deg (a) and 0 deg (b).

13 3.4 False Alarms Caused by Other Metal Objects Figure 9 shows the radiance curves of rusted and shiny metal and aluminum. Shiny metal produces a 35% to 43% difference after noon (not shown). Since IP are very likely to be found rusted with a thin coat of iron oxide, this is important to ensure good discrimination from shiny metal. Aluminum remains colder than the IPs throughout the day. When using the 8-12 micron camera, aluminum pop cans can be mistaken for small IP, but shiny metal shows very distinctly (Figure 10) Radiance :00 08:00 10:00 12:00 14:00 16:00 18:00 20:00 22:00 Time Rusted IP on mud Al plate (ref) Shiny metal (ref) Blue Al can on mud Figure 9. Radiance curves of rusted IP, shiny metal, aluminum pop can and aluminum plate from the 3-5 micron camera Radiance :00 08:00 10:00 12:00 14:00 16:00 18:00 20:00 22:00 Time rusted shiny metal Al plate Al can Al can Figure 10. Radiance curves of rusted IP, shiny metal, aluminum pop can and aluminum plate from the 8-12 micron camera.

14 4 Conclusion The objective of this study was to investigate the feasibility of using off-theshelf broadband thermal IR cameras, one 3-5 micron and the other 8-12 micron, for the aerial detection of unexploded explosives ordnance (UXO). To do so, it was necessary to study the UXO/background thermal behavior to determine the best image acquisition parameters such as altitude, time of day and sky condition. In general, the highest contrast between the targets and their background can be observed in late afternoon and early evening as the natural backgrounds cool down faster than the metal projectiles. This was not always the case for soil buried projectiles which also exhibited a higher contrast near the sun zenith when the background is at its warmest and the target, not heated by the sun, remains cool. For large area mapping, a good compromise between a good spatial resolution and a large field of view is a multi-altitude survey in which a first pass at a higher altitude will produce a base anomaly map that will be used to conduct the subsequent detailed surveys at lower altitudes. The sky condition (clear or cloudy) appears to matter for the vegetation background but not as much for the mud background, although the 8-12 micron camera performs best under a clear sky when the targets are in contact with water. The background moisture was not addressed in this study but should be considered an important limitation. In particular immediately following a rain shower, the thermal contrast would be considerably reduced between the target and the background (Howard, 2001). Other metal objects such as aluminum and shiny metal have a more distinct thermal behavior over time than the rusted projectiles and should not be a significant source of false alarms. The third phase of this project will consist of adding mapping and automatic detection capabilities to the system and evaluate its performance on an operational basis. 5 Acknowledgement This work was funded by the DND UXO Legacy Sites Program. Many people were instrumental in the success of the data collection of Phase I and II. They were, in alphabetical order, Roger Blanchard, Michel Dupuis, Raymond Gagnon, Louis- Philippe Giguère, Stéphane Giroux, Mario Pichette, Sébastien Thibault, and Jacques Tremblay. Special collaboration from Jacques Drouin during Phase I was also greatly appreciated. 6 References Andrews, A., J. Ralston, and M. Tuley, Research on Ground-Penetrating Radar for Detection of Mines and Unexploded Ordnance: Current Status and Research Strategy, Institute for Defense Analyses, IDA Document D-2416, Arcone, S.A., K. O Neill, A.J. Delaney, and P.V. Sellman, UXO Detection at Jefferson Proving Ground Using Ground-Penetrating Radar, US Army Corps of Engineers, Cold Regions Research & Engineering Laboratory ERDC/CRREL Technical Publication TR-00-5, 2000.

15 Howard, G.B., Application of Airborne Thermal Infrared Imaging for the Detection of Unexploded Ordnance, Thermosense XXIII, Proceedings of SPIE, Vol. 4360, pp , Marble, J.A., J.G. Ackenhusen, J.W. Wegrzyn, J. Mancuso, and C. Dwan, Sensor Fusion Performance Gain for Buried Mine / UXO Detection Using GPR, EMI, and MAG Sensors, Detection and Remediation Technologies for Mines and Minelike Targets, Proceedings of SPIE, Vol. 4038, pp , O Neill. K., F. Shubitidze, K. Sun, I. Shamatava, and K.D. Paulsen, EMI Obstruction of Buried UXO by Geophysical Magnetic Permeability, Anthropogenic Clutter, and by Magnitude Disparities, Targets and Backgrounds XI: Characterization and Representation, Proceedings of SPIE, Vol. 5811, pp.87-99, Shamatava, I., F. Shubitidze, C.C. Chen, H.S. Youn, and K. O Neill, Potential Benefits of Combining EMI and GPR for Enhanced UXO Discrimination at Highly Contaminated Sites, Detection and Remediation Technologies for Mines and Minelike Targets, Proceedings of SPIE, Vol. 5415, pp , Sullivan, A., Model-based Summary Conclusions on the Use of UWB Radar for Detecting Unexploded Ordnance, Detection and Remediation Technologies for Mines and Minelike Targets VII, Proceedings of SPIE, Vol. 4742, pp , Sullivan, A., T.R., Damarla, N. Geng, Y. Dong, and L. Carin, Detection of Above Ground Subsurface Unexploded Ordnance Using Ultra-Wideband (UWB) Synthetic Aperture Radar (SAR) and Electromagnetic Modeling Tools, Detection and Remediation Technologies for Mine and Minelike Targets V, Proceedings of SPIE, Vol. 4038, pp , The DND UXO Legacy Sites Program, From Zhang, Y., L. Collins, and L. Carin, Physics Model Based Unexploded Ordnance Discrimination Using Wideband EMI Data, Detection and Remediation Technologies for Mines and Minelike Targets VIII, Proceedings of SPIE, Vol. 5089, pp (2003).