Using Image Processing Techniques for Automated Detection and Annotation of Faulty Regions in Thermal Infrared Images of PV Modules

International Journal of Trend in Scientific Research and Development (IJTSRD)
Volume 4 Issue 1, December 2019 Available Online: www.ijtsrd.com e-ISSN: 2456 – 6470
@ IJTSRD | Unique Paper ID – IJTSRD29749 | Volume – 4 | Issue – 1 | November-December 2019 Page 892
Using Image Processing Techniques for Automated
Detection and Annotation of Faulty Regions in
Thermal Infrared Images of PV Modules
Atilla Ergüzen, Muhammet Sait
Computer engineering, Kirikkale University, Kirikkale, Turkey
ABSTRACT
With the increasingly growing number of solar energy sites, the need for
better and faster fault detection techniques becomes more pressing. Using IR
imaging is reliable and effective but scanning thousands or even hundreds of
thousands of PV modules in mega sites quickly turns to be a time consuming
and tedious task. To save on the spent effort and time digital imageprocessing
techniques can be introduced into the inspection process to help identify and
annotate defects in the thermal footage of PV modules in an automated
manner. The methodology used in the paper relies onanalyzingthehistogram
to choose a suitable thresholding point that would help isolate the potential
faulty areas in a grayscale IR image.
KEYWORDS: Imageprocessing, photovoltaics, solarpower, thermalimaging, fault
detection
How to cite this paper: Atilla Ergüzen |
Muhammet Sait "Using Image Processing
Techniques for Automated Detection and
Annotation of Faulty Regions in Thermal
Infrared Images of PV Modules"
Published in
International
Journal of Trend in
Scientific Research
and Development
(ijtsrd), ISSN: 2456-
6470, Volume-4 |
Issue-1, December
2019, pp.892-895, URL:
www.ijtsrd.com/papers/ijtsrd29749.pdf
Copyright © 2019 by author(s) and
International Journal ofTrendinScientific
Research and Development Journal. This
is an Open Access article distributed
under the terms of
the Creative
CommonsAttribution
License (CC BY 4.0)
(http://creativecommons.org/licenses/by
/4.0)
INTRODUCTION
The world has been increasing its reliance on renewable
energy sources more than ever for supplying its needs of
power in the last years. The motive is to replace as much as
possible of the traditional energy sources such as fossil fuels
and coal because of their environmental impacts and their
finite quantities.
Solar energy is particularly attractive because it’s easy to
install in small and large scale sites alike, and its costs have
been decreasing consistently duringthepastdecade.[1]This
has led to expansion into new markets and deployment of
thousands of systems of various sizes. The use of
photovoltaic power systems in Turkey has been growing
fast, where the cumulative energy production has reached
5.062 GWp at the end of 2018 according to the statistical
data provided by Turkish Electricity Transmission
Corporation. [2]
The use of solar energy has been growing worldwide by
almost 40% per year since 2000. Although this rate may
have slowed down in some regions, it is still quite high in
many countries around the world, like India (42.8% growth
in 2018), China (50.7%), Turkey (173%) and many other
countries [3]. In 2019, it was estimated that PV systems
provides 2.6% of total global energydemandand4.3%ofthe
demand in Europe [4]. Some reports by IEA and IRENA
projected that solar energy would be the dominant
electricity source ahead of other fuelsandrenewablesbythe
2050, providing up to 26% of global energy generation [5],
[6].
All of this has encouraged more research in this field,
especially the improvement of various maintenance and
inspections methods that would help run those power
systems safely and more efficiently.
Using thermal imaging for fault detection
One of the commonly used approaches to examine PV
modules and discover any issues is to use thermal cameras
or thermal handheld devices. This method is quite effective
since it does not require any interruption of the system’s
work.
Cells and modules are scanned at daylight hours during
normal work conditions. Faulty PV cells, modules, or even
strings of modules can be detected as they experience an
abnormal increase in their temperature due to energy being
consumed by the faulty element or not being distributed
correctly throughout the system becauseoffaultywiring[7].
The problem with this approach is that it still requires
manual scanning of the field in case of handheld device use.
IJTSRD29749

International Journal of Trend in Scientific Research and Development (IJTSRD) @ www.ijtsrd.com eISSN: 2456-6470
The alternative is to mount a thermal camera to a UAV
(unmanned aerial vehicle) and fly it in a specified route to
cover the power plant entirely. This new approach has
become more appealing recently because of the increasing
availability of various models of drones and quadcopters
which are both affordable and powerful enough to carry the
imaging device for a sufficient period of time. However, the
footage gathered by the UAV is still manually inspected to
identify any present problems in the scanned modules.
Digital image processing for detecting faults
Many studies in the literature has discussed different
approaches for fault detection using image processing
techniques. In [9], Canny edge detection is used to identify
hot spots in gray thermal images. To reduce the noise in the
obtained images, smoothing and Gaussian filtering is
introduced. For aerial imagerywithinsufficientGSD(ground
sampling distance) this would erase many smaller potential
fault areas.
Environmental conditions affect the quality of the imaging
process, most notably wind which can cause blurring due to
disrupts in UAV stability, and the reflection of solar rays on
modules’ surfaces which results in over illuminated images
[8]. These elements make it more difficult todetecthotspots
using edge detection or other algorithms that would require
low noise levels in the images.
In [11], four sets of features are computed(modulemedians,
grid cell medians, histogram skewness, and vertical
projections) for PV modules detected in the thermal images.
They detect the defective modules based on temperature
abnormalities by conducting a Dixon’s Q test after a Grubbs’
test.
In another study [12], the panel areas are extractedfrom the
UAV thermal images and a methodology for automatic fault
detection is proposed using statistical values of thermal
intensity of each solar panel.
Another method is based on curve fitting of gray thermal
images’ histogram suggested by [13]. Using the least square
method, the histogram is fitted as a smooth curve
eliminating the blurs and local fluctuations present in the
image. A threshold is then selected based on the curve
characteristics in a way that would isolatethehotspotsfrom
the rest of the image.
In this work we rely on histogram analysis to select a
threshold point that would partition the image and isolate
the regions of interest (i.e. hot spots) in a way similar to the
method used in the curve fitting study.
The threshold is calculated from the image histogram using
the following formula, adapted from [13], and it represents
the last “valley” before the first significant “peak” in the
histogram curve:
eq (1)
; is the abscissa of the highest peak in the histogram
; is abscissa of all the “valleys”
This is assuming that the darker the pixel is the higher the
temperature is; which corresponds to the format of the
dataset images studied in this paper [10].
The steps of the complete algorithm used toprocessthedata
are as follows:
1. PV strings extraction: Isolate the solarmodulesfromthe
background;
2. Calculate the histogram for the isolated foreground
image only;
3. Select a threshold point;
4. Convert foreground image into BW using previous
threshold;
5. Highlight the resulting black spots in the original image.
Experimental Results
We tested our method on a thermal dataset of a solar field
installation which was collected in Switzerland by an eBee
Classic drone carrying a sense Fly thermo MAP camera. The
dataset consists of thermal images saved in grayscale.
Ground resolution (GSD) 14.23 cm (5.6 in)/px
Coverage 0.08 sq. km (0.03 sq. mi)
Flight height 70 m (229.6 ft)
Table1 Technical information of the image dataset
We applied the proposed method on a number of the frames
in the dataset. In all of the tested images the total faulty area
was <= 5% of the total area of the inspected solar modules.
Hotspots where represented by darker areas than the
average color tone of healthy PV cells due to their higher
temperatures compared to normally functioning cells.
It is crucial for the defective areas to be relatively small,
since the method used is assuming that the pixels with the
highest frequency of gray level represent healthy areas.
Under these conditions the histogram of the image (minus
any background) takes a shape that clearly identifies the
faulty pixels.

Figure1 A sample image from the dataset (left), isolated modules with background pixels filled with white (right,
border was added for clarity).
In Figure1 A sample image from the dataset (left), isolated modules with background pixels filled with white (right,
border was added for clarity). a sample image from the dataset is preprocessed, to isolate themodulestringsfromanyother
objects in the image. This step is required since we need to apply the next steps on the histogram of the pixels that represent
solar modules only.
Next step can be understood clearly by investigating the histogram of the gray pixels that we have from step 1. In Error!
Reference source not found., we have the histogram corresponding to the image shown earlier, with a zoomed in section to
further illustrate our region of interest. In this particular example, the optimal threshold point would have a gray value of 68.
Accordingly, we continue to partition the image with all pixels that have gray values darker than 68 turned into black, and
everything else turned white.
Figure2 Histogram of the cropped modules image ignoring white pixels (left), Zoomed-in section of the histogram
focusing on the region of interest (right).
After applying this threshold, we acquire an image which only contains potential fault locations. Since these areas still
correspond to their original locations on the starting image, we can highlight these regions in the original image and displayit
to the end user.

Due to the high altitude of the images taken the faulty areas
can be quite small. This has limited our ability to apply any
useful noise reduction steps during the preprocessing of the
images since it may result in losing some of the small hot
spot areas. We opted for an increased number of false
positives in our results instead. Despite that,wefoundin our
tests that the ratio of false positives to correctly marked
areas to be less than 14%. We expect this to improve in case
of application on higher resolutionimageryorloweraltitude
flights which would result in a better ground resolution.
Conclusions
Automated detection of hot spots in PV modules using
histogram analysis is studied in the paper. Experimentation
was done on a dataset of grayscale thermal images of a solar
installation site taken by a drone. Hot spots are isolated
using thresholding based on the histogram of each image.
This method gives good results under certain conditions,
most importantly the defect areas should be a minority
compared to the whole inspected area. However, it is simple
and quite effective compared to other methods in this
research area.
References
[1] S. E. I. Association, “Solar Industry Research Data,”
2019. [Online]. Available:https://www.seia.org/solar-
industry-research-data. [Accessed: 14-Dec-2019].
[2] T. electricity transmission Corporation, “Installationof
Electricity Systems in Turkey and Installed Power by
Resource,” 2018. [Online]. Available:
https://www.teias.gov.tr/sites/default/files/2019-
01/kurulu_guc_aralik_2018.pdf. [Accessed: 14-Dec-
2019].
[3] B. P. l. c., “BP Statistical Review of WorldEnergy,”2019.
[4] I. PVPS, “PVPS Snapshot of Global PV Markets,” 2019.
[5] I. E. Agency, “Energy Technology Perspectives 2014,”
2014.
[6] I. R. E. A. (IRENA), “GLOBAL ENERGY
TRANSFORMATION,” 2018.
[7] M. Cubukcu and A. Akanalci, “Real-time inspection and
determinationmethodsoffaultsonphotovoltaicpower
systems by thermal imaging inTurkey,”Renew. Energy,
vol. 147, pp. 1231–1238, 2020.
[8] M. Aghaei, A. Dolara, S. Leva, and F. Grimaccia, “Image
resolution and defects detection in PV inspection by
unmanned technologies,” in 2016 IEEE Power and
Energy Society General Meeting(PESGM),2016,pp.1–5.
[9] J. A. Tsanakas, D. Chrysostomou, P. N. Botsaris, and A.
Gasteratos, “Fault diagnosis of photovoltaic modules
through image processingandCannyedgedetectionon
field thermographic measurements,” Int. J. Sustain.
Energy, vol. 34, no. 6, pp. 351–372, 2015.
[10] SenseFly, “Solar Panel Installation.”[Online].Available:
https://www.sensefly.com/education/datasets/?datas
et=1416. [Accessed: 14-Dec-2019].
[11] S. Dotenco et al., “Automatic detection and analysis of
photovoltaic modules in aerial infrared imagery,” in
2016 IEEE Winter Conference on Applications of
Computer Vision (WACV), 2016, pp. 1–9.
[12] D. Kim, J. Youn, and C. Kim, “AUTOMATIC FAULT
RECOGNITION OF PHOTOVOLTAIC MODULES BASED
ON STATISTICAL ANALYSIS OF UAV
THERMOGRAPHY,” Int. Arch. Photogramm. Remote
Sens. Spat. Inf. Sci., vol. XLII-2/W6, pp. 179–182, Aug.
2017.
[13] J. Lin, S. Jianhui, and L. Xin, “Hot spots detection of
operating PV arrays through IR thermal image using
method based on curve fitting of gray histogram,”
MATEC Web Conf., vol. 61, 2016.

Using Image Processing Techniques for Automated Detection and Annotation of Faulty Regions in Thermal Infrared Images of PV Modules

More Related Content

What's hot (20)

Similar to Using Image Processing Techniques for Automated Detection and Annotation of Faulty Regions in Thermal Infrared Images of PV Modules (20)

More from ijtsrd (20)

Recently uploaded (20)

Using Image Processing Techniques for Automated Detection and Annotation of Faulty Regions in Thermal Infrared Images of PV Modules