XSpectra®: an Advanced

Real-Time Food Contaminants Detector

B. Garavelli, P. Pozzi, D. Macera, L. Zanotti, A. Mencarelli, G. Bubba, P. Bertoni, M. Sammartini, M. Bettelli, G.
Bertuccio, G. Ghiringhelli, A. Zappettini

 Abstract–An innovative X-ray inspection technology, named

XSpectra®, has been developed with the aim to improve the
current state of art in the field of real-time detection of
contaminants in food products on production lines. The
technology architecture is based on modules equipped with a 128
pixels CdTe array detector each read-out by full-custom Front-
End ASICs. A full-custom Multi-Channel-Analyzer reconstructs
the radiation spectrum, which is then processed by advanced
Neural Network algorithms performing both image
reconstruction and foreign bodies detection. The experimental
characterization of XSpectra® has demonstrated the sensitivity of
the fully operating system to photon energies down to about 10 Figure 1. Simplified drawing of a real-time X-ray inspection system on
keV at events rates up to several millions of photons per second. a production line
A line-width of 8.5 keV FWHM has been measured, at room
temperature, on the 60 keV photo-peak of a synchrotron while they fail in detecting low density contaminants made of
radiation in low-rate conditions. A spectral non-linearity error a wide number of materials including plastic, wood, organic
within ±0.5% has been obtained within the energy range 25 keV and, in general, other small unwanted foreign bodies made of
－ 100 keV. The effective capability of XSpectra® to detect
very-light materials which accidentally end up in the products
currently undetectable low-density contaminants inside real food
products has also been proved.
during their treatment on the production lines. With the aim to
improve the current state of art in the field of industrial
Index Terms–Non-destructive-test X-ray equipment, CdTe controls for food products, we have thus developed XSpectra®
detectors, Multi-spectral analysis [2], a proprietary and novel X-ray inspection technology able
to detect in real-time both low- and high- density contaminants
in foodstuff. In the following paragraphs, the architecture of
XSpectra® and some experimental results will be presented.

I. INTRODUCTION II. XSPECTRA® REQUIREMENTS

F ood contamination is a regular occurrence all over the

world. According to the Rapid Alert System for Food and
Feed (RASFF) [1], 267 alerts about foreign bodies and
Figure 1 shows a simplified drawing of a typical X-ray
inspection system operating on a production line. An X-ray
tube generates a blade-like photons-beam that irradiates the
contaminants found in foodstuff occurred in 2017, with a trend products moving by means of the conveyor belt of the line.
that has not been decreasing in the last 5 years. Among the The beam interacts with the products and is then collected by
broad range of technologies nowadays available for food an X-ray detector for spectroscopic images reconstruction and
inspection, X‑ray detection machines represent by far the best processing. Conventional X-ray inspection machines,
solution, due to their ability to both spot a much wider range available today in the food market, work by just merely
of contaminants and also inspect products at different stages of counting the photons in overall received by the detector
their production process. Despite this, the existing X‑ray (single-energy approach) or at most distinguishing those
technologies are currently not able to completely solve the received in a couple of energy windows (dual-energy
issue of contaminated food as they can detect high density approach). XSpectra® has instead been designed in order to
foreign bodies, such as metal, lead, stones, bones and glass, perform a real-time multi-spectral analysis, up to 1024 energy
bands, with the specific goal to be able to also detect the low-
Manuscript received November 29, 2019. density contaminants that are nowadays invisible; this requires
B. Garavelli, P. Pozzi, D. Macera, A. Mencarelli, G. Bubba, P. Bertoni are XSpectra® to be sensitive to photon energies well below 20
with Xnext s.r.l., via Adelaide Bono Cairoli 30, 20127 Milano, Italy keV. The high speed of the conveyor belts typically adopted
L. Zanotti was with Xnext s.r.l., via Adelaide Bono Cairoli 30, 20127
Milano, Italy. He is now with ASML, North Brabant Province, Netherlands on production lines in the food market (up to 1 m/s) requires
M. Sammartini, G. Bertuccio, G. Ghiringhelli are with Politecnico di the use of high photon fluxes in order to collect spectra with a
Milano, Piazza Leonardo da Vinci 32, 20133 Milano, Italy sufficient statistics; this requires XSpectra® to have
N. Bettelli, A. Zappettini are with IMEM-CNR, Parco Area delle Scienze
37/A, 43124 Parma, Italy
spectroscopic capabilities up to rates of several millions of

978-1-7281-4164-0/19/$31.00
Authorized licensed use limited to: ULAKBIM UASL - Akdeniz ©2019 03,2023
Universitesi. Downloaded on December IEEE at 07:10:20 UTC from IEEE Xplore. Restrictions apply.
 Figure 4. Spectrum at temperature T = +20 °C obtained by irradiating
one pixel with a 100 μm × 100 μm area beam of 60 keV photons incident
on its geometrical center at rates from 0.3×106 up to 2.5×106 cps.
Figure 2. Simplified blocks schematic of XSpectra®
block of the system, collects the data-streams produced by all
photons per second per pixel. Finally, XSpectra® must be also the MCAs and delivers them to the Application Software,
able to work at photons energies up to 130 keV. Putting which is the third and last macro-block. The Application
together all such requirements, makes the XSpectra® design a Software temporally re-orders the data delivered by all the
real challenge. MCAs and processes them by means of advanced proprietary
Neural Network algorithms to both reconstruct the product
III. XSPECTRA® ARCHITECTURE image and also detect the presence of eventual contaminants
Figure 2 shows a simplified blocks schematic of XSpectra®. inside it. In case a contaminant is detected, an EJECT signal is
The first macro block is the Detector Electronic Front-End, generated to drive the line ejector which removes the product
which has a modular architecture. Each module comprises a from the line belt. Figure 3 shows a picture of a Detection Unit
CdTe linear-array of 128 pixels - with 0.8 mm pitch - which in its version equipped with n. 3 Detector Electronic Front-
detects the photons emitted by the X-ray source after having End modules, which has dimensions 36 × 37 × 11 cm3. The
passed through the scanned product. The signals of each pixel placements of the collimated slit for photons entrance, of the
are read-out by a multi-channel full-custom Analog Front-End interface to the Data Collector and of the liquid-cooling plate
Application Specific Integrated Circuit (AFE-ASIC), realized are also shown in the figure.
in CMOS technology and developed in collaboration with
Politecnico di Milano. The AFE-ASIC performs the very first IV. EXPERIMENTAL RESULTS WITH SYNCHROTRON
analog processing on the signals and has been designed in RADIATION
order to obtain the challenging best compromise in terms of XSpectra® modules have been experimentally characterized
industrial reliability, low-power, high photon-rate and low- at beam line BM05 of the European Synchrotron Radiation
noise capabilities. The ASICs feed the Multi Channel Facility (ESRF) in Grenoble (France). The measurements have
Analyzer (MCA) that is composed by a high sampling-rate been performed at temperature T = +20 °C by irradiating a
Analog to Digital Converter (ADC) followed by a FPGA- sub-set of the overall 128 pixels-array by means of a 100 μm ×
based full-custom Digital Signal Processor. Each MCA 100 μm area beam of monochromatic photons, at different
reconstructs the energy spectrum of the X-ray radiation energies and rates, centered in the middle of the tested pixels.
detected by its own 128-pixels array and transfers it to the Figure 4 shows the raw spectra obtained with one of the tested
Data Collector. This last, which constitutes the second macro- pixels by irradiating it with 60 keV photons at incoming rates
ranging from 0.3×106 counts-per-second (cps) up to 2.5×106
cps. The main 60 keV line width resulted to be 8.5 keV
FWHM at the lowest-rate measurement condition and 13.5
keV FWHM at the highest one. The escape-peaks and pile-up
regions can be clearly appreciated. As can be seen in the
figure, the system noise tail affects the lowest-energy part of
the spectrum below about 10 keV at all the tested photon-rate
conditions, thus making the spectral information fully usable
by the algorithms approximately starting from such energy on.

Figure 3. Picture of a Detection Unit equipped with n. 3 Detector

Electronic Front-End modules, whose dimensions are 36 × 37 × 11 cm3.
The collimated slit for photons entrance, the interface to Data Collector
Figure 5. Non-linearity error measured between 25 keV and 100 keV.
and the liquid-cooling plate are also shown.

Authorized licensed use limited to: ULAKBIM UASL - Akdeniz Universitesi. Downloaded on December 03,2023 at 07:10:20 UTC from IEEE Xplore. Restrictions apply.
 figure shows the unprocessed radiographic image as produced
by XSpectra®, while part b) shows how the figure appears
after the processing performed by the XSpectra® algorithms,
which make the presence of the foreign body much more
evident.

VI. SUMMARY AND CONCLUSIONS

An advanced proprietary technology, named XSpectra®, has
been designed for performing quality controls on food
production lines by detecting, in real-time, food contaminants
which still constitute an issue for conventional technologies
Figure 6. Scan of a chopped tomato aluminium can with a rubber band currently adopted in such field. Tests have been carried out in
foreign body inside. Left-side: unprocessed radiographic image as order to characterize both XSpectra® intrinsic spectral
produced by XSpectra®, in which the rubber-band is only barely performances as well as its actual capability to detect critical
noticeable. Right-side: image produced by XSpectra® algorithms, in foreign bodies, principally made of small and light material
which the rubber-band contaminant is fully detected.
pieces which accidentally ends up in the food products
packages while they are managed during their production
The non-linearity error of the system has been measured by process. XSpectra® intrinsic spectroscopic characterization has
irradiating the same subset of pixels by means of photons with been carried out at the European Synchrotron Radiation
energies between 25 keV and 100 keV. The result is reported Facility (ESRF) in Grenoble (France), where a spectral line-
in Figure 5 and shows as the error lies within ±0.5% in all the width of 8.5 keV FWHM has been measured on the 60 keV
range. line of a monochromatic X-ray beam in low photon-rate
conditions, as well as its capability to process the spectral data
V. EXPERIMENTAL RESULTS WITH APPLICATIVE TESTS down to about 10 keV of photons energy, a boundary which is
XSpectra® performances have also been extensively tested extremely useful for the detection of light contaminants. A
at the Xnext s.r.l. facility in Milano (Italy) [2] on many non-linearity error of ±0.5% has also been measured in the
applicative cases directly required by worldwide companies photon energy range from 25 keV to 100 keV. Applicative
operating in the food market. Such tests consisted in using tests have also demonstrated the XSpectra® capability to
XSpectra® at its full functionality to scan real food-products in effectively detect foreign bodies which are nowadays still very
search of real-case contaminants which still represent an issue critical to be detected by conventional technologies.
in the food market as being still undetectable. Figure 6 shows,
as an example, a chopped tomato aluminium can with a VII. ACKNOWLEDGMENTS
contaminant inside as analyzed by XSpectra®. The aluminium Xnext s.r.l. would like to thank E. T. Capria, M. Ruat and S.
can is a cylinder of 7.5 cm diameter × 11 cm height with a Berujon of ESRF for their very precious support in both
side wall thickness of 1.5 mm. The contaminant is a circular organizing and performing the characterization activities on
rubber-band with 1 mm thickness × 5 mm width. The left-side XSpectra® at the synchrotron facility.
of the figure shows the unprocessed radiographic image of the
tomato can as produced by the system. The right-side of the
figure shows the result produced by XSpectra® algorithms, in VIII. REFERENCES
which the foreign body has been clearly detected. Figure 7
[1] https://ec.europa.eu/food/safety/rasff/
shows the results obtained by performing tests on filled-pasta
[2] http://www.x-next.com/
packages with foreign bodies inside; in particular, it shows the
case of a small plastic-band piece, with the size of just few
millimeters, ended up in the pasta package. Part a) of the

Figure 7. Scan of a filled-pasta package containing a few millimeters

sized plastic-bend foreign body. In detail: a) unprocessed radiographic
image as produced by XSpectra®; b) the same image as processed by
XSpectra® algorithms.

Authorized licensed use limited to: ULAKBIM UASL - Akdeniz Universitesi. Downloaded on December 03,2023 at 07:10:20 UTC from IEEE Xplore. Restrictions apply.