Combined High-Resolution Imaging and Spectroscopy System

A Versatile and Multi-modal Metrology Platform

Tobias Hegemann, Fabian B

urger and Josef Pauli

Intelligent Systems Group, University of Duisburg-Essen, Bismarckstraße 90, 47057 Duisburg, Duisburg, Germany

Keywords:

Imaging, Spectroscopy, Surface Analysis, Metrology, Calibration.

Abstract:

Multi- and hyperspectral measurement methods are a versatile approach to analyzing and understanding ma-

terial properties. Especially imaging techniques allow for a precise sensing of surface properties. This paper

presents a novel and multi-purpose metrology platform for high-resolution spectral measurements. The sys-

tem is able to acquire multispectral images with six different spectral channels as well as hyperspectral point

measurements and images. This is realized with a combined measurement head that includes a gray value

camera as well as two spectrometers ranging from 190nm to 1,700nm. A three axis Cartesian robot with

nanometer-precision allows the analysis of large samples with a size up to 40 × 10 × 10 cm and a weight

of up to 25kg. Approaches to automatically focus both the camera and the spectrometers are presented. The

calibration process between the camera and the spectrometers, which is necessary to acquire the full spectral

information corresponding to each pixel of the camera image, is described. Example images and measure-

ments are discussed to show the potential of the metrology platform.

1 INTRODUCTION

The analysis of a wider range of the electromag-

netic spectrum offers the chance to obtain more in-

formation about materials and the quality of sam-

ples. To this effect a unique measurement system was

built, the ﬂexible and multi-modal metrology plat-

form CHRISS: Combined High-Resolution Imaging

and Spectroscopy System. CHRISS allows for multi-

and hyperspectral material analysis which offers more

information than color images and may be employed

in various application ﬁelds such as industrial inspec-

tion, defect detection, material classiﬁcation or food

quality, but also in medical contexts such as tumor

identiﬁcation (Panasyuk et al., 2007). By analyzing

materials with respect to their properties in the elec-

tromagnetic spectrum, it is possible to differentiate

between substances, learn about the composition of

materials in a sample or evaluate the heterogeneity

of a specimen. This provides challenges concerning

both hardware, as specialized equipment is needed,

and software. CHRISS is built to comply with the

hardware challenges, and software solutions were im-

plemented to enable autofocusing both camera and

spectrometer unit as well as providing interaction be-

tween those two units. This paper is organized as fol-

lows: In section 2, spectroscopy in its different types

is presented, followed by an overview of the devel-

oped system in section 3 and an outline of the com-

bined measurement head of camera and spectrometer

unit in section 4. In section 5, the autofocus of the

camera and the spectrometer unit as well as the cali-

bration between the two units is explained. The paper

is concluded with a brief discussion of this work and

a notion of future work is given.

2 ESTABLISHED

MEASUREMENT PRINCIPLES

Spectroscopy is the study of the interaction be-

tween matter and electromagnetic radiation (Skoog

and West, 1980). While there are other types of

spectroscopy such as vibrational (Long, 1977) and

emission spectroscopy (Radziemski and Cremers,

2006), this paper discusses reﬂectance spectroscopy

(Kort

um, 2012) since this type is used in the presented

measurement system. Reﬂectance spectroscopy is

concerned with the materials’ light absorption and

reﬂectance properties. Different substances reﬂect

wavelengths of light distinctly differently so that the

reﬂectance characteristic of an object can give some

indication of the object’s material or at least help to

differentiate between two objects that look alike to the

human eye.

Hegemann T., BÃijrger F. and Pauli J.

Combined High-Resolution Imaging and Spectroscopy System - A Versatile and Multi-modal Metrology Platform.

DOI: 10.5220/0006134602150222

In Proceedings of the 5th International Conference on Photonics, Optics and Laser Technology (PHOTOPTICS 2017), pages 215-222

ISBN: 978-989-758-223-3

215

It is possible to differentiate between 1-

dimensional and polydimensional measurement

techniques. The former denotes a point measurement,

also called a whisk broom scanner (Green et al.,

1998), in which the full spectral data is obtained for

a single point, while the latter applies to line-scan

systems, area scanning and snapshot imaging. A

line-scan system, also known as a push broom

scanner (Weser et al., 2008), is able to acquire the

spectral data for one line instead of a single point.

For an image with a width of x pixels and a height of

y pixels, the push broom scanner obtains spectral data

(λ) for every pixel x in one line (y

) simultaneously.

To acquire the whole data cube, the push broom only

has to scan each of the y lines, making the method

faster than a whisk broom scanner which has to

scan x × y times in order to obtain the whole data

cube. However, the push broom scanner is not as

versatile as a whisk broom scanner, as the width of

the scanning unit is ﬁxed and it is thus not possible

to freely choose the part of the sample for which to

acquire hyperspectral data. Area scanning, also called

spectral scanning (Dickinson et al., 2001), is another

technique which allows for surveying the whole

image monochromatically, meaning that a value for

every pixel of the image is obtained for one certain

wavelength λ. This can be achieved either by using

a broadband light source and an optical band-pass

ﬁlter, which rejects all but one wavelength and may

be tunable or has to be replaced in order to allow for

more wavelengths to be scanned, or alternatively by

using a light source with adaptable spectrum. With

this technique, it is easy to choose certain spectral

bands for the whole image. However, in order to

acquire data of several wavelengths, the station has

to be stationary, as otherwise spectral smearing can

occur, thus deteriorating correlations of an object at

multiple wavelengths. Finally, snapshot imaging, also

referred to as non-scanning (Volin et al., 2001), is a

technique which allows for obtaining the whole data

cube at once. This is naturally very time-efﬁcient, but

computational effort and initial costs are high.

For the metrology platform presented in this pa-

per, a point scanner was chosen as it is the most ver-

satile of the described techniques and also the most af-

fordable. Due to the positioning robot, which allows

for nanometer-exact positioning, a high-resolution

spectral image can be obtained for heterogeneous

samples. Furthermore, the choice of the character-

istics of the spectrometers is ﬂexible and almost any

wavelength range can be analyzed. The only draw-

back is the relatively slow capturing process com-

pared to the other recording approaches.

3 SYSTEM OVERVIEW

Atop of a vibration damped optical table, a three axis

Cartesian robot with nanometer-precision and a com-

bined measurement head for large area gray value

images as well as spectral measurements is placed.

A lighting system encompassing LEDs in different

colors and infrared along with an ultraviolet light

provides the means to acquire multispectral images,

while a broadband halogen light facilitates hyperspec-

tral image acquisition with the aid of spectrometers.

An overview of the system’s exterior, with the impor-

tant instruments and parts of CHRISS, as well as the

system’s main components of the interior, is presented

in Figure 1.

3.1 Enclosure

The core of the system is mounted atop the vibra-

tion damped optical table manufactured by the Opta

GmbH and is surrounded by a light-proof enclosure

construction with a canvas hull. Building vibration

is a problem when dealing with high-precision image

acquisition. In order to handle this problem, a table

which absorbs most of the vibration is used, thus en-

suring that the measurement of the sample is not cor-

rupted. A granite surface plate manufactured by the

company Johann Fischer Aschaffenburg is used as the

precise base for transacting measuring and inspection

tasks. It guarantees a ﬂatness of the measuring sur-

face according to DIN 876 and is thus important for

nanometer-precise measurements.

Additionally, the inﬂuence of variable daylight

has to be minimized to ensure that measurements of

the same sample are always comparable. For that rea-

son the opaque enclosure is important, ensuring that

only the internal lighting system illuminates the sam-

ple. Since the user has to be able to access the interior

to, for example, replace samples, a roller blind has

been installed on the front of the canvas hull which

can be lowered to make the enclosure light-proof.

Due to the fact that the stages of the Cartesian

robot are only guaranteed to operate with nanometer-

precision at 20 degrees Celsius, the room is equipped

with an air conditioning unit. Since the lighting sys-

tem, especially the halogen light, causes the air within

the enclosure to heat up, a ventilation system inter-

changing the warm interior and the cool exterior air

is necessary to ensure the high-precision work of the

positioning robot. A thermometer system with three

measuring points provides information about the tem-

perature within the enclosure so that the user is able

to assess the precision of the positioning system.

PHOTOPTICS 2017 - 5th International Conference on Photonics, Optics and Laser Technology

216

UV spectrometer

Ventilation

system

Light proof

enclosure

Network-

controlled

socket outlet

Ventilation

control

Optical table

Stages

control

system

Color LED

lights

IR light

UV light

Measurement

head

X stage

(a) (b)

Figure 1: Overview of the measurement system. Figure (a) shows the exterior of the measurement system. Figure (b) shows

the main components of the interior. The NIR spectrometer is not visible, because it is behind the enclosure.

3.2 Cartesian Robot

The Cartesian Robot is equipped with a three-axis

high-precision positioning system manufactured by

the company Aerotech. The x-stage may carry a high

payload of up to 25kg over a distance of 40 cm with

a speed of up to 100 mm/s. It is an air-bearing lin-

ear stage which allows for positioning accuracy of

1 nanometer of samples. On the other hand, the y-

and z-stage are connected and hold the measurement

head. The y-stage is a cross roller bearing stage and

is used to position the measurement head over the re-

gion of interest, while the z-stage is an air-bearing lin-

ear stage and its position determines the focus of the

measurement head with respect to the sample. Since

the focal length of both the gray value camera and the

spectrometers is ﬁxed, the distance of the measure-

ment head to the sample has to be adjusted to obtain

images in focus.

3.3 Illumination System

The illumination system consists of four white LED

lights, three of which are equipped with an additional

color ﬁlter to obtain blue, green and red light. Further-

more an infrared (IR) LED and an ultraviolet (UV)

light are installed so that images beyond human vi-

sion can be acquired. The illumination system is man-

aged by a network-controlled socket outlet, the All-

net ALL4076, so that each light can be switched on

and off automatically by the system. Using these six

lights and the gray value camera, it is possible to ac-

quire multispectral images by taking one image with

the camera for each light source thus obtaining a mul-

tispectral image with 6 spectral channels (gray value,

blue, green, red, UV and IR). Additionally, a broad-

band halogen lamp is mounted near the measurement

head and moves together with it in order to produce

light for the spectrometers. The halogen light offers

a broad spectrum of light ranging from approximately

400nm to 1700nm, thereby providing light for the ma-

jority of the acquirable spectrum, while neglecting the

ultraviolet region. The acquisition of a light type more

suitable to the task, which ensures that data in the UV

region as well as in the visible and near IR spectrum

can be obtained, like a confocal broadband ﬁber op-

tics light source, is planned for the future.

4 MEASUREMENT HEAD

The measurement head is a combined system of cam-

era and spectrometer units in which both units are

mounted on rails along the z-axis. The measurement

location of camera and spectrometers are distinct be-

cause of the offset between camera and spectrome-

ter unit along the y-axis, as can be seen in Figure 2.

Since the relative position of camera and spectrome-

ter unit can be calibrated (see section 5.3), it is pos-

sible to scan an area ﬁrst with the camera in order to

ﬁnd the desired section of the sample and then acquire

the full spectral range in one or several deﬁned points

by means of the spectrometer module. This combina-

tion of camera and spectrometer units, along with the

high-precision positioning system, offers a versatile

approach of scanning large areas with the camera to

ﬁnd regions of interest which can be examined more

closely by utilizing the spectrometers.

Combined High-Resolution Imaging and Spectroscopy System - A Versatile and Multi-modal Metrology Platform

217

Camera

Image sensor

Camera lens

Objective tube

Fiber to NIR

spectrometer

=1000-1700nm

Fiber to UVVIS

spectrometer

=190-1000nm

Microscope lens

Dichroic filter

Sample

X stage

Spectrometers

measurement

point/area

Y stage

Z stage

Mounting rails

Objective tube

Camera

area

Light di usor

tube

(a) (b)

Figure 2: Detail description of the measurement head. Figure (a) shows the components and the light paths. Figure (b)

depicts the actual system with the side-mounted light source for the spectrometers.

4.1 Camera Module

For the camera module, the industrial gray value cam-

era Baumer SXG80 with a resolution of 3296 × 2472

pixels (8 megapixel CCD sensor) is used. In com-

bination with the objective inspect.x L 105mm man-

ufactured by the company Qioptiq, which allows for

2x magniﬁcation while featuring very low optical dis-

tortion, it is utilized to obtain high-resolution surface

scans of a sample. The camera module is especially

applicable when inspecting large samples for defects,

as it is used to localize regions of interest. In combi-

nation with the illumination system, it is possible to

acquire multispectral images with 6 waveband chan-

nels, as can be seen in ﬁgure 3. Although the cam-

era may only acquire images of a small section of the

sample, i.e. approximately 2 × 2 cm, the position-

ing system in combination with a stitching module

facilitates the attainment of large-area images. Image

stitching allows for the acquisition of large scale im-

ages, but poses the problem of possibly creating visi-

ble seams where two images meet. Due to the fact that

the camera in CHRISS is not perfectly aligned with

the positioning stage, it is necessary to compensate

the problem of slightly rotated images by cropping

them, thus only using the inner part of the overlap-

ping images. In this way, the stitching module yields

large area scans of the surface of the sample without

visible seams, which can be seen in Figure 4.

In order to be able to reduce the impact of lumi-

nous reﬂectance and unevenness of samples, a light

dispersion unit can be utilized. This unit is placed be-

tween camera objective and sample so that light from

the light source in use does not directly illuminate

the sample but the dispersion unit instead, which pro-

vides a uniform light distribution and ensures a dif-

fuse illumination setting. As a consequence, shad-

ows casted by surfaces with 3D structure can be mini-

mized. Thus, images of samples are generally less af-

fected by problems of uneven surfaces and reﬂections

caused by direct illumination of glossy surfaces. To

this effect, either a semi-transparent cylindrical dis-

persion unit for a diffuse bright ﬁeld or an opaque

cylinder for a dark ﬁeld can be utilized, the results

of which can be seen in Figure 5.

4.2 Spectrometer Module

The presented measurement system utilizes re-

ﬂectance spectroscopy to facilitate a more detailed ex-

amination of samples. The spectrometer module con-

sists of two separate spectrometers which are com-

bined in order to acquire a wide range of spectral data

for a single point. For the purpose of obtaining data

ranging from the near ultraviolet region (190-390 nm)

to the visible spectrum (390-700 nm) up to the begin-

ning of near infrared data (700-1,100 nm) the BaySpec

Super Gumet UV-NIR spectrometer is utilized, while

the BaySpec Super Gamut NIR spectrometer, with a

range of 900 to 1,700 nm, is used to extend the exam-

inable scope of IR data. To acquire the full spectrum

(190-1,700 nm) at one single point, the spectrometers

are combined using a dichroic mirror that serves as a

ﬁlter which reﬂects light of a certain spectrum while

allowing the remaining light to pass through. In this

case a dichroic mirror at λ = 1,000 nm is used, mean-

ing that light below 1000 nm is reﬂected while light

above 1000 nm passes the ﬁlter unimpeded. From the

PHOTOPTICS 2017 - 5th International Conference on Photonics, Optics and Laser Technology

218

(a) Red (b) Green

(e) IR (f) UV

Figure 3: Multispectral image channels with denoted light

sources. The sample is a colored calibration pattern printed

on paper.

dichroic ﬁlter, two ﬁber optic cables lead to the re-

spective spectrometers where the optical signal is dis-

cretized. In this way, it is possible to have two spec-

trometers acquiring data from nearly the same area,

as depicted in Figure 2 (a). In order to minimize the

size of the area, thus allowing for more accurate mea-

surements with higher resolution, a microscope lens,

namely the Nikon TU Plan Fluor 10x, is attached to

the end of the objective tube of the spectrometer head.

In this way, a point resolution of 60 µm is achieved.

A total of 3,904 measurements can be obtained

with the two spectrometers for each sample point,

3,648 of which are conducted by the UV-NIR spec-

trometer, which amounts to 3.97 measurements per

nanometer of the spectral band, while the NIR

spectrometer offers an additional 256 measurements

with a spectral resolution of 0.32 measurements per

nanometer. Even though the resolution of the NIR

spectrometer is inferior to the other spectrometer,

both of them are important to ensure the applicability

of the system to a wide variety of tasks, in particular

those for which the IR spectrum is signiﬁcant, as, for

example, in quality inspection of steel surfaces.

(a)

(b) (c)

Figure 4: (a) Demonstrates the rotation shift between cam-

era and stage. (b) Shows naively stitched image from 4

input images without rotation correction, (c) the result ob-

tained by the presented stitching module with rotation cor-

rections.

5 CAMERA TO SPECTROMETER

CALIBRATION

Calibration is a very important topic in the scope of

CHRISS. This is due to the fact that the ability to ob-

tain the point spectrum for each pixel in the camera

image is desirable. The ﬁrst aspect of the proposed

calibration method is the focus of the camera in or-

der to achieve sharp images. To this effect, an autofo-

cus module has been developed, which is described in

(a) No dispersion (b) Bright ﬁeld (c) Dark ﬁeld

Figure 5: Image of reﬂecting surface with and without dis-

persion units. Shadows and reﬂections can be signiﬁcantly

reduced by using bright and dark ﬁeld. However, more pow-

erful lights are required for the dark ﬁeld.

Combined High-Resolution Imaging and Spectroscopy System - A Versatile and Multi-modal Metrology Platform

219

section 5.1. Since the spectrometer unit is also an op-

tical device, it needs to be focused as well. Section 5.2

provides the speciﬁcation of the spectrometer focus.

Subsequently, the coordinate system of the spectrom-

eter unit can be calibrated to the camera’s coordinate

system, in order to obtain the point spectrum for each

pixel in the camera image.

5.1 Camera Focusing

In order to obtain camera images in focus, the dis-

tance of the camera to the sample is important so that

the focal point of the camera coincides with the posi-

tion of the image plane. Therefore, the camera has to

be moved along the z-axis until the focal length con-

curs with the distance of the image plane to the lens.

To be able to automatically evaluate the image focus,

an image-based approach has been chosen. A focus

metric was implemented based on the Brenner gradi-

ent described in (Yazdanfar et al., 2008). The Brenner

gradient was extended to obtain a metric in which not

only vertical edges are considered but also horizontal

edges. The modiﬁed Brenner metric B is calculated

B =

∑

i=1

∑

j=1

(

|s(i−1, j)−s(i+1, j)|+|s(i, j−1)−s(i, j+1)|

)

where s(i, j) is the gray value of pixel (i, j) and N and

M are the number of horizontal and vertical pixels re-

spectively. Since an edge in an image induces a sig-

niﬁcantly higher contrast between neighboring pixels,

B increases with more edges in an image and since in-

focus images contain more sharp edges, B is accord-

ingly higher for sharp images. The metric B has to

be maximized in order to obtain in-focus images and

according to (Yazdanfar et al., 2008) a focus function

f depending on the vertical position of the camera z

can be approximated by a Lorentzian function

f (z) = y

∗

(z−z

)

+α

with α specifying the width, z

the center and y

the

offset of the curve in y-direction.

A hybrid approach for the autofocus of CHRISS

has been chosen, which comprises a naive and a curve

ﬁtting approach. The naive approach is basically a

grid search for the optimal z so that B is maximized.

The curve ﬁtting approach is faster but less accu-

rate and works with three images taken from differ-

ent heights which sufﬁces to ﬁt a Lorentz function

to the data points using MSE-minimizing curve ﬁt-

ting. A Lorentzian function has a single maximum at

. The z-coordinate corresponds directly to the op-

timal z-position of the camera to obtain in-focus im-

ages. Combining both approaches also combines the

advantages, i.e. the approach is faster than the naive

grid search but gives more accurate results compared

to the curve ﬁtting approach. On average, the cam-

era is positioned within 0.039 mm of the optimal z-

position in 3.67 seconds.

5.2 Spectrometer Focusing

In order to focus the spectrometer unit, an adapta-

tion of the Brenner-based camera image focus pro-

cess is proposed. Since the spectrometers do not cap-

ture an image but instead just a single ”pixel” with

almost 4000 different wavelengths, the focusing ap-

proach described above cannot be applied without ad-

justments. However, the principle of examining the

sharpness of edges is still valid. The idea is to use

a sample pattern with two distinct areas, a black and

a white one. After positioning the spectrometer head

over the white part of the sample in an initial z po-

sition, the measurement head is moved in the y di-

rection, to the black area of the sample. In between,

several measurements are carried out so that the inten-

sity value for each waveband and each position on the

y-axis is obtained. Using only the wavelength with

the highest variation, it is possible to locate the point

of transition from white to black by determining the

maximum change between two measurements. This

is reasonable as it can be assumed that the difference

between two measurements is higher for sharp tran-

sitions and lower for blurry edges. The spectrome-

ter head is moved in the z-axis after the procedure of

ﬁnding the maximal intensity difference ∆I

, and the

next maximal intensity difference is determined. In

this way it is possible to ﬁnd the z-position in which

the spectrometer unit delivers sharp measurements by

determining the z-position in which the maximum

change max(∆I

) occurs.

Since the measurement system comprises two sep-

arate spectrometers and since both of the optical ﬁber

cables were manually positioned and ﬁxated, and are

thus not centered perfectly on the optical axis, the

focus point of the two spectrometers differs slightly.

It is therefore possible to ﬁnd different optimal z-

positions for both spectrometers, however, to allow

for faster intensity value acquisition of all wave-

lengths in one point, the mean value of the two best

z-positions is taken as the optimal z

best

5.3 Point Calibration

Point calibration describes the process of ﬁnding the

same point in a camera image and with the spectrom-

eter unit. This should be possible so that regions of

interest, which were found by analyzing the image ob-

tained with the camera, can be examined more closely

by means of utilizing the spectrometers. In order to

PHOTOPTICS 2017 - 5th International Conference on Photonics, Optics and Laser Technology

220

(a) Coarse calibration pat-

tern.

(b) Fine calibration pattern.

Figure 6.

ﬁnd the same point with camera and spectrometers,

the coarse calibration pattern, which is shown in Fig-

ure 6 (a), is used. It is easily possible to ﬁnd the cor-

ner at the center of the pattern using a regular image,

however, since the spectrometers only scan one point

at a time, it is a challenging task to ﬁnd the corner

using the spectrometers. In order to do so, the user

places the calibration pattern approximately under the

spectrometer head and a number of measurements are

taken along the y-axis for the purpose of ﬁnding the

horizontal edge, similarly to detecting the edge when

focusing the spectrometers. When the y-coordinate of

the edge is found, the measurement head is moved in

the x-axis and the process of ﬁnding the edge is re-

peated so that several y-coordinates are found corre-

sponding to the horizontal edge. In a respective way

several x-coordinates are determined for the vertical

edge. The intersection point of the two lines obtained

in this way corresponds to an approximation of the

target corner of the calibration pattern. Since this is

only a coarsely estimated target point, it is possible

to improve upon the results by taking an array of hy-

perspectral measurements in the vicinity of the found

intersection point, as can be seen in Figure 7 (a). The

obtained data can be regarded as a large number of

images by considering a single wavelength at a time,

thus obtaining a total of 3,904 images. Each image is

preprocessed with morphological ﬁlters using struc-

turing elements (Efford, 2000). By utilizing a combi-

nation of the Canny edge detector (Canny, 1986) and

Hough transform (Duda and Hart, 1972), it is pos-

sible to ﬁnd a proper representation of the two per-

pendicular lines in each image. The intersection of

the two lines is determined for each image and the

median coordinates of the intersection points are cal-

culated in order to ﬁnd an accurate representation of

the target corner point. To ﬁnd the exact offset of the

camera to the spectrometer head, several more aspects

are taken into consideration. It is necessary to ﬁnd

the size of pixels in the camera image to translate a

distance in the image coordinate system (CS) to the

positioning stage CS and also determine the rotation

(a)

(b)

(c)

(d)

(e)

Figure 7: Calibration results. (a) shows the selected area

of the image with green dots indicating spectrometer mea-

surement locations. (b) and (c) reveal the results for UV-

NIR and NIR spectrometer respectively after coarse cali-

bration, i.e. blurry lines and corner does not match center

point (black plus). (d) and (e) show improved results with

sharper lines and lower offset between center and corner

for the two spectrometers with wavelengths of 702 nm and

1360 nm respectively.

between camera and the stage. Since the camera was

installed manually, it is not perfectly aligned with the

axis of the stage, therefore a rotation parameter has

to be found in order to move the stage properly with

respect to the information provided by the camera im-

age. It is also necessary to consider that the found

intersection point differs for the two spectrometers,

as explained above, hence the mean value of the two

coordinates of the intersection points is used as an es-

timate of the corner. To improve the results, a ﬁne

calibration pattern

, cf. see Figure 6 (b), manufac-

tured by means of a laser instead of a printer and thus

more precise, is used. The results of calibrating with

both patterns can be seen in ﬁgure 7. The calibration

process takes about 67 minutes both for coarse and

ﬁne calibration, but it only has to be done once un-

USAF-1951 Standard Target (T-20).

Combined High-Resolution Imaging and Spectroscopy System - A Versatile and Multi-modal Metrology Platform

221

less the hardware setup of the measurement system

is changed. Experiments show that the hyperspectral

images reach a lateral resolution of 60 micrometers.

6 CONCLUSIONS

With CHRISS, a unique, ﬂexible and versatile mea-

surement system has been developed which contains

a high-resolution camera able to obtain multispectral

images, also considers a wide area of wavebands due

to the combined spectrometer unit and utilizes a high-

precision positioning system which allows for thor-

ough analysis of large samples. The hardware of the

attained system was presented in detail, and solutions

to the major problems of calibration and focus have

been described. Construction and implementation of

the CHRISS project is ﬁnished in general but can be

extended in the future to improve the performance, for

example by installing a more suitable light source or

by using pinholes at the junction of the spectrometer

head and the optical ﬁber cables of the spectrometers.

This would improve the lateral resolution of hyper-

spectral point measurements, but is only applicable in

combination with a more powerful light source as less

light is able to reach the ﬁbers and thus the spectrom-

eters.

The system may now be used to acquire multi- or

hyperspectral data which can be analyzed. Current

projects work on clustering of hyperspectral image

data and classiﬁcation of different materials in multi-

spectral space by means of Convolutional Neural Net-

works (CNN). The applicability of CNNs to hyper-

spectral image data and its classiﬁcation results will

be explored in the future in order to create a measure-

ment system that holistically analyzes a large sample,

detects regions of interest and is able to examine them

more closely to detect anomalies in an otherwise ho-

mogeneous sample.

ACKNOWLEDGEMENTS

This work was funded by the European Commis-

sion within the Ziel2.NRW programme “NanoMikro

+Werkstoffe.NRW”.

REFERENCES

Canny, J. (1986). A computational approach to edge de-

tection. IEEE Transactions on pattern analysis and

machine intelligence, (6):679–698.

Dickinson, M., Bearman, G., Tille, S., Lansford, R., and

Fraser, S. (2001). Multi-spectral imaging and lin-

ear unmixing add a whole new dimension to laser

scanning ﬂuorescence microscopy. Biotechniques,

31(6):1272–1279.

Duda, R. O. and Hart, P. E. (1972). Use of the hough trans-

formation to detect lines and curves in pictures. Com-

munications of the ACM, 15(1):11–15.

Efford, N. (2000). Digital image processing: a practical

introduction using java (with CD-ROM). Addison-

Wesley Longman Publishing Co., Inc.

Green, R. O., Eastwood, M. L., Sarture, C. M., Chrien,

T. G., Aronsson, M., Chippendale, B. J., Faust,

J. A., Pavri, B. E., Chovit, C. J., Solis, M., et al.

(1998). Imaging spectroscopy and the airborne vis-

ible/infrared imaging spectrometer (aviris). Remote

Sensing of Environment, 65(3):227–248.

Kort

um, G. (2012). Reﬂectance spectroscopy: principles,

methods, applications. Springer Science & Business

Media.

Long, D. A. (1977). Raman spectroscopy. New York, pages

1–12.

Panasyuk, S. V., Yang, S., Faller, D. V., Ngo, D., Lew, R. A.,

Freeman, J. E., and Rogers, A. E. (2007). Medical hy-

perspectral imaging to facilitate residual tumor iden-

tiﬁcation during surgery. Cancer biology & therapy,

6(3):439–446.

Radziemski, L. J. and Cremers, D. A. (2006). Handbook of

Laser Induced Breakdown Spectroscopy. John Wiley

& Sons, West Sussex, England.

Skoog, D. A. and West, D. M. (1980). Principles of in-

strumental analysis, volume 158. Saunders College

Philadelphia.

Volin, C. E., Garcia, J. P., Dereniak, E. L., Descour, M. R.,

Hamilton, T., and McMillan, R. (2001). Midwave-

infrared snapshot imaging spectrometer. Applied op-

tics, 40(25):4501–4506.

Weser, T., Rottensteiner, F., Willneff, J., Poon, J., and

Fraser, C. S. (2008). Development and testing of a

generic sensor model for pushbroom satellite imagery.

The Photogrammetric Record, 23(123):255–274.

Yazdanfar, S., Kenny, K. B., Tasimi, K., Corwin, A. D.,

Dixon, E. L., and Filkins, R. J. (2008). Simple and ro-

bust image-based autofocusing for digital microscopy.

Opt. Express, 16(12):8670–8677.

PHOTOPTICS 2017 - 5th International Conference on Photonics, Optics and Laser Technology

222