Ali Cafer Gurbuz, Waymond R. Scott Jr. and James H. McClellan LOCATION ESTIMATION USING A BROADBAND ELECTROMAGNETIC INDUCTION ARRAY Ali Cafer Gurbuz, Waymond R. Scott Jr. and James H. McClellan School of Electrical and Computer Engineering Georgia Institute of Technology
Outline Introduction Method Examples Conclusions Basis for Location Estimation Dictionary Orthogonal Matching Pursuit (OMP) Examples Simulated Laboratory Field Conclusions Scott, Georgia Tech SPIE, April 2009
Array System Used for this Work A broadband EMI system has been built Single transmit and three receive coils Broadband: 330 Hz to 90 kHz High fidelity measurements Battery Power EMI Array Head Broadband EMI Head Broadband EMI System at Field Site Scott, Georgia Tech SPIE, April 2009 3 Scott, Georgia Tech SPIE, April 2009
Introduction Broadband Electromagnetic Induction (EMI) Sensors Effective for detecting buried metallic targets Effective at discrimination of metallic targets Compatible and often used with other subsurface sensors This work is focused on estimating the locations of multiple metallic targets including their depth and cross ranges with an acceptable accuracy Estimates are valuable in themselves making it easier to remediate targets Estimates could significantly improve the results for the fusion of the EMI data with other sensor such as GPR Scott, Georgia Tech SPIE, April 2009
Outline Introduction Method Examples Conclusions Basis for Location Estimation Dictionary Orthogonal Matching Pursuit (OMP) Examples Simulated Laboratory Field Conclusions Scott, Georgia Tech SPIE, April 2009
Basis for Depth Estimation Graph of the intensity of the down track response at a single frequency with depth as a parameter when the target passes under the center of the array The relative size of the EMI response of the three array heads changes with depth The shape of the down track response changes with depth Shallow Shallow Shallow Deep Deep Deep Left Channel Center Channel Right Channel Scott, Georgia Tech SPIE, April 2009
Basis for Depth Estimation Graph of the intensity of the down track response at a single frequency with cross range is a parameter when the target passes under the center of the array The relative size of the EMI response of the three array heads change with depth The shape of the down track response changes with depth Far Right Center Far Left Far Left Far Right Far Left or Right Left Channel Center Channel Right Channel Scott, Georgia Tech SPIE, April 2009
How to use this info? To use the information we followed a basis pursuit method: Discretize the target space Generate EMI target response model for each possible target space point, generating a dictionary of responses Use orthogonal matching pursuit to extract the optimal representation of the received signal in terms of the dictionary elements finding the target location information Scott, Georgia Tech SPIE, April 2009
Dictionary Generation: EMI Model Target is assumed to be infinitesimal and described by the magnetic polarizability tensor M Magnetic fields calculated using the Bio-Savart law Using reciprocity that the received voltage due to the target is Scott, Georgia Tech SPIE, April 2009
Dictionary Generation Discritizing the target space and calculating the model data for each possible target location for any scan position of the EMI sensor generates a dictionary of responses However generating the dictionary for 3D target localization increases the size of the dictionary making it unrealistic to work with it For 1m3 volume with 1 cm resolution the dictionary would have 1 million columns !!! Scott, Georgia Tech SPIE, April 2009
Use Translational Symmetry As the array scans targets having shifted y locations but the same cross range and depth values ,i.e., same x and z locations, result in shifted responses of each other in the receiver coils. It is sufficient to create a dictionary for y = 0 cross range slice and use it to locate targets in 3D. Scott, Georgia Tech SPIE, April 2009
Orthogonal Matching Pursuit Scott, Georgia Tech SPIE, April 2009
Outline Introduction Method Examples Conclusions Basis for Location Estimation Dictionary Orthogonal Matching Pursuit (OMP) Examples Simulated Laboratory Field Conclusions Scott, Georgia Tech SPIE, April 2009
Simulated Scanning Example #1 A running scan is simulated Horizontal line denotes the center line of the EMI array Although target position estimates are done for the current measurements, (shown by green stars), they are not accepted (shown by red) until they fall in the acception region. Acception region is taken as 25 cm behind of the center line of array. The model effects of accepted targets are taken out of measurements for detecting future targets. Scott, Georgia Tech SPIE, April 2009
3D Scanning Examples - 2 Scott, Georgia Tech SPIE, April 2009
Experimental Laboratory Results For the laboratory results, the EMI head is stationary and the target is scanned Lessons the effects of nearby metal objects Single Loop Z Yaw Three Loop Pitch Y X Five Degree-of-Freedom Non-metallic Positioning System Broadband EMI Sensor Scott, Georgia Tech SPIE, April 2009
Example: Single Loop 22 AWG wire with a diameter of 31.8 cm For the laboratory results, the EMI head is stationary and the target is scanned x=17 cm, yaw=0, pitch=0, z = −8 cm to z = −20 cm with 1 cm increments Scan y = −50 to y = 50 cm with 0.5 cm increments Accurate Location estimated for all depths A loop target model oriented in z direction is used for dictionary generation Scott, Georgia Tech SPIE, April 2009
Example: Single Loop 22 AWG wire with a diameter of 31.8 cm For the laboratory results, the EMI head is stationary and the target is scanned x=0 and 32 cm, yaw=0, pitch=0, z = −8 cm to z = −20 cm with 1 cm increments Scan y = −50 to y = 50 cm with 0.5 cm increments X = 2 X = 32 Scott, Georgia Tech SPIE, April 2009
Example: AP Landmine Medium metal AP Landmine For the laboratory results, the EMI head is stationary and the target is scanned x=17 cm, yaw=0, pitch=0, z = −8 cm to z = −20 cm with 1 cm increments Scan y = −50 to y = 50 cm with 0.5 cm increments Accurate Location estimated until ̴ 15 cm depth A loop target model oriented in z direction is used for dictionary generation Scott, Georgia Tech SPIE, April 2009
Example: AP Landmine Medium metal AP Landmine For the laboratory results, the EMI head is stationary and the target is scanned x=7 and 28cm, yaw=0, pitch=0, z = −8 cm to z = −20 cm with 1 cm increments Scan y = −50 to y = 50 cm with 0.5 cm increments X = 7 X = 28 Scott, Georgia Tech SPIE, April 2009
Example: Triple Loop Three loops of copper wire that are at right angles to each other: zr = 10.1, 50.2, and 172 kHz. For the laboratory results, the EMI head is stationary and the target is scanned x=17 cm, yaw=0, pitch=0, z = −8 cm to z = −20 cm with 1 cm increments Scan y = −50 to y = 50 cm with 0.5 cm increments Poor Location estimated for all depths A loop target model oriented in z direction is used for dictionary generation Scott, Georgia Tech SPIE, April 2009
Field Data The broadband array EMI system is scanned over the grid squares Battery Power EMI Array Head Broadband EMI System at Field Site Scott, Georgia Tech SPIE, April 2009
Field Data Results Measurements taken over an AP mine with the quadrapole array in the field. Depth estimates for 6 different instances of an AP mine in the field Scott, Georgia Tech SPIE, April 2009
Conclusions and Summary A promising location estimation method using electromagnetic induction array data is proposed. Simulated, Laboratory and Field Data Results indicate that depth and cross range estimation of targets could be obtained within an acceptable accuracy Future work requires modeling the target for generalization of the method Scott, Georgia Tech SPIE, April 2009
Questions Scott, Georgia Tech SPIE, April 2009