Vous êtes sur la page 1sur 10

Infrared Physics & Technology 55 (2012) 236–245

Contents lists available at SciVerse ScienceDirect

Infrared Physics & Technology


journal homepage: www.elsevier.com/locate/infrared

Review

Recent progress in diagnosing the reliability of electrical equipment


by using infrared thermography
Mohd Shawal Jadin a,⇑, Soib Taib b
a
Faculty of Electrical & Electronic Engineering, Universiti Malaysia Pahang, 26600 Pekan, Pahang, Malaysia
b
School of Electrical and Electronic Engineering, Universiti Sains Malaysia, 14300 Nibong Tebal, P. Pinang, Malaysia

a r t i c l e i n f o a b s t r a c t

Article history: Infrared thermography (IRT) has gained more attention and become an interesting method in electrical
Received 16 September 2011 preventive maintenance due to its high precision and sensitivity imaging characteristics. This paper pro-
Available online 24 March 2012 vides a review of the application of IRT for diagnosing electrical equipment, including their thermal
anomalies and methods of measurement. Improvement of the inspection techniques is highlighted in
Keywords: order to investigate the reliability of electrical equipments due to the effect of the environmental factors
Infrared thermography and equipment condition. Factors related to the target equipment and the inspection tool together with
Thermal anomaly
their characteristics is also presented. Due to the complex analysis, various automatic diagnostic systems
Image processing
Artificial intelligence
are proposed for faster and more accurate analysis. Typical engineering solutions using recent technolo-
Qualitative gies are reviewed which could be used to improve the quality of IRT inspection.
Quantitative Ó 2012 Elsevier B.V. All rights reserved.

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236
2. Application of IRT for diagnosing electrical equipments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
2.1. Thermal anomalies in electrical equipment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
2.2. Improving inspection techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
2.3. Method of measurement and analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
3. Automated diagnostic system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
4. Future recommendations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
4.1. Advanced technology of infrared camera with the higher detector resolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
4.2. High resolution and continuous monitoring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
4.3. Accurate methods of fault diagnosis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
4.4. Intelligent diagnostic system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
4.5. Models sustainability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244
5. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244

1. Introduction thermal image on a video monitor and perform temperature calcu-


lations. Due to its advantages in terms of being non-contact, free
IRT is about the science of acquisition and analysis of thermal from electromagnetic interference, safe, reliable and can provide
information from non-contact thermal imaging devices. This de- large inspection coverage, IRT has taken a very important role in
vice is used to detect infrared energy (i.e. the heat) and converts predictive and preventive maintenance programs [1–5]. It can
it into an electronic signal, which is then processed to produce a instantly visualize and verify thermal profiles, quickly locate the
hot spots, determine the severity of the problem and help estab-
lishing the time frame in which the equipment should be repaired.
⇑ Corresponding author. Furthermore, inspections can be done without shutting down the
E-mail address: mohdshawal@ump.edu.my (M.S. Jadin). system operation.

1350-4495/$ - see front matter Ó 2012 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.infrared.2012.03.002
M.S. Jadin, S. Taib / Infrared Physics & Technology 55 (2012) 236–245 237

IRT can monitor the thermal behavior of electrical equipment 2.1. Thermal anomalies in electrical equipment
by sensing the emission of infrared energy (i.e. temperature) of
the equipment. Although heat is not a perfect indicator of all Electrical devices are usually rated for power, which indicates
problems in electrical systems, but heat produced by abnormally the amount of energy that the device can conduct without being
high electrical resistance often precedes electrical failures [6]. It is damaged. If the device is operated at a power above its specifica-
well understood that the life of electrical equipments is drasti- tions, the excess power causes overheating on the device which
cally reduced as temperatures increase. Condition monitoring is in turn reduces its life cycle and efficiency. To an extend this con-
using IRT images can reveal the presence of any thermal anoma- dition can potentially lead to fire ignition especially for old build-
lies in electrical equipments, produced by the defect on the sur- ings with outdated electrical wiring that is deteriorating,
face of the equipment. The defect will normally alter the inappropriately amended, or insufficient to support the existing
thermal signature of the surface due to the change in the amount electrical loads. However, new buildings or constructions also are
of heat generated and the heat transfer properties of the equip- not immune from this condition. Over the time, the electrical
ment [7]. installations and their contact surfaces will begin to deteriorate,
Although the inspection technique of electrical equipment is resulting in an increased resistance.
quite straightforward, there are several things that need to be con- Faults in electrical power systems can be classified into a few
sidered. Many factors, such as environmental effects and equip- categories, such as poor electrical connections, short or open cir-
ment’s condition will normally affect the results of analysis, cuits, overloads, loads imbalance and improper equipments instal-
especially during an outdoor inspection such as in power substa- lation [1,3,5,8–10]. In most cases, poor electrical connections are
tions. Direct inspection without considering these factors will def- among the common problems in transmission and distribution
initely lead to inaccurate measurements and incorrect decision lines of electrical power systems [4]. According to an IRT survey
making. A good electrical system thermographer must contend conducted during the period of 1999–2005 [11], it was found that
with several problems related to the IRT instrument, the electrical almost halves of the thermal problems were found in conductor
equipments to be inspected and the skill for data interpretation. connection accessories and bolted connections. The problems are
Due to these reasons, recent researches have shown an interest mainly caused by loose connections, corrosion, rust, and non-ade-
on more advanced inspection techniques by implementing an quate use of inhibitory grease. This kind of problem can be recog-
automated diagnosing system for assessing the equipment’s condi- nized by inspecting the heat pattern via IRT camera where the
tion. This will involved various fields of studies especially image highest temperature point indicates the location of the problem.
processing and artificial intelligence. Fig. 1 shows an example of oxidized connection of a miniature cir-
The rest of this paper discusses the recent progress of IRT appli- cuit breaker with its corresponding infrared image. The oxidation
cation for diagnosing and analyzing the reliability of electrical cable connection has caused a hot spot temperature exceeding to
power equipment. Various types of faults that are commonly occur 76.6 °C during the time of inspection and has resulted a burning
in electrical power equipments and components are given in Sec- sign. This condition requires an immediate attention and repair.
tion 2. This section also discuses about the measurement tech- It is recommended that, if oxidation and arcing damage cannot
niques as well as the methods for improving the quality of be repaired, parts and cable should be replaced.
inspection. Section 3 focuses on the recent development on auto- Ideally, electrical systems are designed and sized to meet the
mated system for detecting and evaluating the thermal fault in load demands for a facility. However, over the time and due to
electrical equipments. Some limitations of the current researches building expansion, additional loads have caused the electrical
and recommendations for future development are highlighted in system equipments reaching beyond its original design capacity.
the Section 4. Concluding remarks appears in Section 5. Improper and under sizing of electrical equipments and compo-
nents can also cause an overload conditions. Overload condition
usually shows a uniformed heat pattern appears through the entire
2. Application of IRT for diagnosing electrical equipments circuit. Total heat generated from the equipments depends on the
load and the ambient operating temperature of the equipment.
This section focuses on the application of IRT for diagnosing the Fig. 2 depicts a single phase circuit breaker with an overload
reliability of electrical equipments and components. The typical condition. The most left circuit breaker has risen in temperature
examples of thermal faults that commonly occur in electrical for more than 67 °C while the rest of the breakers operate below
equipments are briefly explained. The other part of this section will 40 °C.
look into methods of improving the quality of inspections and the In three phase electrical system, a perfect system should have
techniques used to evaluate the severity level of the thermal similar operating condition on each phase or at least not vary
anomaly.

Fig. 1. Oxidized connection of a miniature circuit breaker with its corresponding infrared image.
238 M.S. Jadin, S. Taib / Infrared Physics & Technology 55 (2012) 236–245

greatly load differences between phases. IRT camera can clearly


shows the condition where the higher loads phase will be warmer
than other phases. As illustrated in Fig. 3, the whole B-phase circuit
shows an elevated heating compared to R-phase and Y-phase.
Other problems that can contribute to thermal anomalies in
electrical equipments are open circuits, harmonics, inductive heat-
ing and defective equipments. Under the normal operating condi-
tion, an open circuit conductor shows colder thermal pattern
when compared to adjacent conductor under similar loads. This
is due to lack of current flow through that conductor. A hot spot
in electrical equipments can also caused by harmonic problems
which are normally created by non-linear loads such as solid-state
electronics, electronic lighting ballast, variable frequency drives,
motors and so on. In order to confirm harmonics problem, it is
important to verify load conditions whether it is running within
Fig. 2. Typical heat pattern of an overloaded circuit breaker.
specified ratings and close to balanced. If not, the problem could
be due to overloading or load imbalance condition. Another testing
methods using harmonic analyzer is required in order to confirm
this problem. Inductive heating is resulted from the proximal
interaction of non-carrying electric devices with the magnetic field
around conductors. The heat intensity is depending on the amount
of current passing through the conductor and the affected electri-
cal components can reach temperatures up to several hundred de-
grees. Finally, IRT can also be used to detect the defective electrical
equipments such as power transformer, uninterruptible power
supply (UPS), current transformer, lightning arrester and many
others. The defective electrical equipments will display abnormal
thermal distribution on its surfaces. In inspecting a power trans-
former, for example, the common causes of failure in transformers
are due to oil leakage and inferiority in internal insulation, which
can lead to catastrophic destruction and power outage [1,10,12].
For normal operation of a transformer, operating temperatures rise Fig. 3. Typical heat pattern of load imbalance.
over ambient 65 °C for oil-filled ones and 150 °C for air-cooled
transformers [3]. Temperatures above these operating points will
cause the transformer to fail due to breakdown in the insulation object being inspected, and the IRT camera specifications
of the transformer’s winding, causing an electrical short-circuit. [6,9,15]. Without considering all these factors, results may be mis-
To summarize, IRT is a very useful tool for diagnosing the con- interpreted or incorrect.
dition of electrical equipments. The hot spot in electrical equip- For an outdoor or uncovered inspection, such as in power sub-
ments can be easily located and the problems prior to fatal station, environmental effect is a very critical issue. The data
failure can be identified. Nevertheless, in some cases, IRT images regarding the environmental factors are very important, and
cannot be interpreted directly except by experienced and qualified should be collected prior to inspection. Table 1 summarizes all
thermographers. Some IRT characteristics and many variables that the factors regarding the environmental effects that need to be
might affect the measurement should be well understood. considered when doing an IRT inspection [3,6,9,14–16].
Furthermore, there are other factors that can affect the thermal
2.2. Improving inspection techniques measurement. A prior study should be made before starting any IRT
inspection. In this particular case, most thermographers need to get
Newer developments in modern IRT equipment have improved some information about the target location. Sometimes, the history
the quality of measurements. Most modern IR imagers can resolve of the target location and electrical power equipments also needs to
surface temperature differences of 0.1 °C or less [6] [13]. Despite be taken into consideration. The load variations, type of equip-
the advantage of modern designs of IRT cameras associated with ments, as well as the materials used in building the equipment
advanced image processing and display system, there are still sev- are among other important data needed for an inspection. In order
eral factors that need to be considered when doing an inspection to get the best and accurate measurement, the right and suitable
even if the temperatures can be measured accurately [14]. This is tool should be selected. It is recommended that for an extensive
a very critical aspect, especially for an outdoor inspection such as outdoor inspection, especially during sunny periods, long-wave
in substation, underground distribution, and aerial distribution (generally 8–14 lm) sensing of IRT systems will give superior re-
[4]. In this section, all related factors that are directly or indirectly sults. This is because in this wave band, the thermal detectors have
affect the results of an inspection will be discussed. When an accu- greater sensitivity to ambient temperature objects and have good
rate measurement is required, all these factors may first need to be transmission through smoke [3,8]. Short-wave systems should be
identified at the time the IRT image is captured. used only on a limited basis, or if loads and other conditions allow,
Generally, the factors that affect the accuracy of IRT measure- on overcast days or at night [6]. Table 2 summarizes all the factors
ments can be categorized as procedural, technical and environ- related to the target equipment and the inspection tools.
mental/ambient conditions [6,9]. The procedural factor concerns
the thermographer itself. This factor can be minimized if certified 2.3. Method of measurement and analysis
or qualified personals are employed. For technical factors, most
of the information needed concerns the emissivity of the equip- There are two ways to measure the thermal condition of electri-
ment under inspection, load current variation, distance of the cal equipments. The first is known as quantitative, which is to take
M.S. Jadin, S. Taib / Infrared Physics & Technology 55 (2012) 236–245 239

Table 1
Environmental factors to be considered during inspection.

Environmental factor Effect on IRT measurement


Ambient air temperature An increase in air temperature will result in an increase in the measured temperature of equipment. At very high or very low-
temperatures, the IR system becomes less stable
Precipitation/humidity (snow, The temperature can be dramatically reduced, leading to a misinterpretation of the data. The equipments that are only slightly
rain, fog, etc.) warm may be cooled below the abnormal point
Wind or other convection Wind speeds as low as 1–5 mph can have a significant cooling effect on a high resistance fitting. Above 5 mph, it can reduce the
temperature difference between the equipments and ambient to a few degrees above the ambient
Sun or solar radiation Solar heating of equipments, especially those with a high absorption of the sun’s energy (such as aged conductors), will mask small
thermal differences

Table 2
Factors related to the target equipment and the inspection tool with their characteristics.

Equipment and Characteristic


tool
Electrical loads Temperature of the connection will increase as the load increases. When the loads are light, problems in the early stages of failure will be less
thermally obvious. It is recommended that during the inspection, the load on the line should be at least 40% [6,15,17]
Equipment Most of the conductors have quite low emissivities, typically 0.1–0.3. While greasy, black, overheated and aged conductors can have emissivity
emissivity values as high as 0.97, it is often difficult to assess this visually in the field from a distance [15,6]
Thermal gradient The heat of high resistance is usually being generated at some internal point to the surface. There exists a thermal gradient between the hottest
spot inside the equipment and the surface being viewed [16]
IRT device Factors that must be considered are resolution, both spatial and measurement, as well as the detected waveband [14,16]. Sensitivity and signal
(camera) processing speed must also be considered for good performance [6]
Distance and The resolution of the IRT image decreases with distance. Acute angles present less information than images taken at right angles [5]
angle

Table 3
the exact temperature of the equipments. The second type is qual-
Maintenance testing specifications for electrical equipment.
itative, which takes the relative temperature values of a hotspot
with respect to other parts of the equipment under the similar Priority DT between similar DT over Recommended action
components under ambient
conditions [18]. Some applications do not require obtaining exact
similar load (°C) temperature
surface temperatures. It is sufficient to acquire thermal signatures, (°C)
which are characteristic patterns of relative temperatures of the
4 1–3 1–10 Possible deficiency,
equipments. This method of qualitative visual inspection is expedi- warrants investigation
ent for collecting a large number of detailed data and conveying 3 4–15 11–20 Indicates probable
them in a fashion that can be easily interpreted. In contrast, deficiency; repair as
accurate quantitative thermography demands a more rigorous time permits
2 – 21–40 Monitor until
procedure to extract valid temperature maps from raw thermal
corrective measures
images [13]. can be accomplished
Qualitative analysis is a widely used method in electrical equip- 1 >15 >40 Major discrepancy;
ment inspection [19]. The analysis is done by employing the DT cri- repair immediately
teria [20,21]. The severity or the level of overheating of the
electrical equipments will be referred to the temperature-rating
table. This table is usually divided into three or four different derived by selecting an area or by locating a point at the compo-
categories to indicate the maintenance priority based on the nents or equipments as illustrated in Fig. 4a. The temperature dif-
equipment’s temperature rise with respect to similar reference ferences between all phases can be observed clearly by using one
equipment or component [21]. There are many standards such as line temperature profile (Fig. 4b). Based on the DT method, the
InterNational Electrical Testing Association (NETA) [22], American temperature rises between the suspected phase and reference is
Society for Testing & Materials (ASTM) – E 1934 [23] and National calculated as:
Fire Protection Association (NFPA) – NFPA 70-B [24] provide a
guidelines for IRT inspection. Standards are employed to provide DT ¼ T 3  T 1 ð1Þ
a simple, consistent and accurate method of evaluating the severity
of the problems located [14]. In assessing the level of severity, where T3 is the maximum temperature value of Area 3 or at Point 3
NETA standard is widely used for acceptance testing specifications while T1 is the maximum temperature value of Area 1 or Point
for electrical equipment by using the delta T criteria as shown in 1which is assigned as the reference temperature. The severity level
Table 3. These deltas T are based on temperature rise above a de- of the temperature elevated is then checked using Table 3 under the
fined reference, which is typically the ambient air temperature, a column ‘DT between similar equipments and under similar load
similar component under the same conditions or the maximum (°C)’. Maintenance action should be taken according to the level of
allowable temperature of the component [22]. Some thermogra- priority.
phers prefer to make their own testing specification table based The advantage of qualitative method is that it is a practical
on their experiences instead of using the available standards. method to establish ‘‘failure’’ or ‘‘no failure’’ and the emissivity
In common practice, the severity level assessment is evaluated has only a minor impact on the result [20]. A drawback is that
based on the maximum temperatures for both qualitative and the temperature tables are usually only found in handbooks and
quantitative measurement. Using any commercial infrared image guidelines; thus there is a lack of a recognized standard [14]. More-
analysis software, the maximum temperature values can be over, the DT criterion does not say anything about whether the
240 M.S. Jadin, S. Taib / Infrared Physics & Technology 55 (2012) 236–245

Fig. 4. (a) Techniques for extracting temperature values and (b) the temperature profile across Line 1.

equipment temperature limits are actually exceeded. Furthermore, Developing an automated diagnosing system for analyzing
using the DT criteria will not expose systematic failures affecting infrared images of electrical equipments usually consist of three
all three phases [21]. Therefore, quantitative measurement is re- major steps as illustrated in Fig. 5. The primary step is finding re-
quired to confirm the level of failure in electrical equipments. gion of interest (ROI) which is more related to image segmentation.
In quantitative measurement, ambient temperature value is The purpose of segmentation is to subdivide an image into its con-
taken as the reference. The observation is established by measuring stituent regions or objects. After finding ROI, the descriptive infor-
the absolute temperature of the equipment. However, it is not easy mation within the regions should be extracted in such a way that it
to determine precisely the true temperatures of the equipment or can be distinctive enough to be classified in the next process.
component. It requires an even greater understanding of the Finally, classification step which regard to decision making process
variables influencing radiometric measurement including object’s to give result whether this image contain a possible thermal fault
emissivity, transmissivity and reflectivity as well as a grasp of its or not.
limitations that have been discussed in previous section. Therefore, The key success of decision making process depends on the cor-
some related data must be collected prior to IRT inspection and rect ROI detection. Incorrect or inaccurate region identification will
adjustments should be made accordingly. Since this measurement affect to information extraction and classification process. There-
forms the basis for subsequent calculations, it is very important fore, finding ROI has taken very important role in this particular
that it is as accurate as possible. Using the same image as depicted application. Based on previous studies, the method used to find
in Fig. 4a, to assess the suspected phase which is located in Area 3, the ROI and the extraction of features is dependent on the method
the maximum temperature rises relative to ambient temperature of analysis, whether qualitative or quantitative. For quantitative
is given as: analysis, global region segmentation was normally used to identify
the warm regions. The most popular and the simplest methods are
DT ¼ T 3  T amb ð2Þ by using thresholding technique. The possible thermal anomalies
in electrical equipments can be detected by filtering the image
where T3 and Tamb are the maximum temperature value of Area 3 or
using a certain threshold value. If the original image is I(x, y), the
at Point 3 and the ambient temperature respectively. Again, the col-
threshold image, G(x, y) is defined as
umn ‘DT over ambient temperature’ in Table 3 is used for testing
specifications. It is also important to check the maximum allowable 
1 if Iðx; yÞ > T
temperature rating for electrical equipments. Gðx; yÞ ¼ ð3Þ
0 if Iðx; yÞ 6 T

3. Automated diagnostic system where T is the threshold value. The segmentation result is in the
binary image form where 1 represents the target object or region
Most of IRT cameras today come with their analysis software and 0 represent for background image.
and have the capability to prepare the inspection report including The hotspot region can be extracted using morphological seg-
the commercial stand alone software. However, despite multi func- mentation where the maximum gray pixel value determines the
tions and ease to use of the software, the evaluation process are maximum temperature of the hotspot region [20,37–39]. In other
very time consuming for a manually analyzing and preparing the work, Moja and Willis used two levels image segmentation for
report even conducted by a qualified or experienced personnel. detecting the warm regions in IRT images [40]. Image thresholding
Furthermore, for a maintenance program of a large electrical sys- technique was also implemented for diagnosing electrical appara-
tem where hundreds of equipments need to be inspected, the anal- tus using image comparison method [41,42]. This method used to
ysis becomes more complicated. Therefore, applying an automated compare an infrared image to a single baseline or reference image.
diagnosis system will offer a faster and more accurate of decision A shortcoming of this technique is that the statistical variation in
making. Rapid development in computer vision based on image thermal patterns due to changing equipment loading, environmen-
processing techniques and the integration of artificial intelligence tal conditions, varying mechanisms of equipment deterioration
has provided advantages in monitoring and diagnosing problems and changing measurement position are not catered for. Further-
of the power equipments. Currently, automatic inspections of IRT more, template matching technique requires a precise alignment
images are widely applied in medical imaging [25–32], non-
destructive testing [33,34], defect detection in structures [35,36]
and so on. In electrical power applications, automated diagnosis
of IRT images employing intelligent systems is still in the early
stages. This is due to the complex analysis and various factors that
need to be considered in developing such a system. Fig. 5. General step of an automated diagnosis system.
M.S. Jadin, S. Taib / Infrared Physics & Technology 55 (2012) 236–245 241

of images and is limited to equipment that is really identical to temperature distribution of each detected region is more practical
each other [40]. because the actual severity level of the equipments could be accu-
The problem with simple thresholding method is that the image rately predicted. Another straightforward method of evaluation is
tends to be over segmented. This is due to the nature of an infrared by analyzing the real temperature values for each pixel in the im-
image which is quite different from that of visual light image. The age by extracting directly from its RGB data [52]. This method is
formation of a thermal image is purely based on the heat distribu- quite simple but has a problem with high processing time due to
tion of an object. It brings some difficulties to image segmentation the large feature vectors to be computed by a classifier algorithm.
due to its over-centralized intensity distribution and low intensity For qualitative analysis, an automated diagnosing system can be
contrast [43]. Furthermore, extracting the hot regions within an implemented to detect and evaluate the similar equipments or
infrared image is a very challenging task, especially when the im- components under the same condition. Since the qualitative eval-
age contains a very complex background and low signal-to-noise uation needs to identify the differences between normal and
ratio (SNR) [44]. Another reason of producing the low-quality faulted regions, it brings some difficulties in finding ROIs. This is
images such as blurring effect, low target-to-background contrast due to variation of intensity levels of the image that make it diffi-
and noises in infrared images is due to the limitation of infrared cult to detect the faulted and normal regions simultaneously. As for
camera technology availability [45]. Some examples of IRT image example, hot fuse in Fig. 7 appears brighter intensities as the nor-
segmentation results using various thresholding methods are mal one. Looking at the same image, some parts of the fuse show
shown in Fig. 6 with their corresponding threshold values. Com- little bit dimmer (at the mid of hot fuse) while other parts show
pared to ground truth image, most of the segmented images tend brighter intensities making it more difficult to segment the fuse
to be over segmented. This will lead to inaccurate ROIs detection. as one region. The problems also become more complicated if there
Therefore, another method of thresholding-based segmentation is other heat source outside the equipment.
should be explored. There are various thresholding techniques Using the conventional segmentation based on bottom-up ap-
for image segmentation that have been proposed and published proach, there are two typical segmentation mistakes that can hap-
[46]. pen. The first is the target equipments or components are
Another possible approach for finding ROIs of electrical equip- segmented into multiple regions. The second one is some parts of
ments in an IRT image is to use region based segmentation such the equipments or components are merged with the background
as Watershed Transform [51]. Even watershed segmentation has image. Fig. 8 illustrates this situation by giving examples of bot-
the advantages of noise and illumination robust, in practical appli- tom-up based segmentation results using Normalized Cut segmen-
cation it tends to be over-segmented especially for more complex tation [56] and Edge Detection and Image SegmentatiON (EDISON)
background in IRT image. algorithm [57].
In quantitative analysis, ROIs are normally the warm regions Therefore, a method based on top-down approach for object
which are segmented from an IRT image. It can be in different sizes recognition is one of the best choices. Smedberg [53] and Wretman
and shapes depending on the temperature distribution of the im- [54] have proposed a method based on top-down oriented. The
age and the method used for segmentation. The features are usu- appropriate steps of this method are shown in Fig. 9. By assuming
ally extracted from the geometrical properties of the region such that all objects in an infrared image of electrical equipments are in
as moment invariant [37] [39], Zernik moment [38], and Fourier repetitive structures, the ROIs are identified by detecting the local
transform as the textural feature descriptor [40]. Taking the geo- features and grouping the similar distinctive features of each ob-
metrical properties of the regions as classification parameters, a ject. Scale Invariant Feature Transform (SIFT) algorithm [55] was
proper justification is needed because some images may appear employed to find the repeated structure by identifying its distinc-
hot but in actual it can be considered as normal temperature due tive features, describe the features (Fig. 9b) and matching each fea-
to the dynamic appearance of IRT images. The problem also arises tures in order to detect similar object within the image (Fig. 9c).
when there are other sources of heat outside the target equipment Here, SIFT algorithm was modified in such a way that the distinc-
since in quantitative image analysis; the decision is made for the tive similar features are matched and paired within a single image
whole image. Therefore, it is recommended to also consider the rather than between images. Finally, after having a set of matching
statistical temperature distribution as the region features [41,42]. features, these set of features are grouped and the region of re-
Assessing the electrical equipments by analyzing its statistical peated structure can be extracted (Fig. 9d) [54]. This segmentation

Fig. 6. Example results of infrared image segmentation using thresholding method: (a) original infrared image, (b) original image histogram, (c) ground truth segmentation,
(d) Otsu (T = 79) [47], (e) Hamadani (T = 136) [48], (f) Minimum Error Thresholding (MET) (T = 41) [49], (g) Kapur (T = 138) [50], (h) Fuzzy C-Mean (T = 99).
242 M.S. Jadin, S. Taib / Infrared Physics & Technology 55 (2012) 236–245

for monitoring the thermal anomaly in electrical equipment [59].


Image registration methods are divided into three steps: First,
based on (regional) gray similar to the method of selection of such
methods by a similarity measure, with time as a measure of simi-
larity of two images. The second is image transformation by using
phase correlation based on Fourier transform algorithm. Finally,
they use Harris corner point detection for matching the images.
After detecting the image of the target equipment, the image is
then compared with the similar template in order to define the
apparatus condition.
There are various techniques that can be used to determine the
thermal severity of electrical equipment through qualitative IRT
Fig. 7. Heat distribution of fuses.
image analysis. One of the techniques is a direct interpretation
technique has shown a good ROIs identification and repeated ob- by identifying the real maximum temperature for each of electrical
ject detection. However, this technique is only practical for a very equipment and evaluating their condition based on the DT criteria.
look like similar structure of electrical equipment or component. The maximum temperature is determined by finding the highest
The limited texture in infrared image of electrical equipments pixel value within the selected region. However, this technique
has lead to limited number of features detection. some time can be misleading. This is because, a maximum temper-
In another attempt, Ishino [58] proposed a method for detecting ature of 80 °C which is quite high for electrical equipment if the
the thermal fault in electrical apparatus by comparing the refer- ambient temperature is 30 °C compare to an ambient temperature
ence thermal pattern of an apparatus using high-order local auto- of 60 °C. Therefore, relative temperature different against ambient
correlation features. In this study, similar apparatus are detected temperature has also to be considered. Calculating the histogram
using Manhattan distance between the reference feature and the or histogram distance is another method that can be used for find-
feature at each attention area. The similar is defined as ing the similarity between two objects. In this case, the histogram
for each region is computed and compared with other regions in
P
35
k k order to get the DT. Another recommended approach is to analyze
Similarity ¼ jfeatureinput  featurereference j ð4Þ
k¼1 the gradient of the segmented region. One of the advantages of uti-
k k
lizing the gradient analysis technique is that the source of the hot-
where featureinput is the kth feature of input image, and featurereference spot in electrical equipment can be identified [53]. All of these
is the kth feature of high-order local autocorrelation of the refer- parameters can be used as input features for classification or deci-
ence thermal image. Once the target region is detected, local accu- sion making process.
mulated gradient of brightness value is used to define the condition Either qualitative or quantitative analysis of IRT images, most of
of the electrical apparatus. Local accumulated gradient of brightness the available classification algorithms can be successfully imple-
value is given as mented for classifying the condition of electrical equipments. An
7 artificial neural network (ANN) algorithm that learns complex in-
Agb ¼ f jIn  In1 j ð5Þ put–output relationships of the data has been used for learning
n¼2
and classifying the thermal condition of electrical equipments
where In is the brightness value at the nth point. [40,52,53,60]. Of all ANN algorithms, multi-layer perceptron
Lin et al. implement image registration method based on object- (MLP) model was widely used due to its flexibility, high computa-
oriented division and the image matching adjustment algorithm tional rate and robust. In another attempt, support vector machine

Fig. 8. (a) Original IRT image, (b) result of Normalize cut segmentation [56], (c) result of EDISON segmentation [57].

Fig. 9. Finding ROI using top-down approach: (a) Original image of fuse, (b) SIFT key points and feature descriptors, (c) matching most similar features, (d) final ROIs
detection after grouping the set of features [54].
M.S. Jadin, S. Taib / Infrared Physics & Technology 55 (2012) 236–245 243

(SVM) algorithm can be employed to determine the condition of to use the latest technology for IRT cameras. Besides the resolution,
the electrical equipment [37,38]. SVM widely used tool for data the modern IRT cameras have very high thermal sensitivity and
classification, function approximation, etc., due to its generaliza- within some product the cameras have the capability to adjust
tion ability [61]. SVM is a learning system that uses a hypothesis the data measurements on screen, like object emissivity, tempera-
space of linear functions in a high dimensional feature space to ture, etc.
estimate decision surfaces directly rather than modeling a proba-
bility distribution across training data. Fuzzy-based decision mak-
ing approach was also tested for classifying power apparatus 4.2. High resolution and continuous monitoring
condition. So et al. proposed a fuzzy logic-based expert system
which can extract the major features within the IRT images and The adoption of continuous thermal imaging deliver extra ben-
recommends appropriate actions for maintenance [41]. While efits over periodic thermal inspection, especially relates to electri-
Laurentys Almeida et al. [51] implemented neuro-fuzzy network cal equipment durability. There is an advantage for the continuous
for detecting and classifying fault in surge arrester. The developed thermal monitoring, since faults can occur at any time. In addition,
system achieved 90% of fault classification based on tested datasets it is not operator dependent, nor is it dependent on the periodical
of more than 760 measurements. The only thing that distinguishes inspection, especially during heavy load. In other aspect, the, real-
between the proposed systems from others is that the input vari- time monitoring can alert the signals or alarms if anomalies occur
ables for neuro-fuzzy network can consider the environmental fac- unexpected and therefore appropriate action can be taken simulta-
tors parameters, IRT variables and surge arrester specifications neously. On top of this, integration with the existing Supervisory
such as pollution index, rated voltage, material and manufacturer Control and Data Acquisition (SCADA) systems will enable the
of the equipment. The whole system diagram and the input vari- real-time remote monitoring without the need for separate system
ables are depicted in Fig. 10a and b respectively. or report, which is practically, cannot be achieved with periodic
thermal inspections.

4. Future recommendations
4.3. Accurate methods of fault diagnosis
There are several methods that can be implemented to improve
the quality of IRT inspections for the electrical equipments and IRT can often be use to compare between similar equipments to
components. This includes the technology of IRT equipment as well detect faults, but there may be a need to show the severity of the
as the method of inspection. Since the demand for preventive fault. Fault diagnosis in mechanical and electrical systems often
maintenance and condition monitoring of electrical equipment is depends on the prediction of the hot spot at full load and the dura-
gaining attention and interesting, a robust and rapid analytical tool bility of the equipment at that temperature. New methods in pre-
is required for the inspections. The following items are highly rec- dicting the reliability and service life of the equipments had to be
ommended for improving the quality of IRT inspection and condi- explored because there is a need to predict the maximum temper-
tion monitoring particularly in assessing the reliability of electrical ature when the system operated at full load.
equipment.
4.4. Intelligent diagnostic system
4.1. Advanced technology of infrared camera with the higher detector
resolution Due to the excessive demand for preventive maintenance in
electrical power equipments, there is an immediate requirement
The main factor that mostly affects the quality of inspection is to have more reliable and robust intelligent systems. To date, the
the IRT equipment itself. Advances in manufacturing processes of developed intelligent systems could only be used for specific elec-
infrared detectors have dramatically increased both the yield and trical equipments due to the different features of the equipments.
quality while reducing the production costs. However, the quality Therefore, a smart intelligent system model has to be designed and
of inspection is related to the image resolution. Poor resolution will developed which will also cater for the problem with image
produce bad interpretation of inspection images. Therefore, for quality. For inspections being done outdoors, the captured
more accurate and correct data interpretation, it is recommended image will normally be affected by noises. Therefore, an advanced

Fig. 10. (a) System diagram of intelligent surge arrester diagnosis, (b) input variables for neuro-fuzzy classifier [51].
244 M.S. Jadin, S. Taib / Infrared Physics & Technology 55 (2012) 236–245

image processing techniques and new algorithms had to be ex- in: Proceedings of the International Conference on the Properties and
Applications of Dielectric Materials, 2009, pp. 1055–1058.
plored to overcome that could solve these problems.
[13] B. Griffith, D. Türler, H. Goudey, Infrared thermographic systems: a review of
IR imagers and their use, in: J.P. Hornak (Ed.), The Encyclopedia of Imaging
Science and Technology, John Wiley and Sons, Berkeley, CA, 2001, pp. 1–53.
4.5. Models sustainability
[14] J. Snell, R.W. Spring, The new approach to prioritizing anomalies found during
thermographic electrical inspections, in: Proceedings of the SPIE 5073, 2003,
Models developed should be able to accommodate the complex pp. 222–230.
[15] S. Gasperic, B. Sprajcar, R. Bernard, R. Glavic, The analysis of thermographic
interaction of the various factors and can be sustain for future
reports with emphasis on emissivity coefficient determination of bare
expansion. This inspection result should include the impact of conductors and joints, in: Proceedings of the International Conference on
wind speed, variation of electrical loads, ambient air temperature, Electricity, Distribution, 2007, pp. 1–4.
object emissivity, resistance and so on. [16] J. Snell, J. Renowden, Improving results of thermographic inspections of
electrical transmission and distribution lines, in: Proceedings of the
International Conference on Transmission and Distribution Construction,
Operation and Live-Line, Maintenance, 2000, pp. 135–144.
5. Conclusion [17] Z. Korendo, M. Florkowski, Thermography based diagnostics of power
equipment, Power Engineering Journal 15 (1) (2001) 33–42.
It is evident that early prevention is required to avoid possible [18] T.J. Hurley, Infrared qualitative and quantitative inspections for electric
utilities, in: Proceedings of the SPIE, Thermosense XII: An International
faults due to the thermal impact and at the same time it will Conference on Thermal Sensing and Imaging Diagnostic Applications, vol.
increase the reliability of electrical power equipments. The use of 1313, 1990, pp. 6–24.
IRT provides effective and reliable information, particularly for [19] R.N. Wurzbach, R.G. Hammaker, Role of comparative and qualitative
thermography in predictive maintenance, in: Proceedings of the SPIE 1682,
preventive maintenance programs and on line monitoring. Modifi-
1992, pp. 3–11.
cation and improvements in the algorithm and analysis methods [20] Y.C. Chieh, L. Yao, Automatic diagnostic system of electrical equipment using
need to be considered in order to avoid misinterpretation or inac- infrared thermography, in: Proceedings of the International Conference of Soft
curate analysis of IRT data. Recent trends in IRT inspection of Computing and, Pattern Recognition, 2009, pp. 155–160.
[21] T.M. Lindquist, L. Bertling, R. Eriksson, Estimation of disconnector contact
electrical equipments show that there is a quick demand for an condition for modeling the effect of maintenance and ageing, in: Proceedings
intelligent system. The use of advanced system could improve of the Russia Power Tech., 2005, pp. 1–7.
the quality of inspections. Therefore, a systematic approach must [22] Standard for Infrared Inspection of Electrical Systems & Rotating Equipment,
Infraspection Institute, 2008.
be design for a new IRT inspection and a sustainable intelligent [23] ASTM, ASTM E 1934: Standard Guide for Examining Electrical and Mechanical
system. For a complex image analysis, an advanced and effective Equipment with Infrared Thermography, West Conshohocken, Pennsylvania,
image processing techniques should be implemented. Further ASTM International, 2005.
[24] NFPA, NFPA 70B: Recommended Practice for Electrical Equipment Maintenance,
development could embed the automatic processing capabilities Quincy, Massachusetts, National Fire Protection Association, 2006.
in the form of automatic recognition of the measured objects and [25] T. Tan, C. Quek, G. Ng, E. Ng, A novel cognitive interpretation of breast cancer
their critical parts. thermography with complementary learning fuzzy neural memory structure,
Expert Systems with Applications 33 (3) (2007) 652–666.
[26] G. Schaefer, M. Závisek, T. Nakashima, Thermography based breast cancer
Acknowledgements analysis using statistical features and fuzzy classification, Pattern Recognition
42 (6) (2009) 1133–1137.
[27] C. Quek, W. Irawan, E. Ng, A novel brain-inspired neural cognitive approach to
The authors gratefully acknowledge Universiti Sains Malaysia SARS thermal image analysis, Expert Systems with Applications 37 (4) (2010)
(USM) under Postgraduate Research Grant Scheme (USM-RU- 3040–3054.
[28] E. Ng, A review of thermography as promising non-invasive detection modality
PRGS), FRGS Grant and Universiti Malaysia Pahang (UMP) for their
for breast tumor, International Journal of Thermal Sciences 48 (5) (2009) 849–
financial support and facilities. 859.
[29] P.J. Lisboa, A.F. Taktak, The use of artificial neural networks in decision support
in cancer: a systematic review, Neural Networks 19 (4) (2006) 408–415.
References [30] T.Z. Tan, C. Quek, G.S. Ng, K. Razvi, Ovarian cancer diagnosis with
complementary learning fuzzy neural network, Artificial Intelligence in
[1] Y. Cao, X. Gu, Q. Jin, Infrared technology in the fault diagnosis of substation Medicine 43 (3) (2008) 207–222.
equipment, in: Proceedings of the China International Conference on [31] B. Wiecek, Review on thermal image processing for passive and active
Electricity, Distribution, 2008, pp. 1–6. thermography, in: Proceedings of the Annual International Conference of the
[2] X. Maldague, Applications of infrared thermography in non destructive Engineering in Medicine and Biology, 2006, pp. 686–689.
evaluation, in: P. Rastogi (Ed.), Trends in Optical Nondestructive Testing, [32] N. Selvarasu, S. Vivek, N. Nandhitha, Performance evaluation of image
2000, pp. 591–609 (invited chapter). processing algorithms for automatic detection and quantification of
[3] C.A. Balaras, A.A. Argiriou, Infrared thermography for building diagnostics, abnormality in medical thermograms, in: Proceedings of the International
Energy and Buildings 34 (2) (2002) 171–183. Conference on Computational Intelligence and Multimedia Applications, 2007,
[4] Z. Azmat, D.J. Turner, Infrared thermography and its role in rural utility pp. 388–393.
environment, in: Proceedings of the Rural Electric Power Conference, 2005, pp. [33] X. Maldague, Y. Largouët, J. Couturier, A study of defect depth using neural
B2/1–B2/4. networks in pulsed phase thermography: modelling, noise, experiments,
[5] M.A. Kregg, Benefits of using infrared thermography in utility substations, in: Revue Générale de Thermique 37 (8) (1998) 704–717.
Proceedings of the SPIE 5405, 2004, pp. 249–257. [34] F. Just-Agosto, D. Serrano, B. Shafiq, A. Cecchini, Neural network based
[6] C. Hellier, Handbook of Nondestructive Evaluation, first ed., McGraw-Hill nondestructive evaluation of sandwich composites, Composites Part B
Professional, 2001. Engineering 39 (1) (2008) 217–225.
[7] F. Lizak, M. Kolcun, Improving reliability and decreasing losses of electrical [35] M. Selek, O.S. Sahin, S. Kahramanli, Thermographical investigation of crack
system with infrared thermography, Acta Electrotechnica et Informatica 8 (1) initiation using artificial neural networks, in: Proceedings of the International
(2008) 60–63. Conference on Computer as a Tool, 2007, pp. 270–275.
[8] R. Epperly, G. Heberlein, L. Eads, A tool for reliability and safety: predict and [36] R. Heriansyah, S. Abu-Bakar, Simulating and estimating spalling defect depth
prevent equipment failures with thermography, in: Proceedings of the in passive thermography using neural network techniques, International
Petroleum and Chemical Industry Conference, 1997, pp. 59–68. Journal of Simulation: Systems, Science and Technology 9 (2) (2008) 23–30.
[9] L. dos Santos, E.C. Bortoni, L.E. Souza, G.S. Bastos, M.A.C. Craveiro, Infrared [37] L. Baoshu, Z. Xiaohui, Z. Shutao, N. Wendong, HV power equipment diagnosis
thermography applied for outdoor power substations, in: Proceedings of the based on infrared imaging analyzing, in: Proceedings of the International
SPIE 6939, 2008, pp. 69390R. Conference on Power System Technology, 2006, pp. 1–4.
[10] H. Niancang, The infrared thermography diagnostic technique of high-voltage [38] A. Rahmani, J. Haddadnia, O. Seryasat, Intelligent fault detection of electrical
electrical equipments with internal faults, in: Proceedings of the International equipment in ground substations using thermo vision technique, in:
Conference on, Power System Technology, 1998, pp. 110–115. Proceedings of the International Conference on Mechanical and Electronics,
[11] J. Martínez, R. Lagioia, Experience performing infrared thermography in the Engineering, 2010, pp. V2-150–V2-154.
maintenance of a distribution utility, in: Proceedings of the International [39] K.C. Lin, C.S. Lai, Fault recognition system of electrical components in scrubber
Conference on Electricity, Distribution, 2007, pp. 1–4. using infrared images, in: V. Palade, R.J. Howlett, L. Jain, Knowledge-Based
[12] N. Utami, Y. Tamsir, A. Pharmatrisanti, H. Gumilang, B. Cahyono, R. Siregar, Intelligent Information and Engineering Systems, vol. 2773, Springer-Verlag,
Evaluation condition of transformer based on infrared thermography results, Berlin, Heidelberg, 2003, pp. 1303–1310.
M.S. Jadin, S. Taib / Infrared Physics & Technology 55 (2012) 236–245 245

[40] N.T. Moja, A.J. Willis, Classification of thermally condition-monitored [51] C.A. Laurentys Almeida, A. Braga, S. Nascimento, V. Paiva, H. Martins, R. Torres,
components using statistical and neural network techniques, in: Proceedings W. Caminhas, Intelligent thermographic diagnostic applied to surge arresters:
of SPIE, San Diego, CA, USA, 1997, pp. 573–581. a new approach, IEEE Transactions on Power Delivery 24 (2) (2009) 751–757.
[41] A.T.P. So, W.L. Chan, C.T. Tse, K.K. Lee, Fuzzy logic based automatic diagnosis of [52] M. Shafi’i, N. Hamzah, Internal fault classification using artificial neural
power apparatus by infrared imaging, in: Proceedings of the Second network, in: Proceedings of the International Power Engineering and
International Forum on Applications of Neural Networks to Power Systems, Optimization Conference, 2010, pp. 352–357.
1993, pp. 187–192. [53] M. Smedberg, Thermographic Decision Support – Detecting and Classifying
[42] Wenjun Zhang, Remote malfunction diagnosis system based on infrared Faults in Infrared Images, Master Thesis, Royal Institute of Technology,
thermal imaging and RIA, in: Proceeding of Symposium on Photonics and Stockholm, Sweden, 2006.
Optoelectronic, 2010, pp. 1–5. [54] D. Wretman, Finding Regions of Interest in a Decision Support System for
[43] Chen Cong-ping, Qin Wu, Fang Zi-fan, Zhang Yi, Infrared image transition Analysis of Infrared Images, Master Thesis, Royal Institute of Technology,
region extraction and segmentation based on local definition cluster Stockholm, Sweden, 2006.
complexity, in: Proceeding of International Conference on Computer [55] D.G. Lowe, Distinctive image features from scale-invariant keypoints,
Application and System Modeling, 2010, pp. V3-50–V3-54. International Journal of Computer Vision 60 (2) (2004) 91–110.
[44] Y. Li, X. Mao, An efficient method for target extraction of infrared images, in: [56] Jianbo Shi, J. Malik, Normalized cuts and image segmentation, IEEE
F.L. Wang, H. Deng, Y. Gao, J. Lei, Artificial Intelligence and Computational Transactions on Pattern Analysis and Machine Intelligence 22 (8) (2000)
Intelligence, vol. 6319, Springer-Verlag, Berlin, Heidelberg, 2010, pp. 185–192. 888–905.
[45] S. Fan, S. Yang, P. He, H. Nie, Infrared electric image thresholding using two- [57] D. Comaniciu, P. Meer, Mean shift: a robust approach toward feature space
dimensional fuzzy Renyi entropy, Energy Procedia 12 (2011) 411–419. analysis, IEEE Transactions on Pattern Analysis and Machine Intelligence 24 (5)
[46] M. Sezgin, B. Sankur, Survey over image thresholding techniques and (2002) 603–619.
quantitative performance evaluation, Journal of Electronic Imaging 13 (1) [58] R. Ishino, Detection of a faulty power distribution apparatus by using thermal
(2004) 146–165. images, in: Proceedings of IEEE Power Engineering Society Winter Meeting,
[47] N. Otsu, A threshold selection method from gray-level histograms, IEEE 2002, pp. V2-1332–V2-1337.
Transactions on Systems, Man and Cybernetics 9 (1) (1979) 62–66. [59] L. Lin, D. Wu, J. Liu, X. Zhang, A substation infrared temperature monitoring
[48] N.A. Hamadani, Automatic Target Cueing in IR Imagery, Master Thesis, Air and warning system with object separation and image registration, in:
Force Inst. of Tech, Wright-Patterson AFB, Ohio, 1981. Proceedings of the International Conference on Image Processing and Pattern
[49] J. Kittler, J. Illingworth, Minimum error thresholding, Pattern Recognition 19 Recognition in, Industrial Engineering, 2010, pp. 78202B.
(1) (1986) 41–47. [60] S. Haykin, Neural Networks: A Comprehensive Foundation, second ed.,
[50] J.N. Kapur, P.K. Sahoo, A.K.C. Wong, A new method for gray-level picture Prentice Hall, New Jersey, 1998.
thresholding using the entropy of the histogram, Computer Vision, Graphics, [61] V.N. Vapnik, The Nature of Statistical Learning Theory, Springer-Verlag, New
and Image Processing 29 (3) (1985) 273–285. York, 1995.

Vous aimerez peut-être aussi