Physical Realization of an Optical Automatic Target Recognition System

(1)

[ title ]

(2)

[ abstract ]

(3)

Acknowledgments

This Master Thesis was carried out at Lappeenranta University of Technology.

I would like to thank my supervisor Erik Vartiainen for his consultations and advices on this work. I am also very grateful to Erkki Lahderanta for the wonderful possibility to study here in LUT.

If you are reading this, you possibly deserve the gratitude too, as you are at least interested in the topic of this work.

Lappeenranta, 2010 Vladimir Osipov

(4)

List of Figures

1.1 Proposed algorithm for reected light processing . . . 13

2.1 Absorption of pure water . . . 16

2.2 Modern laser diode structure . . . 20

2.3 Cotton reectance . . . 22

2.4 Nylon reectance . . . 23

2.5 Nylon webbing reectance . . . 23

2.6 Polyester reectance . . . 24

2.7 Interference lter operation . . . 27

2.8 Interference lter transmission . . . 28

3.1 Experimental setup for reection spectra measurements . . . 30

3.2 Spectrum of a typical incandescent lamp . . . 31

4.1 Measured incandescent lamp light spectrum . . . 38

4.2 Measured laser light spectra . . . 39

4.3 Normalized reection spectra of monochromic samples . . . 40

4.4 Normalized reection spectra of patterned samples . . . 41

(7)

List of Tables

1 Atmosphere composition near the surface . . . 15

2 Fractions of materials used in gear . . . 21

3 Probing wavelengths . . . 32

4 Samples to be measured . . . 33

5 Filters peak transmission . . . 34

6 Characteristic numbers for 10% reector . . . 39

7 Characteristic numbers for monochromic samples . . . 41

8 Characteristic numbers for patterned samples . . . 41

9 Monochromic samples in the articial environment . . . 42

10 Patterned samples in the articial environment . . . 42

11 Absorption in the vaporous gap . . . 43

12 Monochromic samples in the articial environment w/o vapour . . . 43

13 Patterned samples in the articial environment w/o vapour . . . 43

14 Correlation between samples . . . 44

15 Equivalence hypotheses . . . 44

(8)

Nomenclature

Symbols

A atmosphere losses

C concentration

Hr relative humidity

I intensity

M molar mass

N density of absorbers N^∗ eective refractive index

N e external medium refractive index

R reection coecient, distance from scatter R₀ total reectance

Rscat scattered reectance

R_q surface roughness parameter

T transmission coecient, temperature

V volume

kB Boltzman constant l optical path, distance

m mass

p_A vapour pressure of A

x some part

α absorption coecient, molecule polarizability θ angle of incidence

λ wavelength

σ absorption cross-section

(9)

Abbreviations

ATR Automatic target recognition

CP Cotton-polyester

FIR Far infrared

FLIR Forward looking infrared FWHM Full-Width at Half-Maximum GPIB General Purpose Interface Bus

ID Identier

IFF Identication friend-or-foe

IR Infrared

HV High voltage

LED Light-Emitting Diode

NIR Near infrared

NYCO Nylon-cotton

OD Olive drab

PMT Photomultiplier tube

UV Ultraviolet

eV electron-Volt (1.6·10⁻¹⁹ J) ppmv parts per million of volume

w/o without

(10)

1 INTRODUCTION

Throughout the history of the humankind there has been a demand for a supe- riority in the warfare. It were mostly the inventions that were initially intended for military, that later got widely applied and carried on the progress. Moreover, since the ancient times nations have been spending a huge part of their resources for military purposes. This is still actual today.

In the modern warfare actively develops a new trend connected with a robotics warfare. A considerable amount of researches is done in this area. One of the critical elements of robotics warfare systems is an automatic target recognition (ATR). It is a device, or primarily an algorithm, allowing to recognize objects based on the data received from sensors. Without this part it is impossible for an unmanned vehicle to perform combat tasks, involving interaction with enemy units, without an assistance from a human operator. The primitive ATR system is just an identication friend or foe (IFF) transceiver. It utilizes radio frequency encrypted data exchange between the operator and the target. Such systems have a number of disadvantages for application in ATR systems and are mainly used for friend or foe recognition of aircraft or vehicles. Its main disadvantage is a necessity to have a transmitter on-board for every allied unit and even in such a case neutral units will be always considered non-friendly. This problem is solved by applying passive methods of data acquisition from an object. The best way for such acquisition is to utilize light, especially invisible. It is hardly detectable by detached observers, provides high localization and allows very fast scanning and thus determination of an essence of an object.

The main aim of this work is to search for a possibility of a physical realization of an optical automatic target recognition system. It is proposed to investigate theoretically and experimentally the possibility of a distant recognition of various modern infantryman gear materials and, moreover, various colouring patterns of those materials. For this purpose the following algorithm is proposed: to highlight

(11)

Figure 1.1: Proposed algorithm for reected light processing.

the object with light of an infrared (probably near infrared) laser, operating in a multi-mode regime, lter the reected beam on necessary wavelengths, mathe- matically post-process the obtained data and look, if it is possible to distinguish clearly one sample from another. Fig. 1.1 shows the proposed processing of the reected light. To implement this idea it is necessary to solve a number of auxiliary problems, that is going to be done in this work.

(12)

2 ESSENTIAL THEORY

2.1 Reection and scattering from rough surfaces

The reection from uneven or granular surface diers from the specular reection, i.e. reection from smooth and plain surfaces. The incident ray in the case of rough surface reects at a great number of angles. In case of textile surface particles are usually much larger in diameter than the wavelength, and we deal with geometric scattering, which is neither Rayleigh nor Mie, and the laws of geometric optics are mostly sucient here to describe the interaction of light with the material particles.

The fraction of light, scattered from any custom rough surface can be found as [1]

Rscat

R₀ = R0−Rspec

R₀ = 1−exp²(−4πRq

λ ), (2.1)

whereR_qis a surface roughness parameter,R_scatis scattered reectance andR₀is a total reectance. In case, when we deal with textile materials, roughness parameter is usually much more, than the wavelength of, for example, a NIR light. Thus it is seen that for Rq > λ we have the fact that most of the light is scattered, neither reected specularly.

2.2 Atmosphere absorption and scattering

2.2.1 The phenomenon of atmosphere absorption

There are two reasons for absorption processes in the atmosphere the absorption by gaseous molecules of an atmosphere itself and the absorption by various aerosols.

An aerosol is a suspension of ne solid particles or liquid droplets in a gas. There are not too many major gaseous components. Table 1 shows us the composition of the Earth atmosphere near the surface. This composition is almost always constant

(13)

with the exception of water part. The amount of water is dened by humidity and may vary from 0.5% to 4% [2], depending on the atmosphere conditions. At the same time the aerosol composition is never the same. Sand and dust, suspended in the air, may create serious diculties for the light beam to propagate over long distances.

Table 1: Earth atmosphere near the surface. [2]

Gas Volume, ppmv

N2 780840

O2 209460

H2O ≈ 20000

Ar 9340

CO2 387

Ne 18

Molecular absorption turns the radiation energy into the excitation energy of the molecules, the incident beam looses its energy. The energy of gas molecules is normally dened by the following mechanisms:

• translational motion of the molecule mass center, where the average kinetic energy in this case is equal to ^k^B₂^T, where T is the gas temperature;

• rotation of the molecule about an axis through its center of mass;

• vibration of the molecule atoms about their equilibrium positions with stretch- ing of chemical bonds between them;

• energy states changing of the molecule atoms electrons.

As we are going to use NIR region that lays in 0.75 - 1.4 µm range [3], we are mainly interested in rotational and vibrational transitions of the molecules, as the

(14)

translational energy is too low (25.85meV at300 K) and electron energy is usually higher (> 1 eV). The atmosphere components, primarily responsible for absorption in this range, are water (H2O) and carbon dioxide (CO2) [4]. NIR absorption bands for water are 0.94, 1.1 and 1.38 µm and for carbon dioxide 1.4 µm. The carbon dioxide absorption at the concerned band is comparatively weak [5]. More- over, as seen from Table 1, the concentration of carbon dioxide is normally 50 times lower, than the concentration of water vapour. So, for the rst approximation, carbon dioxide may not be taken into the consideration, if assumed that the air is clear.

Figure 2.1: Absorption of pure water [6].

To describe the absorption process Beer-Lambert (or BeerLambertBouguer) law is used. It introduces an absorption coecient that is a property of a medium and links it to transmission. For gases it looks like

T = I

I₀ =e^−αl =e^−σlN, (2.2)

where α is the absorption coecient, l is an optical path of a beam inside the medium, σ is an absorption cross-section and N is a density of absorbers. Beer- Lambert law is also capable to describe scattering, but this will be shown later.

(15)

2.2.2 The phenomenon of atmosphere scattering

Scattering in the atmosphere also happens due to two separate mechanisms scattering by the gaseous components and scattering by the aerosols. Although the light energy is not lost after scattering, the eective power goes down, as the light is being redistributed into dierent directions. The scattering by gaseous molecules happens in accordance to Rayleigh law. It describes scattering by particles much less in size, than the wavelength [7]

I =I₀8π⁴α²

λ⁴R² (1 + cos²θ), (2.3) where α is a molecule polarizability,R is a distance from scatter andθ is an angle between incident and scattered rays. The main atmosphere scatterer, nitrogen (N₂) has a cross-section of5.1·10⁻³¹ m² at 532 nm [8]. This gives the transmission loss of about10⁻⁵ per every meter. As seen from equation (2.3), scattering intensity in the considered case is reciprocal to the fourth power of a wavelength, thus nitrogen scattering for 1 µm wavelength will be approximately eight times less. This fact makes possible to neglect the inuence of scattering by gaseous atmosphere components for NIR radiation.

For the aerosol scatterers, whose particles have typical sizes about the light wavelength, the scattering behaviour is described by the Mie (or Lorenz-Mie-Debye) theory. This theory is too complex to introduce it here, so one may refer to [9]

for the detailed explanation. Just note that the scattering is proportional to the scatterer size and reciprocal to the incident light wavelength. It is usually hard enough to calculate scattering from the particles with a complex surface, so it is proposed to use numerical description of the aerosol scattering process. As noted above, it is possible to take scattering inuence into account by the Beer-Lambert law by describing scattering as a usual loss by means of the absorption coecient.

It is now clear that for the NIR radiation in case of normal air, not polluted by

(16)

complex chemicals, but only usual dust, the reasons for energy losses are: absorption by water vapour, absorption by aerosols and scattering by aerosols.

Let's consider the dependence of an absorption coecient of an air lled with water vapour as a function of temperatureT and relative humidity H_r

H_r = p_H₂_O p^∗_H

2O

·100%, (2.4)

where p_H₂_O is a water vapour pressure and p^∗_H₂_O is a saturation water vapour pressure. Saturation water vapour pressure for temperature T is [10]

p= exp(20.386−5132/T). (2.5) 1 Torr is equal to 133.322 Pa. From ideal gas law

pV =mRT /M, (2.6)

wherep,V, T, mand M are correspondingly pressure, volume, temperature, mass and molar mass of a gas and R is a gas constant, equal to 8.31 _{K mol}^J and from equations (2.4) and (2.5)

x= 133.322·18

8.31·100%·10⁶ · Hr

T ·e(20.386−5132/T)

, (2.7)

where x is a part of water in a cubic meter of medium. Assuming absorption to be linearly dependent on a fraction of an absorber we have α^x_H

20 = α_H₂₀·x. This relation is quite useful, as atmosphere parameters are always given as a temperature and a relative humidity, but not as a water vapour mass or else.

So, the transmission coecient of such semi-ideal medium will be −(α^x_H₂₀+α_a,a+ α_a,s +α_i), where α^x_H

20 represents water absorption in a functional form, α_a,a is an aerosol absorption, α_a,s is an aerosol scattering, α_i is an intrinsic losses due to all other factors (necessary in case of deeper modelling). Some authors [11]

also propose to introduce optical depth parameter equal to R

αdl to describe the propagation in the non-homogeneous atmosphere, but in our case it is not necessary, as the gas and aerosol distribution is considered as uniform. We will also omit

(17)

the airmass factor, because of the same considerations. Finally, the transmission fractionT rfor the wavelength λ, aerosol composition and concentrationC, relative humidity H_r, temperature T and light path inside the medium l will be

T r(λ, l, H_r, T, C) = exp[−l·(α^x_H

20(λ, H_r, T) +α_a,a(λ, C) +α_a,s(λ, C))]. (2.8)

2.3 Laser beam spectrum and geometry

Laser radiation wavelength depends on properties of a medium used for lasing. To reduce the amount of a theory, let us consider only lasers that will be probably applied in case of the development of a real system, i.e. solid-state semiconductor lasers. Although we will consider only one type of laser, it is supposed not to be restricted to use other types of lasers for the measurement purposes during the planned experiment. The reason why it is the best to use lasers is because of their ability to produce almost all of the output power on the several and known wavelengths. Due to this, one can investigate precisely necessary monochromatic points or regions without having powerful light sources.

Semiconductor lasers are based on semiconductor LEDs, but utilize the same stimu- lated emission eect as all other lasers do. Semiconductor lasers can be fabricated for almost any wavelength in a reasonable range by varying the composition of semiconductors used [12]. Anyway, NIR belongs already for a long time to this range.

Due to the fact that most semiconductor lasers, except of vertical-cavity surface- emitting lasers (VCSEL), have emitting area longer in the lateral direction, the emitted beam diverges more in a vertical direction after leaving the laser [13].

Thus the beam has an elliptical prole. To make the prole circular one needs to use optics like cylindrical lenses. Fig. 2.2 shows a structure of a typical modern semiconductor laser with an emitting area marked red.

(18)

Figure 2.2: Modern laser diode structure.

2.4 Gear fabrics

2.4.1 Modern military gear materials

All the materials, used for production of military gear and apparel can be divided onto the groups of apparel materials, highly durable materials and protective materials.

All of these three types of materials are present in the equipment of a modern soldier. Apparel materials are represented by classical cotton, polyester (polyethy- lene terephthalate (C10H8O4)n [14]), polar eece (a sort of polyester [15]) and nylon (polymerized mixture of hexamethylene diamine C6H16N2 and adipic acid C6H10O4 [16]). Durable fabrics are mostly done from polyamides (a group of carbon-nitrogen based polymers), like Cordura nylon. Sometimes leather may be utilized in small amounts. Bulletproof protective materials include various aramid (a sort of polyamide [17]) synthetic bers and carbon composites. Among the most

(19)

well-known aramides are Nomex and Kevlar. Weatherproof protective materials like GoreTex are usually done from multilayered fabrics, in which the top layer is a polyamide one and the active, second one is polytetrauoroethylene (CnF2n+2) [18].

Likely for the current research, protective materials rarely appear on the very surface of the equipment and are always covered by durable wear-proof cover, usually made from Cordura, thus it is not needed to consider their optical behaviour.

Let us estimate the fraction of all these materials in the wearing of an modern infantryman. Military apparel is done from either 65% of cotton and 35% of polyester or 50% of cotton and 50% of polyester or 50% of cotton and 50% of nylon [19].

Let us assume also that the gear and the armour take together about 60% of body surface. Let us assume that all the gear and armour covers are done from nylon. In the equipment there also normally exist gloves, made from Nomex or/and leather, but the area of gloves is negligible in comparison to the area of the whole body.

Let us also neglect the area of the uncovered part of the head.

Not to get deep into the details, we are interested in the optical behaviour of the following materials: cotton, polyester and nylon. Table 2 shows the approximate fraction for each of them for the case of a plain textile and the case of an equipped unit.

Table 2: Assumed fractions of materials on the gear "surface".

Material Fraction for a cloth, % Fraction for an equipped unit, %

cotton 50 - 65 20 - 26

polyester 0 - 50 0 - 20

nylon 0 - 50 60 - 80

(20)

2.4.2 Gear materials optical properties

Fig. 2.3 shows us the reection spectrum for the cotton fabric, coloured with white colour. The composition of the dye is unknown, so let us consider its own reection spectrum to be uniform in the NIR region. It can be seen that the reection of a cotton is almost the same throughout the whole NIR region and is approximately equal to 63%. Nevertheless it varies a bit. Below this variation may be taken into account while measurement data processing.

Figure 2.3: Reectance of the cotton, coloured white [20].

Fig. 2.4 shows us the reection spectrum for the nylon fabric, coloured with black colour. The composition of the dye is again unknown, so let us consider its own reection spectrum to be uniform in the NIR region. Reection varies from 35% to 64%. Below this data will be digitized and taken into account while measurement data processing. We may also need data for OD coloured nylon webbing, as it is found often in the equipment. Fig. 2.5 shows us its reectance. This data is not too useful, as is "spoiled" by the OD dye spectrum.

Fig. 2.6 shows us the reection spectrum for the polyester fabric, coloured with

(21)

Figure 2.4: Reectance of the nylon, coloured black [20].

Figure 2.5: Reectance of the thick nylon webbing, coloured OD [20].

black colour. The composition of the dye is again unknown, so let us consider its own reection spectrum to be uniform in the NIR region. The reection stays more or less similar near the level of 70%. Below this data will be also digitized and taken into account while measurement data processing.

(22)

Figure 2.6: Reectance of the polyester, coloured black [20].

2.5 Patterns and fabric dyes

To narrow the scope of the research let us consider only camouages used nowadays by the armies (even only ground forces) that do participate more than others in the ISAF¹ mission [21]. This way, it is proposed to investigate desert camouages of three states: USA, UK and Germany. We will not take into account non-ocial camouages, partly used by these armies, like Multicam or Digital DPM.

US ground forces (U.S.Army) are using ACUPAT (Army Combat Uniform PAT- tern) pattern for their equipment since 2004 for both homeland and expeditionary forces. On contrary to USMC (Marine Corps) there are no separate patterns for woodland, arid and urban areas. Pattern consists of three colours: Desert Sand 500, Urban Gray 501, and Foliage Green 502 [19]. Uniform is made of 50% of cotton and 50% of nylon. Gear is manufactured from Cordura nylon. It is prohibited to wear the eld jacket with the rolled sleeves [22], so no necessity to calculate arms area separately. From the assumptions, made in the part 2.4.1 and information from this paragraph we have about 80% of nylon and 20% of cotton for the surface

1International Security Assistance Force

(23)

of a unit.

UK ground forces (British Army) have got issued uniforms and gear of two types of patterns: DPM (Disruptive Pattern Material) is used for the troops, deployed in woodland areas, and DDPM (Desert Disruptive Pattern Material) is designed mainly for arid regions usage. We are interested in the second one, as are consid- ering the ISAF case. Pattern consists of two colours: Desert Tan 492 and Coyote Brown 476². Combat uniform is made of 50% of cotton and 50% of nylon.

Germany ground forces (Heer) have got also issued uniforms and gear of two types of patterns Flecktarn for woodland areas deployed troops and Tropentarn (or Wustentarn) for arid areas deployed troops. We are again interested in the second one. Pattern consists of three colours: Khaki (Tan 380), Brown 383, Dark Green 483. Combat uniform is made of 50% of cotton and 50% of polyester. One note should be done here: German soldiers are not so armoured as UK and US, so we will minimize the area, covered by gear and armour down to 40%. Overall we have about 40% of nylon, 30% of cotton and 30% of polyester for the surface of a unit.

Industrial colouring of cotton is usually done with the reactive dyes like Procion, polyester is coloured with disperse dyes and nylon, being easyly colourable, is often processed by acid dyes like anthraquinone (9,10-dioxoanthraceneC14H8O2 [23]) based dyes. [24] Unfortunately, it was not found what exact dyes are used for cre- ating patterns described above, so we will have to assume their optical properties basing on other methods. Anyhow, some components of the dyes, used for military, can be found in [25].

2Unfortunately no proof link here information was got from an oral source.

(24)

2.6 Optical bandpass lters

Optical lters selectively transmit light with certain properties (usually with the certain wavelength). On this criterion lters are divided onto longpass, i.e. trans- mitting only longer wavelengths, shortpass, i.e., on contrary, attenuating longer wavelengths and bandpass lters. Bandpass lter allows the transmission only in the exact passband. These lters are done, either by combining the materials of longpass and shortpass lters or by more complex techniques, as shown below.

Optical ltering mechanism can be either absorptive or interferential. The operation of an absorptive lter is based on the classical absorption of light by matter.

Varying the chemical composition of an absorber used, one may vary optical properties of the lter. These lters have plastic substrate that makes them cheap and lightweight. But they do not possess as ne properties, as interference lters do.

Interference lter consists of a number of thin lm layers (see Fig. 2.7) and utilizes an eect of thin-lm interference. The more layers the lter has the greater se- lectivity can be achieved [26], but this makes the production of such lters more expensive. Interference lters can be fabricated for almost any passband.

Transmission characteristic of a typical interference lter is shown on Fig. 2.8.

Filter central wavelength can be tuned by changing the incident angle λθ =λ0

q

1−(N e/N^∗)²·sin²θ, (2.9) where λ_θ is the wavelength at angle of incidence; λ₀ is the wavelength at normal incidence; N eis the refractive index of external medium;N^∗ is the eective refractive index of the lter and θ is the angle of incidence [27]. It is seen that tuning can be done only in the direction of shorter wavelengths, thus, in case of standard lter usage, it is necessary to choose a lter with a central wavelength, higher or equal to the incident light wavelength.

(25)

ρ_a = (n₀−n₁)/(n₀+n₁); δ₁ = 2πn₁d₁/λ;

ρ_b = (n₁−n₂)/(n₁ +n₂); δ₂ = 2πn₂d₂/λ;

. . .

ρm = (nm−1−nm)/(nm−1+nm); δm = 2πnmdm/λ;

Figure 2.7: Interference lter operation [26].

Out-of-band transmission for the described lter is 1· 10⁻⁴ from X-ray to FIR region. Minimum peak transmission in the NIR is from 35% to 50%, refractive index is 2.05. Full-width at half-maximum value for these serially produced lters varies from 100 down to 1 nm [27].

(26)

Figure 2.8: Typical interference lter transmission [27].

(27)

3 THE EXPERIMENT

3.1 Experimental setup

The setup described below, was used to perform the measurements of reection spectra. The setup consists of the laser source, optical system, sample holder, monochromator, photomultiplier, voltage amplier, radiometry voltmeter and a computer with the data acquisition software. Fig. 3.1 shows all these components.

Monochromator MS257 is the main part of the setup. It provides a wavelength resolution of 0.1 nm in the 170 1600 nm range. Monochromator has a quadruple grating with 1200 lines/mm and a motor driven slit assembly on the input and each of the two outputs. There are one input and two output ports. The default conguration of the monochromator supposes the output beam to pass out via the axial port. It is required to replace side exit mirror to adopt monochromator for the use with another output port. Quadruple-grating turret drive, ip mirror drive, slit assembly drive and integrated shutter are controlled automatically from the computer via the GPIB (IEEE-488) interface [28, 29].

Laser source illuminates the sample, the reected light gets through the diaphragm and the chopper to the monochromator input slit. The diaphragm is needed to re- move the light, reected from the unnecessary parts of the sample. The chopper, making 25 rotations per second, is needed to calibrate dynamically the zero input level it periodically hides the slit from the incident light and at these moments light level is considered to be zero and during the next cycle this value is subtracted from the received one.

The light of the desired wavelength on the output of the monochromator gets to

(28)

Figure 3.1: Experimental setup for reection spectra measurements.

the photomultiplier tube that is supplied with a high-voltage source. It is possible to vary the applied voltage in a wide range, but it is not desired, as the PMT amplication dependency on the voltage is unknown. Signal, produced by PMT, is amplied and fed to the radiometry system Merlin that, being also controlled by a computer via GPIB interface, processes the resulting signal and sends it to the PC. The intermediate amplier has a variable amplication coecient from 10⁴ to 10⁹. It is useful, as the Merlin system has a maximum input value of 5 V and it may be necessary to decrease the signal by a known factor in case of large signals.

The data acquisition software TracQ controls all the measurement characteristics and collects all tha data from Merlin system, allowing its preliminary processing.

The resulting data is stored on the disk in a simple and convenient format.

3.2 Light sources

For the experimental measurements it was decided to use not only laser source, but also a light of an incandescent lamp for the following purposes. The spectrum of

(29)

an incandescent lamp light is wide and smooth in the visible and infrared regions and has its maximum in the NIR. It can be approximately described by a Planck's law for the black body radiation

I(ν, T) = 2hν³ c²

1

exp(_kT^hν)−1. (3.1)

Figure 3.2 shows an intensity distribution of energy through the wavelength for a typical lamp. That is why such a lamp could be used as a polychromatic source, although not so powerful on a single frequency as a laser. This information was not given above, in the theoretical part, as the author does not consider it to be important for an explanation of an operation and so on.

Figure 3.2: Spectrum of a typical incandescent lamp. [30]

As a laser source there was chosen a 10 mW He-Ne laser with the main wavelength of 632.8 nm. Harmonics peaks of 1.5x and 2x wavelengths: 949.2 nm and 1265.6 nm were utilized to get into the NIR region.

(30)

3.3 Working wavelengths

As the nal design of a device, using the proposed algorithm, is supposed to measure the incoming light spectrum not integrally, but on few exact wavelengths, we have now also to use only some separated frequencies. Let us, this way, choose the amount of points that would be enough to cover NIR region more or less evenly.

It is necessary to take into the consideration the form of materials spectra given in part 2.4.2 and dyes spectra, given in part 2.5. It is good to remember that the increasing of data points amount will lead, in case of a device, to the increasing of the channels used, thus increasing the cost and lowering the reliability of the whole system. We may and should utilize the He-Ne laser, thus already providing ourselves with two working wavelengths 949.2 and 1265.6 nm.

Table 3: Probing wavelengths.

Window center, nm Source 750 Incandescent lamp 850 Incandescent lamp

949.2 He-Ne laser

1050 Incandescent lamp 1150 Incandescent lamp

1265.6 He-Ne laser

1350 Incandescent lamp

Thus, we provide a coverage of a NIR region with near 100 nm steps and do not get into local H2O absorption peaks (Fig. 2.1). It is proposed to set a window width for lamp-source points to be 10 nm from the following considerations:

• lamp source has no such localized power as a laser, so need to widen the window;

(31)

• there are serially produced lters for such windows [27];

• materials and dyes have no sharp spectrum areas in NIR;

• this is just an experiment, not a production.

For laser-source points it is proposed to choose the window, equal to the width of the corresponding laser peak. The exact values will be seen later from the preliminary measurements. We are not intended to use laser sources for all the measurements, only for method testing, as we do not have lasers for all the wavelengths and at the same time already have another sucient light source.

3.4 Samples selection

As written in part 2.5, it is proposed to choose from a variety of patterns only the camouages patterns of the armies, participating in the ISAF mission, as in this case it is possible to obtain extra information on pattern usage in reality, if necessary. Moreover, it is these armies that are usually pioneers in the introducing of a robotic warfare. This way, we have three types of patterns for investigation.

Table 4: Samples to be measured.

No Material Pattern/Colour

1 C65 / P35 ACUPAT

2 C65 / P35 DDPM

3 C65 / P35 Tropentarn

4 C65 / P35 OD

5 Nylon OD

6 Polyester OD

To measure dierent patterns it is proposed to x the structure of a fabric not to deal with dierent absorption for dierent materials. Vice versa, for measuring materials it is proposed to x colour, although it is not completely possible, as dyes

(32)

for each material will dier, as written in part 2.5. Table 4 shows, what samples are going to be measured. For patterned samples separate measurement of each colour of a pattern will be done in addition to standard all-sample-area measurement.

It is also proposed for this time not to utilize laser for full-pattern measurements, as we anyway have not got lasers for all the necessary wavelengths and use laser measurements only for the verication of the method proposed.

3.5 Filter modelling

For the current experimental purposes there is no necessity to have real lters, as there will be no real receivers and computing electronics. So we can only make a mathematical estimation of a lter behaviour. It is possible to simulate a lter by means of an equation, derived from the considerations of a bandpass lter operation, but it would be much convenient just to have a scalable piecewise linear function, representing a lter. Such a function or a dataset could be in our case acquired by digitizing the spectrum of a typical interference lter, shown on Fig. 2.7.

From [27] can be seen the average peak transmission coecients, Tx, for the desired lters with FWMH of 10 nm. This is given in Table 5.

Table 5: Filters peak transmission.

λ, nm 750 850 950 1050 1150 1265 1350

Tx, % 55 55 55 50 40 40 35

(33)

3.6 Simulating humidity

For the purpose of simulating humid atmosphere a simple water evaporating setup can be used. As shown in part 2.2.2, the amount of water vapour at xed relative humidity grows exponentially with the temperature, so we have just to boil the water with an approximately constant temperature to have a constant and humid enough atmosphere at the gap between the light source and the sample. My 1 kW kettle can serve the science a bit. The diameter of vapour-enriched area in this case is 11 cm.

Overheat air from the boiler will also introduce some geometric scattering. It seems to be more a feature, than a bug, because in natural conditions there can be found eects similar to road mirages [31].

3.7 Signal processing

In this work we are also going to simulate the signal processing in a real system, including normalization and sample determination. For the purpose of normalization it is proposed to perform additional measurement with a sample, having constant reection coecient, e.g. R= 10%and divide each spectrum (consisting, as agreed, of seven measured values) on the spectrum of that reference one.

Light intensity equation for the optical part with xed environmental conditions will be

I_λ =I₀(λ)A(λ, l)R(λ)A(λ, l)T(λ) = I₀(λ)A²(λ, l)R(λ)T(λ), (3.2) where I_λ is an incoming integral light intensity for the λ-window, R is a target reection coecient,T is a lter transmission coecient,Ais an atmosphere losses, I₀ is a source integral light intensity for theλ-window. From (2.2) and taking into

(34)

consideration thatexp²(x) = exp(2x), we have

Iλ =I0(λ)A(λ,2l)R(λ)T(λ), (3.3) where l is a distance to the target. We may note also electronic part sensitivity and amplication S for each of the measurement channels

I_i =I_0,iA_i(2l)R_iT_iS_i. (3.4) HereI_λ is changed toI_i to emphasize that it is already a digital signal of a channel i, ready for computer processing. Reference spectrum from 10% reective sample will look like

I_i,ref = 0.1I_0,iA_i(2l)T_iS_i. (3.5) Dividing (3.4) on (3.5) will give the plain values of the reectance

R_i = 0.1I_i

I_i,ref. (3.6)

The value of reference reectance (0.1) may not be taken into account, as the results are going to be compared between each other, having already been normalized. The formulas, derived above, may be later used for deeper investigations, like atmosphere dust eects, power calculations and so on.

After acquisition, the data from each channel is packed into a set. To compare these datasets between each other it is proposed to use Pearson's correlation coecient

r_xy =

Px_iy_i−xny p(P

x²_i −nx²)(P

y_i²−ny²), (3.7) wheren equals to the number of channels, i.e. seven. Then it is proposed to prove or reject a hypothesis of equivalence of dataset X and Y for various condence levels. Usually in such experiments the condence equals to 95%, but we will also check the hypothesis with a 99% precision. Using the Student's criterion

t= r_xy√ n−2

p1−r_xy² = 2.236 r_xy

p1−r_xy² . (3.8)

The criticalt-value for n= 7 and 95% is equal to 2.571 and for 99% it is 4.032 [32].

All the mathematical calculations can be done using MATLAB software.

(35)

3.8 The algorithm overview

Here is the step-by-step algorithm for carrying out the experiment described, including data acquisition and postprocessing.

1. Measure laser spectrum and verify that the are required peaks.

2. Find out windows widths for laser peaks.

3. Measure intensity ³ I_ref from reference 10% reector.

4. Adjust amplication coecient and PMT voltage to make maximum value from previous measurement to be near 10 - 15% of the scale. It is preferred to maximize the PMT voltage and minimize the amplication coecient here to keep the possibility of a precise multiplication in case of weak signals.

5. Measure intensity I_pat from entire patterns.

6. Calculate normalized intensity for all the samples as I =I_pat/I_ref. Let then these be Database samples.

7. Simulate high temperature and humidity by means of water vapour, increase distance between the setup and the sample, change angle on small random values.

8. Repeat measurements for the main types of samples. Let then these be Target samples.

9. Calculate correlation coecients.

10. Prove or reject hypotheses of equivalence of similar and non-equivalence of dierent Database and Target samples for 95% and 99% probabilities.

3 Measuring intensity means here and below in this part measuring intensity of a reected light at predened points, discussed in part 3.3, with a chosen window width.

(36)

4 RESULTS AND DISCUSSION

4.1 Light source spectra

Firstly light spectrum from the incandescent lamp was measured, using a simple mirror instead of a sample (see Fig. 3.1). Mirror, having non-uniform reection spectrum, can bring in some inaccuracy into the measurements. This situation is hardly acceptable, as this data is used for normalization. To get rid of possible errors the same spectrum was measured with a standard reector having a reection coecient of 10% for the NIR range. As seen from Fig. 4.1 mirror gave reection almost equal to 100%, nevertheless there is some additional absorption near 900 nm wavelength.

Figure 4.1: Incandescent lamp light spectrum got from a mirror and a 10% reector.

Reference data for the laser was taken only with a 10% reector, as shown, that measurements, involving mirror, may be not correct. Fig. 4.2 shows spectra of 949.2 nm and 1265.6 nm He-Ne laser peaks.

(37)

Figure 4.2: Laser light spectra. Upper axis is for 949.2 nm.

Integral average for the 949.2 nm peak appeared to be 0.4967 V and the same parameter for the 1265.6 nm peak is 0.0719 V. This value was calculated as

I¯=

R I(λ)dλ

∆λ . (4.1)

where ∆λ is the measurement range. It is used to characterize the power level in the required gap integrally with only one number. Let us call it characteristic numberA. For incandescent lamp light these characteristic numbers are calculated in a similar way, but in accordance to the bandwidth of the lters used in a real system (see part 3.5). Table 6 shows this data.

Table 6: Characteristic numbers for 10% reector.

λ, nm 750 850 950 1050 1150 1250 1350 A, mV 1.692 1.246 2.343 0.753 0.318 0.413 0.300

Note that characteristic numbers for the laser and for the incandescent lamp are in- comparable to each other and used only for further normalizations of acquired data.

(38)

4.2 Reection intensity spectra

4.2.1 In the pure environment

Firstly, let us take a glance at pure, not patterned materials, coloured in olive drab colour. Fig. 4.3 shows us the comparison of normalized reection spectra of nylon, eece (polyester) and cotton-polyester fabrics.

Figure 4.3: Normalized reection spectra of monochromic samples.

The shapes are not important much now, as are not being a scope of our investigation. Let us calculate characteristic numbers for the samples. When calculating, it is better to use normalized data than non-normalized, as described in part 3.8, although the second ones can also be used for comparison in our case.

Table 7 shows characteristic numbers for monochromic nylon (N), eece (F) and cotton-polyester (CP) fabrics samples.

Now let us nd out the similar characteristic numbers for all the three patterned

(39)

Table 7: Characteristic numbers for monochromic samples.

λ, nm 750 850 950 1050 1150 1250 1350 Fleece 0.330 0.293 0.342 0.316 0.278 0.292 0.416 CP 0.198 0.330 0.613 0.586 0.506 0.532 0.572 Nylon 0.296 0.394 0.450 0.505 0.537 0.561 0.593 samples. Fig. 4.4 shows us the spectra of these samples.

Figure 4.4: Normalized reection spectra of patterned samples.

Table 8: Characteristic numbers for patterned samples.

λ, nm 750 850 950 1050 1150 1250 1350 ACUPAT 0.259 0.386 0.524 0.518 0.414 0.438 0.484

DDPM 0.200 0.300 0.433 0.542 0.520 0.469 0.496 Tropentarn 0.178 0.500 0.575 0.445 0.292 0.329 0.367

It is seen that acquired spectra has no peaks and that the slopes are not steep, except of probably that of 850 nm. So there is no necessity to measure the whole spectrum in further experiments and is enough just to track the data at the xed wavelengths.

(40)

4.2.2 In the articial environment

The similar measurements, performed for monochromic and patterned spectra, but at the hot and humid articial environment gave us the following data. Tables 9 and 10 shows us the results for monochromic and patterned samples accordingly.

Table 9: Monochromic samples in the articial environment.

λ, nm 750 850 950 1050 1150 1250 1350 Fleece 0.292 0.320 0.360 0.226 0.182 0.236 0.171 CP 0.164 0.261 0.599 0.495 0.426 0.370 0.187 Nylon 0.223 0.333 0.314 0.362 0.340 0.300 0.212 Table 10: Patterned samples in the articial environment.

λ, nm 750 850 950 1050 1150 1250 1350 ACUPAT 0.219 0.325 0.547 0.367 0.294 0.243 0.209

DDPM 0.207 0.358 0.415 0.508 0.403 0.358 0.196 Tropentarn 0.168 0.601 0.574 0.349 0.213 0.193 0.118

Before statistical processing let us take into account the absorption of the water vapour. For this purpose we have to divide the charateristic numbers of the samples onto the appropriate numbers of water absorption spectrum. The spectrum was found in part 2.2. Table 11 shows us the estimated absorption in the vaporous gap with given diameter.

Meanwhile, this estimation is very rough due to uncontrolled vapour temperature and unknown vapour spread distance. This estimation only helps us to clear the data more or less. Tables 12 and 13 shows us the results for monochromic and patterned samples accordingly.

Now let us calculate the correlation coecients between the data, acquired from the measurements in the pure environment, and the measurements in the articial humid environment. Let us also calculate cross-correlation coecients, i.e. corre-

(41)

Table 11: Absorption in the vaporous gap.

λ, nm exp(−α·l)

750 0.989

850 0.981

950 0.862

1050 0.939 1150 0.726 1250 0.675 1350 0.353

Table 12: Monochromic samples in the articial environment w/o vapour.

λ, nm 750 850 950 1050 1150 1250 1350 Fleece 0.296 0.326 0.418 0.241 0.250 0.349 0.484 CP 0.166 0.266 0.695 0.527 0.587 0.547 0.530 Nylon 0.226 0.340 0.364 0.386 0.468 0.445 0.600 Table 13: Patterned samples in the articial environment w/o vapour.

λ, nm 750 850 950 1050 1150 1250 1350 ACUPAT 0.222 0.331 0.594 0.451 0.405 0.359 0.503

DDPM 0.209 0.365 0.481 0.541 0.555 0.530 0.554 Tropentarn 0.169 0.613 0.666 0.371 0.293 0.286 0.334

lation between dierent materials. Both types of coecients are used in order to prove or to reject the equivalence hypotheses. Table 14 shows us these coecients.

Cross-correlation is high enough sometimes, as in the case of patterned materials we have similar spectra.

Using the data from part 3.7 it can be shown, whether the samples can be clearly distinguished one from another or not. Table 15 claries the situation.

(42)

Table 14: Correlation between samples.

Sample Correlation Cross-correlation

Fleece 0.789 0.314

CP 0.950 0.740

Nylon 0.931 0.432

ACUPAT 0.912 0.561

DDPM 0.979 0.338

Tropentarn 0.946 0.496

Table 15: Equivalence hypotheses.

Sample True at 95% True at 99% Can be missed? (99%)

Fleece yes no no

CP yes yes no

Nylon yes yes no

ACUPAT yes yes no

DDPM yes yes no

Tropentarn yes yes no

4.3 It happened that. . .

With all the assumptions done the samples can be distinguished one from each other with at least 95% probability. For cotton-polyester samples this probability can be increased up to 99%. One should also take into account that in the calculations above we have used water vapour absorption approximation, which could be not precise for the conditions used. Anyway, the quality of distinguishing can be increased either by increasing the number of channels, or by some intelligent selection of sensing wavelengths.

(43)

5 CONCLUSION

The objective of this work was to try to nd out the possibility of creation an optical target recognition system, using target aimed scanning at several xed wavelengths. It was desired to use near infrared light as the most convenient from all of the points of view. Materials and patterns were selected in a way to represent better some modern military gear and equipment.

The series of conducted experiments and the data that was obtained, showed us that the proposed algorithm can be used for the purposes declared. In other words one can take a sample, measure it, process, compare to all the references, stored in the database, and denitely say, which one was measured. Of course, in the case when the samples are very similar, this algorithm will not work. But in our case it was not required to distinguish similar targets as dierent ones.

Many factors had not been taken into consideration during the experiments. In a real life situation the atmosphere may dier a lot from those two that were used while testing. There may be dust. Unfortunately it was not possible to simulate dusty air conditions, as this would have required special equipment. There still re- mains a problem of proper calibration of such a system. Within the bounds of this work the simplest calibration method was proposed. It will be dicult to calibrate this way in non-laboratory conditions. No electronic part was considered, although it can bring in additional errors into the incoming data too.

The overall price of such a system, if we will take into consideration only optical and electronic components, can be even less than 1000 euros. Unfortunately, we can not compare this price, as the prices of the systems analogous to this are unknown.

(44)

This work may be helpful for those, who conducts investigations in the area of ATR or in the area of distant optical measurements. Also it may be tried to assemble the proposed system for optimisation and further research. Denitely, those ones will be in demand in future.

(45)

References

[1] J. M. Bennett, L. Mattsson, Introduction to surface roughness and scattering.

Optical Society of America, Washington D.C., 1989.

[2] E. Grayzeck, NASA Fact Sheets. Earth Fact Sheet, 2009.

[3] Roy Henderson, Wavelength Considerations. Instituts fur Umform- und Hochleistungs, 2003.

[4] S.C.Liew, Absorption of Radiation. Centre for Remote Imaging, Sensing and Processing, Singapore, 2001.

[5] S.C.Liew, IR Absorption Bands of Water and Carbon Dioxide. Centre for Remote Imaging, Sensing and Processing, Singapore, 2001.

[6] G. M. Hale, M. R. Querry, Optical constants of water in the 200 nm to 200 µm wavelength region, Applied Optics, vol. 12, pp. 555563, 1973.

[7] C.F. Bohren, D. Human, Absorption and scattering of light by small particles.

John Wiley, New York 1983.

[8] Maarten Sneep, Wim Ubachs, Direct measurement of the Rayleigh scattering cross section in various gases, Journal of Quantitative Spectroscopy and Radiative Transfer, vol. 92, pp. 293310, 2005.

[9] Raymond L. Lee, Jr., Mie theory, Airy theory, and the natural rainbow, Applied Optics, vol. 37, pp. 15061519, 1998.

[10] Holger Vomel, Saturation vapor pressure formulations. CIRES, University of Colorado, Boulder, November 2006.

[11] Houghton, John Theodore, The Physics Of Atmospheres. Cambridge Univer- sity Press, 2002.

[12] Mark Csele, Fundamentals of Light Sources and Laser. Wiley, 2004.

(46)

[13] Kenichi Iga, Y. Kokubun, Encyclopedic Handbook of Integrated Optics. CRC Press Taylor & Francis Group, LLC, 2006.

[14] N. Ogata, K. Sanui, H. Tanaka, S. Yasuda, Synthesis of polyester by direct polycondensation with triphenylphosphine, Polymer Journal, vol. 13, pp. 989911, 1981.

[15] J. Hu, B. Xin, H.J. Yan, Classifying eece fabric appearance by extended morphological fractal analysis, Textile Research Journal, vol. 72, pp. 879 884, 2002.

[16] M.P. Stevens, Polymer Chemistry. Oxford University Press, 1999.

[17] Committee on High-Performance Structural Fibers for Advanced Polymer Ma- trix Composites, High-Performance Structural Fibers for Advanced Polymer Matrix Composites. 2005.

[18] Robert W. Gore et al., U.S. Patent number 4194041. Waterproof laminate, March 1980.

[19] Program Executive Oce Soldier, U.S.Army, Army Combat Uniform, Oc- tober 2009.

[20] Clark et al., Digital Spectral Library, DS231. U.S. Geological Survey, 2007.

[21] United Nations Security Council, Resolution 1386 (2001), December, 20 2001.

[22] U.S. Army Publishing Directorate, Army Regulation 6701, 2005.

[23] H. W. Underwood, Jr. and W. L. Walsh., Quinone, Organic Syntheses, vol. 2, p. 553, 1943.

[24] W. R. Remington, E. K. Gladding, Equilibria in the Dyeing of Nylon with Acid Dyes, J. Am. Chem. Soc., vol. 72, p. 25532559, 1950.

[25] U.S. Army Natick Soldier Research, Development and Engineering Center, USIFI Mil Spec Conference, Natick, MA, February 2006.

(47)

[26] Hugh Angus Macleod, Thin-lm optical lters. Third Edition. Institute of Physics Publishing, Philadelphia, PA, USA, 2001.

[27] LOT-Oriel Group, Standard Bandpass Filters, http://www.lot- oriel.com/site/pages_uk_en/products/optical_lters/bandpass_spec_799.php (checked 2009.12.28).

[28] LOT-Oriel Group, Oriel MS257^TM Monochromator and Imaging Spectro- graph, http://www.newport.com/Oriel-MS257-1-4-m-Monochromator-and- Imaging-Spect/378367/1033/catalog.aspx (checked 2009.12.29).

[29] Andrey Tomilov, Electroluminiscence of silicon- and some metal oxides, Mas- ter's thesis, Lappeenranta University of Technology, 2009.

[30] William J. Sersen, Articial Light The Options Available, Gemological Digest, vol. 3, p. 4556, 1990.

[31] Andrew T. Young, An Introduction to Mirages. San Diego State University, 1999.

[32] Murray R. Spiegel, Larry J. Stephens, Statistics. Fourth edition.. Schaum's outlines, 2007.