Sequential Adaptive Multi-Modality Target
This presentation is the property of its rightful owner.
Sponsored Links
1 / 27

Sequential Adaptive Multi-Modality Target Detection and Classification Using Physics-Based Models PowerPoint PPT Presentation


  • 53 Views
  • Uploaded on
  • Presentation posted in: General

Sequential Adaptive Multi-Modality Target Detection and Classification Using Physics-Based Models. K. Sarabandi I. Koh, B. Lyons, H. Mosallaei, M. Casciato. Radiation Laboratory The University of Michigan, Ann Arbor, MI 48109-2122 [email protected] Outline: Motivation

Download Presentation

Sequential Adaptive Multi-Modality Target Detection and Classification Using Physics-Based Models

An Image/Link below is provided (as is) to download presentation

Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author.While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server.


- - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - -

Presentation Transcript


Powerpoint presentation

Sequential Adaptive Multi-Modality Target

Detection and Classification Using Physics-Based Models

K. Sarabandi

I. Koh, B. Lyons, H. Mosallaei, M. Casciato

Radiation Laboratory

The University of Michigan, Ann Arbor, MI 48109-2122

[email protected]


Powerpoint presentation

  • Outline:

  • Motivation

    • Detection of targets camouflaged under foliage using multi-frequency, -polarization, -incidence angle SAR/INSAR sensors.

  • Physics-based scattering and propagation modeling of clutter

  • Model reduction (extraction of channel parameters)

  • Scattering models for hard targets under trees

  • High resolution SAR/INSAR image simulator

  • 3-D SAR at MMW for target detection and identification


Powerpoint presentation

  • Motivation

  • A reliable approach for detection and identification of targets camouflaged under foliage with an acceptable false alarm rate and probability of detection has not yet been developed.

  • Due to the complexity of the problem, i.e.

    • Signal attenuation, phase-front distortion, poor signal-to-clutter ratio, etc., single sensor approaches (optical, IR, radar) do not produce satisfactory results.

  • “Capable sensors” operating in diverse modality in conjunction with novel algorithms can drastically enhance FAR and PD.

    • Polarization diversity, Multi-frequency, Multi-static, Multi-incidence angle, Interferometric


Powerpoint presentation

Phenomenology of Wave Scattering & Propagation In Forest

Satellite

UAV

receiver

  • Forest is a complex random medium composed of lossy scatterers arranged a semi-deterministic

    • Foliage cause significant attenuation, scattering, field fluctuation

  • To assess performance of radar sensors and target detection algorithms phenomenology of EM wave interaction with foliage must be understood.

    • Scattering from foliage (clutter)

  • Target is in the close proximity of many scatterers

    • Distortion of phase front and the scattered field from target

  • Signal level, fluctuations, polarization state,

    impulse response, spatial coherence etc. depend on:

    • Tree density

    • Tree type

    • Tree height and structure


Powerpoint presentation

3D Tree Generation

  • Lindenmayer systems allow generation of complex tree structure using only a few parameters

  • An algorithm based on self-similarity

  • Gross structure: columnar, excurrent, decurren

  • Biophysical parameters include tree height, trunk diameter(dbh) and branching angle

Tree structure G = G(V,,P)

Axiom  = X

Productions:

p1: X FF{-X}F{++X}F{+X}{-X}

p2: F FF


Powerpoint presentation

rb

ra

r0

  • Tree Type: Coniferous and Deciduous

  • Inclusion of Botanical Information

  • –Tapering in Length and Diameter

  • *Law of conservation of cross section area

  • –Stochastic Processing

  • –Leaf Arrangement

  • •Computer Implementation

  • –Tree DNA generation and structure visualization

  • Forest stand generation and visualization

  • (scaling and view angle)

r02 = ra2 + rb2

Red Maple

Red Pine


Powerpoint presentation

Put Matts stuff here

Fractal tree details

GUI

Still scenario

Movie

Tree generation graphical user interface

The Radiation Laboratory


Powerpoint presentation

Scene Generation and Visualization

  • Land cover

  • DEM

  • Tree stand placement

  • Vehicles, transmitter, and receiver placement


Powerpoint presentation

Propagation & Scattering Model for Forest Canopies

• Scattering from discrete scatterers- Trunk: stratified dielectric cylinder- Branch: homogeneous dielectric cylinder- Leaf: dielectric disk or needle- Ground: layered dielectric half-space

•Single Scattering is invoked

•Four Scattering Mechanisms are included

Height

Attenuation rate NP/m

Rough Interface


Source or observation point in the forest

Source or Observation Point in the Forest

  • Near-field calculation is required

  • Approximate analytical formulations for near-field scattering from branches and tree trunks are derive.

  • Coherent summation of scattered field from all tree components. (Coherence is important at S-band and lower)

E i

E s

E d

E r

Single scattering theory Interaction among tree structures are ignored.


Backscattering coefficient of red pine forest

45

Backscattering Coefficient of Red Pine Forest

150 trees & 100 realization.

Density: 0.1/m2

Red pine

Tree height: 15.3 m

Crown Height: 9.5 m


Time domain response at a fdtd grid point

attenuation

dispersion

attenuation

Time-Domain Response at a FDTD Grid Point

Frequency: 30MHz – 100MHz

10 trees are considered.

Dielectric constants: 21.7 + i14.6 for branch

9.8 + i1.7 for ground.

Height of tree: 15m, Diameter of trunk: 22cm.

45o Incidence angle.

Note: Small effect of forest


Time domain response

Ez

Ez

V/m

V/m

Severe distortion due to trees

& Dispersion

Time[ns]

Time[ns]

Time-domain Response

Observation point is 1m above the ground inside a pine forest.

v-pol. wave is incident at 40o, and BW 1GHz (1GHz – 2GHz).


Interaction of foliage and target hybrid frequency time domain simulations

h-pol. or v-pol.

including all interactions

Interaction of Foliage and Target Hybrid Frequency/Time-Domain Simulations

  • Using the forest model, calculate time domain response of several trees in the proximity of the target at FDTD grids on a box (excitation).

2.Using FDTD, compute scattering from the target on the same grid points.


Powerpoint presentation

Hybrid Frequency/Time-Domain Simulations (Cont.)

  • To calculate the interaction between the target and the forest, reciprocity theorem is used. After exchanging observation & source points, use the previously calculated scattering property of the forest to obtain the final backscattering result.

Observation point

exchanging

Source point

Note: Using this procedure, interaction between forest & target is taken account into up to first order.


Powerpoint presentation

Validation of Hybrid Frequency/Time-Domain Modeling

(x)

(z)

Validation

A 2x2x2 FDTD mesh is used to model free space within the forest (in the absence of any vehicles).

The same problem is solved by a pure MoM code.

Results of the two methods are in excellent agreement.

A FDTD box around the observation point

(y)

2x2x2 FDTD mesh and electric field component that is plotted in the figure on the right.

FDTD simulation parameters :

Dx = Dy = Dz = 0.3 m

Dt = 0.314 nsec


Powerpoint presentation

Bistatic Scattering From HUMVEE

Discretized HUMVEE for FDTD Analysis

z

Ei

x

400

y


Powerpoint presentation

Preliminary Results

Current Distribution over the HUMVEE


Field distribution on the fdtd box 2 ghz

Field Distribution On the FDTD Box(2 GHz)

h-pol. incidence

v-pol. incidence

Note : Considerable distortion due to trees.


Powerpoint presentation

Macromodeling of Field Statistics – Mean Field

  • Mean Field at receiver is of the form:

  • Functional to macromodel mean field should be of similar form

    such as Prony’s exponential series expansion:

  • According to Foldy’s approximation a < S > in forward direction, remembering:


Pine forest mean field

Pine Forest - Mean Field

i = 45o

<EThy>

Simulated Data

Macromodel

Prony’s order = 3

<ETvx>

<ETvz>


Powerpoint presentation

Model reduction: Field STD

  • Standard deviation is a smooth function of f

  • It is therefore possible to macromodel the standard deviation

    with a functional in the form of a Taylor series polynomial:

std(Etvx)

std(Ethy)

std(Etvz)


Spatial correlation

a = 1.07

a = 1.58

v-pol.

h-pol.

1

1

Calculated

1/(1+ ax)

Calculated

1/(1+ ax)

0.8

0.8

0.6

0.6

0.4

0.4

0.2

0.2

0

-5

0

5

0

l

-5

0

5

l

Model Reduction: Spatial Correlation

Spatial Correlation

Note: Since observation line is inside the shadow region, field should be highly correlated.

Observation line


Powerpoint presentation

Michigan SAR Image Simulator Geometry

HH Polarization

VV Polarization

Direction of Flight

Increasing Range

0dBsm Point Target

0dBsm Point Target


Future works

Future Works

  • Use physics-based model for generating synthetic multi-modal data

    • Statistics of clutter scattering and channel (Monte Carlo simulations)

    • Hard target interaction with foliage (model reduction)

  • Improvement of Forest Model Accuracy

    • Including the effects of near-field multiple scattering among vegetation components.

  • Hard target model reduction (scattering centers)

  • Implement hybrid foliage/hard target interaction.

  • Improve computation time: Parallel processing


Powerpoint presentation

Shuttle Radar Topography Mission

50 Km X-band

225 Km C-band Swath Width


  • Login