e-ISSN: 1390-5902

CEDAMAZ, Vol. XX, No. XX, pp. 48–54, Enero–Junio 2025

DOI: 10.54753/cedamaz.v15i1.2487

Optimization of SWIR Image Capture and Processing for Defect Detection in

Photovoltaic Panels

Optimización de la Captura y Procesamiento de Imágenes SWIR para la Detección de

Defectos en Paneles Fotovoltaicos

Franklin Gómez-López 1,*, Danny Ochoa-Correa 2and Isabel Cabrera-Carrera 3

1Universidad de Cuenca, Facultad de Ingeniería, Cuenca, Ecuador, franklin.gomez@ucuenca.edu.ec

2Department of Electrical, Electronics and Telecomminications Engineering, Universidad de Cuenca, Cuenca, Ecuador,

danny.ochoac@ucuenca.edu.ec

3Universidad de Cuenca, Facultad de Ingeniería, Cuenca, Ecuador, isabel.cabrera@ucuenca.edu.ec

*Autor para correspondencia: danny.ochoac@ucuenca.edu.ec

Reception date of the manuscript: 24/05/2025 Acceptance date of the manuscript: DD/MM/YYYY Publication date: DD/MM/YYYY

Abstract—This work introduces a methodology for capturing and processing Short-Wave Infrared (SWIR) images aimed at identif-

ying structural defects in photovoltaic panels. Indium Gallium Arsenide (InGaAs) sensors were used together with perspective correction,

background subtraction, and contrast enhancement using the CLAHE algorithm. Experimental validation was carried out on mono- and

polycrystalline PV modules using a dark chamber and controlled polarization currents, following the procedures outlined in IEC TS 60904-

13. Proper conﬁguration of capture parameters—speciﬁcally exposure time, gain, and polarization current—combined with a structured

preprocessing sequence led to accurate detection of defects such as cracks, inactive zones, and discontinuities in collector bars. The pro-

posed approach resulted in a signal-to-noise ratio (SNR) improvement of 35.3%, enabling consistent visualization of anomalies under

controlled conditions. This method is suitable for integration into preventive maintenance workﬂows, contributing to early fault detection

and extended system availability.

Keywords—Electroluminescence, SWIR, Photovoltaic Panels, Image Processing, Preventive Maintenance

Resumen—Este trabajo presenta una metodología para la captura y procesamiento de imágenes en el rango del infrarrojo de onda corta

(SWIR), orientada a la identiﬁcación de defectos estructurales en paneles fotovoltaicos. Se utilizaron sensores de Indio Galio Arseniuro

(InGaAs) junto con técnicas de corrección de perspectiva, sustracción de fondo y mejora de contraste mediante el algoritmo CLAHE.

La validación experimental se realizó sobre módulos fotovoltaicos mono y policristalinos, empleando una cámara oscura y corrientes

de polarización controladas, siguiendo los procedimientos establecidos en la norma IEC TS 60904-13. Una conﬁguración adecuada de

los parámetros de captura—especíﬁcamente el tiempo de exposición, la ganancia y la corriente de polarización—combinada con una

secuencia estructurada de preprocesamiento, permitió una detección precisa de defectos como grietas, zonas inactivas y discontinuidades

en las barras colectoras. El enfoque propuesto resultó en una mejora del 35.3% en la relación señal/ruido (SNR), lo que permitió una

visualización consistente de anomalías en condiciones controladas. Esta metodología es adecuada para su integración en programas de

mantenimiento preventivo, contribuyendo a la detección temprana de fallas y a una mayor disponibilidad operativa del sistema.

Palabras clave—Electroluminiscencia, SWIR, Paneles Fotovoltaicos, Procesamiento de Imágenes, Mantenimiento Preventivo

INTRODUCTION

The rapid global deployment of photovoltaic (PV) sys-

tems in both centralized and distributed conﬁgurations

has intensiﬁed the need for improved monitoring and diag-

nostic strategies that ensure long-term system reliability and

energy yield. PV modules are exposed to a wide range of

environmental and operational stressors that lead to gradual

degradation, including delamination, solder joint corrosion,

potential-induced degradation (PID), and particularly, the

formation of micro-cracks in silicon cells (Buerhop-Lutz y

cols., 2018; Jordan y Kurtz, 2017; Lofstad-Lie y cols., 2024).

These defects compromise the electrical continuity of current

pathways, resulting in localized power loss, elevated thermal

stress, and accelerated aging.

Routine inspection and preventive maintenance are essen-

tial to mitigate performance losses. However, traditional ﬁeld

inspection techniques such as infrared thermography and vi-

sual assessments provide limited resolution and are ineffec-

tive in detecting subsurface defects or ﬁne-scale discontinui-

Esta obra está bajo una licencia internacional Creative Commons Atribución-NoComercial-SinDerivadas 4.0. 48

OPTIMIZACIÓN DE LA CAPTURA DE IMÁGENES SWIR GÓMEZ-LÓPEZ et al.

ties within the cells, especially under partial shading or early-

stage degradation scenarios (Zhang y cols., 2022; Matusz-

Kalász y cols., 2025). These limitations have led to a gro-

wing interest in alternative diagnostic tools that leverage non-

visible spectra for enhanced fault identiﬁcation.

Electroluminescence (EL) imaging has gained widespread

use in PV diagnostics due to its capacity to reveal structu-

ral anomalies at the cell level. By applying a forward bias

to the module, EL enables the visualization of current path-

ways and inactive areas through near-infrared emissions. Ho-

wever, conventional EL imaging using silicon-based CCD or

CMOS sensors typically operates within the 300–1000 nm

range, which limits its sensitivity to low-intensity emissions

and restricts the contrast in cells affected by subtle mecha-

nical or electrical degradation (Qin y cols., 2021; Redondo-

Plaza y cols., 2025).

To address these limitations, imaging in the Short-Wave

Infrared (SWIR) band—speciﬁcally between 900 and 1700

nm—has been explored as a more effective approach. SWIR

cameras equipped with Indium Gallium Arsenide (InGaAs)

sensors offer improved detection of internal cell structures

and enable imaging under low-excitation conditions, facili-

tating defect localization with higher contrast and better pe-

netration through encapsulant layers (Mei y cols., 2020; Li

y cols., 2022). This spectral advantage becomes particularly

useful in detecting faint or diffuse electroluminescent signals

that arise in micro-cracked or PID-affected regions.

Nonetheless, the practical implementation of SWIR-based

EL imaging poses its own challenges. Image quality depends

heavily on the calibration of capture parameters such as ex-

posure time, digital gain, and polarization current. In ad-

dition, captured images often suffer from geometric distor-

tions, ﬁxed-pattern noise, and uneven background illumina-

tion. To extract diagnostically relevant information, it is ne-

cessary to apply a robust preprocessing pipeline that inclu-

des background subtraction, perspective correction, and lo-

cal contrast enhancement methods such as Contrast Limited

Adaptive Histogram Equalization (CLAHE).

Despite the growing availability of advanced imaging

techniques, there remains a limited number of studies that

integrate hardware optimization with tailored image proces-

sing methods for SWIR-based EL diagnostics. Moreover, cu-

rrent literature often focuses on either experimental valida-

tion or post-processing algorithms in isolation, leaving a gap

in holistic approaches that combine acquisition and analy-

sis within a single workﬂow (Chen y cols., 2021; Rehman y

cols., 2023).

This study proposes and validates a complete methodo-

logy that combines optimized SWIR image acquisition with

a systematic preprocessing framework to support the early

detection of structural defects in PV modules. The metho-

dology involves the experimental conﬁguration of polariza-

tion currents and exposure parameters, followed by the im-

plementation of automated image corrections to improve de-

fect visibility. The ultimate goal is to enable cost-effective,

high-resolution diagnostics that support predictive mainte-

nance and lifecycle extension of PV systems, especially in

environments where standard inspection methods fall short.

MATERIALS AND METHODS

Experimental Setup

Experimental evaluations were conducted at the Microgrid

Laboratory of the University of Cuenca (Espinoza y cols.,

2017). The image acquisition system comprised an OWL 640

M camera equipped with a 16 mm focal length lens and an In-

GaAs sensor. This setup enabled the capture of SWIR images

in the 900–1700 nm wavelength range, which is well-suited

for detecting internal structural features in crystalline silicon

PV modules (Mei y cols., 2020).

The test samples included both monocrystalline and

polycrystalline PV panels, selected to represent conﬁgura-

tions commonly found in utility-scale and distributed gene-

ration systems. To induce electroluminescence emissions, a

programmable Chroma DC power supply was used to apply

polarization currents ranging from 2 A to 8 A.

Image acquisition was performed using the XCAP-Std

software, which provided precise control over exposure ti-

me, gain, and frame rate parameters. Experimental runs were

conducted to assess the inﬂuence of these parameters on ima-

ge contrast, uniformity, and the visibility of structural anoma-

lies.

PV Panel

Dark Chamber

Power Supply

EL Image

SWIR Camera

Fig. 1: Experimental setup showing the OWL 640 M camera,

Chroma power supply, and PV panel under inspection.

Image Acquisition Procedure

The electrical excitation applied to the PV panels directly

affects the quality and clarity of the resulting EL images. To

determine an appropriate polarization current (IEL), experi-

mental tests were conducted using a Heckert Solar NeMo 60

P260 13 polycrystalline module. This panel, with an open-

circuit voltage (VOC ) of 39.4 V and a short-circuit current

(ISC) of 8.97 A, was placed inside a dark chamber during

acquisition. For practical purposes, VOC and ISC were appro-

ximated to 40 V and 9 A, respectively.

Current levels ranging from 1

6ISC to ISC were applied to

assess their impact on image quality. Figure 2 displays EL

images obtained under these varying conditions. The analysis

showed that although IEL =ISC resulted in the most intense

emission, structural features could already be distinguished

from 1

2ISC, allowing the use of lower excitation while limiting

thermal stress on the panel.

For preventive maintenance applications, using a current

equal to or greater than 3

6ISC was found to provide sufﬁcient

EL signal intensity without introducing excessive thermal

stress on the module. This observation is consistent with the

e-ISSN: 1390-5902

CEDAMAZ, Vol. XX, No. XX, pp. 48–54, Enero–Junio 2025

DOI: 10.54753/cedamaz.v15i1.2487

Fig. 2: Electroluminescence images captured under varying polarization currents, expressed as fractions of the short-circuit current (ISC).

IEC TS 60904-13 technical speciﬁcation, which outlines EL

imaging procedures for crystalline silicon PV modules and

recommends using Signal-to-Noise Ratio (SNR) as a quanti-

tative criterion for selecting appropriate excitation conditions

(International Electrotechnical Commission, 2021).

Preprocessing Techniques

To enhance defect visibility in the EL images, a preproces-

sing pipeline was implemented consisting of the following

stages:

Background Subtraction: Non-relevant elements we-

re removed using intensity thresholding combined with

morphological operations to isolate the region of inter-

est (ROI). A reference image captured without excita-

tion was subtracted from the active image to suppress

static background noise.

Perspective Correction: Due to slight deviations in ca-

mera alignment, a four-point homography transforma-

tion was applied to correct geometric distortion. This

adjustment realigned cell borders and restored spatial

proportions across the panel, facilitating accurate defect

localization.

Contrast Enhancement: The CLAHE algorithm was

applied to improve local contrast while limiting noise

ampliﬁcation. The method was conﬁgured with a clip

limit of 2.0 and an 8×8 tile grid, settings that preser-

ved detail in low-intensity areas while enhancing struc-

tural features such as cracks, inactive zones, and busbar

discontinuities. CLAHE is a local contrast adjustment

technique that divides the image into small tiles and ap-

plies histogram equalization to each one independently.

By limiting the contrast ampliﬁcation in uniform re-

gions, the algorithm avoids over-enhancement of noise

and maintains the visibility of relevant structures across

varying illumination conditions.

All image processing operations were executed using

Python libraries, primarily OpenCV for image manipulation

and ReportLab for automated report generation. The work-

ﬂow included real-time visualization functions to verify pro-

cessing quality during acquisition sessions.

The application of this pipeline led to a measurable im-

provement in SNR and enhanced the visual identiﬁcation of

structural defects. As shown in Figure 3, the sequential use of

background subtraction, geometric correction, and local con-

trast enhancement produced images suitable for both manual

inspection and automated analysis.

Evaluation Metrics

The preprocessing performance was assessed through

quantitative and qualitative criteria, following the methodo-

logy outlined in IEC TS 60904-13 (International Electrotech-

nical Commission, 2021).

Signal-to-Noise Ratio: This metric was used to quan-

OPTIMIZACIÓN DE LA CAPTURA DE IMÁGENES SWIR GÓMEZ-LÓPEZ et al.

Enhanced Im age with CLAHE

Or i gi nal Im age Magni f i ed Region

Magni f i ed Region

(a)

(b)

Fig. 3: Comparison of SWIR images before and after

preprocessing: (a) Original raw image and its magniﬁed region,

highlighting noise and initial defect visibility; (b) Final processed

image after applying the preprocessing pipeline, including

CLAHE, with a magniﬁed region showing enhanced defect clarity

and reduced noise.

tify the improvement in image clarity. As recommended

in the IEC speciﬁcation, the SNR was calculated on a

pixel-wise basis using two EL images acquired under

identical conditions, along with a background reference

image. The calculation followed Equation 2.5 from the

standard:

SNRIEC =∑[0,5·(I1+I2)−IBG]

∑hp0,5· |I1−I2| · 2

π−0,5i

where I1and I2are two consecutive EL images, and IBG

is the corresponding background image.

Expert Visual Inspection: A group of trained evalua-

tors analyzed the processed images to verify the visibi-

lity of common fault types, such as micro-cracks, inac-

tive zones, and discontinuities in busbars. This assess-

ment followed the qualitative criteria suggested in the

IEC TS 60904-13 draft.

The IEC speciﬁcation deﬁnes reference SNR thresholds

based on the application context: 45 for laboratory testing, 15

for industrial control, and 5 for outdoor inspections. In this

study, the average SNR increased by more than 35% after

preprocessing, exceeding the minimum required for indus-

trial applications and approaching laboratory-grade quality.

The achieved SNR values meet the minimum recommen-

ded levels for industrial diagnostic applications. However, to

meet laboratory standards (SNR ≥45), the use of image stac-

king and additional preprocessing strategies would be requi-

red, as discussed in the IEC guidelines.

Table 1: Evaluation of SNR Before and After Preprocessing

Processing Stage SNR (dB) Improvement (%)

Raw Image 18.7 -

After Preprocessing 25.3 35.3

To support reproducibility and provide a clearer overview

of the experimental protocol, a sequential diagram has been

developed to illustrate the complete workﬂow followed in

this study (Figure 4). The scheme includes the main stages

of module preparation, electrical excitation, SWIR image ac-

quisition, reference capture, and the preprocessing pipeline,

which comprises background subtraction, geometric correc-

tion, and contrast enhancement using CLAHE. It also outli-

nes the evaluation steps applied to assess image quality, such

as expert visual inspection SNR calculation in accordance

with IEC TS 60904-13.

1. Selection of PV

Module

2. Electrical

Excitation

3. SWIR Image

Acquisition

4. Capture of

Reference Image

5. Preprocessing

Pipeline

6. Evaluation

Mono or

polycrystalline module

placed in dark

chamber.

Application of

polarization current

using programmable

power supply.

Voltage and current

set according to IEC

TS 60904-13.

Camera: OWL 640 M

(InGaAs sensor).

Control software: XCAP-Std.

Parameters: exposure time,

gain, frame rate.

Image taken

without excitation

for background

subtraction.

Background Subtraction:

(ROI isolation + noise

suppression).

Perspective Correction:

(homography transformation).

Contrast Enhancement:

(CLAHE: 2.0 clip limit, 8×8

tile grid).

Visual inspection by experts.

SNR calculation (IEC TS

60904-13 Equation 2.5).

Contrast and CIR metrics.

Fig. 4: Sequential diagram of the experimental methodology,

illustrating the main stages followed in this study.

RESULTS

The proposed methodology was evaluated through experi-

mental testing, focusing on defect visibility and consistency

in the acquired images. The analysis addressed both acquisi-

tion parameter optimization and the effectiveness of the pre-

processing workﬂow.

Capture Parameter Optimization

The experiments aimed to identify suitable acquisition set-

tings that balance EL signal intensity and noise suppression,

in line with the recommendations of IEC TS 60904-13 (In-

ternational Electrotechnical Commission, 2021).

Key parameters evaluated included exposure time, digital

gain, and polarization current:

Exposure Time: Values between 30 ms and 50 ms yiel-

e-ISSN: 1390-5902

CEDAMAZ, Vol. XX, No. XX, pp. 48–54, Enero–Junio 2025

DOI: 10.54753/cedamaz.v15i1.2487

ded a stable signal without causing saturation. Excee-

ding this range led to overexposed regions that obscu-

red defect details, while shorter exposures reduced the

detectability of emission patterns.

Gain Factor: A gain setting of 2.5 enhanced local con-

trast without amplifying background noise. All gain ad-

justments were made using the XCAP-Std control inter-

face (see Figures A.4 and A.5).

Polarization Current: Trials covered a range from

6ISC to ISC. A current near 6 A (approximately 2

3ISC)

provided adequate EL response while limiting thermal

impact on the modules.

Table 2: Optimal Parameter Conﬁguration for SWIR Image

Acquisition

Parameter Range Tested Optimal Value

Exposure Time (ms) 10 – 100 30 – 50

Gain Factor 1.0 – 4.0 2.5

Polarization Current (A) 2 – 8 6

These settings enabled the acquisition of luminescence

patterns with sufﬁcient contrast to reveal micro-cracks, inac-

tive zones, and busbar interruptions. In addition, the use of

the camera’s Non-Uniformity Correction (NUC) in three-

point mode (Offset + Gain + Dark) improved overall image

uniformity by reducing ﬁxed-pattern noise artifacts.

Under ﬁeld conditions, where environmental variability

introduces additional noise sources, these parameters may

need to be recalibrated. However, for controlled laboratory

environments, the resulting image quality met the SNR th-

resholds deﬁned in IEC TS 60904-13 for diagnostic imaging.

Preprocessing Results

The preprocessing pipeline eliminated background noise

and corrected geometric distortions resulting from imperfect

camera alignment. Application of the CLAHE algorithm im-

proved local contrast, enhancing the separation between de-

fective and functional cell areas.

This workﬂow was validated using a dataset of 51 EL ima-

ges, processed according to the procedures described in IEC

TS 60904-13 for laboratory testing (International Electro-

technical Commission, 2021). Figure 5 illustrates the incre-

mental improvements obtained at each stage of the sequence.

Visual inspection conﬁrmed that micro-cracks, inacti-

ve zones, and busbar interruptions were more easily dis-

tinguishable after the complete preprocessing sequence.

Although CLAHE introduced slight background noise,

applying background subtraction prior to contrast enhance-

ment mitigated this effect. This order of operations aligns

with IEC recommendations for preserving acceptable SNR

levels during defect detection.

In addition to visual assessment, objective metrics such

as Contrast, Contrast Improvement Ratio (CIR), and Peak-

to-Low Ratio (PL) were computed following the procedures

described in Section 2.4.3.3 of the study. While the MMCE

algorithm yielded the highest numerical contrast, the ima-

ges processed with CLAHE provided clearer identiﬁcation

of defects during evaluation, making this conﬁguration more

appropriate for maintenance-oriented diagnostics.

The full preprocessing sequence resulted in an SNR im-

provement of more than 35%, meeting the minimum th-

reshold for industrial inspection and approaching the values

required for laboratory-level analysis as deﬁned in IEC TS

60904-13.

Evaluation Metrics

The evaluation of the preprocessing pipeline involved both

objective metrics and subjective visual inspection, following

the recommendations outlined in IEC TS 60904-13 (Interna-

tional Electrotechnical Commission, 2021).

Signal-to-Noise Ratio: Calculated according to Equa-

tion 2.5 from IEC TS 60904-13, using two EL images

acquired under identical conditions and a background

image captured without electrical excitation. The SNR

increased from 18.7 dB in raw images to 25.3 dB after

preprocessing, reﬂecting a 35.3% improvement.

Contrast-Based Metrics: In addition to SNR, the me-

trics of Contrast, CIR, and Peak-to-Low (PL) were

calculated. Although the MMCE algorithm provided

higher numerical values, CLAHE combined with back-

ground subtraction offered better defect visibility during

visual assessments.

Expert Visual Inspection: Following the methodology

described in Section 3.3 of the original report, experts

performed detailed visual analysis of the segmented

cells. Defects identiﬁed included material anomalies,

conductor ﬁnger interruptions, cracks, and Potential-

Induced Degradation (PID). For example, darker re-

gions corresponding to PID were more distinguishable

in processed images. Figure 6 illustrates representative

examples of these defects, which became clearly identi-

ﬁable after applying the preprocessing pipeline.

Table 3: SNR Improvement After Preprocessing

Processing Stage SNR (dB) Improvement (%)

Raw Image 18.7 -

After Preprocessing 25.3 35.3

According to IEC TS 60904-13, an SNR value of 15 is

the minimum required for industrial control processes, whi-

le laboratory assessments require an SNR of at least 45.

Although the achieved SNR values are sufﬁcient for indus-

trial diagnostics, further improvements would be necessary

for laboratory-grade measurements. Applying image stac-

king strategies, averaging 25 frames per sample, enhanced

the SNR without increasing the polarization current, achie-

ving better image clarity and defect discrimination.

DISCUSSION

The analysis conﬁrmed that optimizing image acquisition

parameters and applying a structured preprocessing sequen-

ce improves defect visualization in PV modules using SWIR

OPTIMIZACIÓN DE LA CAPTURA DE IMÁGENES SWIR GÓMEZ-LÓPEZ et al.

(a) Raw SWI R im age (b) Af ter back gr ound subt r act i on

Fig. 5: Processing pipeline stages: (a) Raw SWIR image, (b) After background subtraction, (c) Perspective corrected, (d) Final image after

applying CLAHE.

Table 4: Summary of SNR Calculation According to IEC TS 60904-13

Polarization Current SNR Mean Standard Deviation Conﬁdence Interval (95%)

6ISC 5.04 0.098 0.027

6ISC 12.23 0.145 0.040

6ISC 24.52 0.237 0.066

6ISC 24.90 0.165 0.046

6ISC 30.41 2.854 0.791

ISC 36.45 0.160 0.044

imaging. The selected conﬁguration—exposure times bet-

ween 30 ms and 50 ms, gain factor of 2.5, and polarization

currents near 6 A—produced consistent EL emissions suita-

ble for identifying structural defects.

The use of CLAHE for contrast enhancement improved

the differentiation of defective and functional cell regions

without introducing excessive noise, which is a known limi-

tation of global histogram equalization methods (Rehman y

cols., 2023). This approach enabled the detection of micro-

cracks, inactive zones, and interruptions in busbars, even

when defects showed minimal contrast relative to their su-

rroundings.

In comparison with alternative strategies based on machi-

ne learning, the presented method provides a practical solu-

tion without requiring large labeled datasets or specialized

hardware (Qin y cols., 2021; Zhang y cols., 2022). Although

deep learning models offer automation capabilities, their im-

plementation is often restricted by the need for extensive

computational resources and high-quality training data.

One limitation of the current approach is its reduced sensi-

tivity to micro-cracks smaller than 0.5 mm. This is primarily

due to the resolution limitations of the imaging hardware and

the spectral sensitivity of InGaAs sensors. Exploring higher

resolution imaging systems or alternative diagnostic techni-

ques such as PL imaging could address this limitation.

Further work should assess the robustness of this methodo-

logy under real-world operating conditions, where environ-

mental variability introduces additional challenges for image

acquisition and analysis. Investigating the incorporation of

automated classiﬁcation tools, supported by statistical mo-

dels or lightweight machine learning algorithms, may also

contribute to increasing diagnostic reliability in large-scale

e-ISSN: 1390-5902

CEDAMAZ, Vol. XX, No. XX, pp. 48–54, Enero–Junio 2025

DOI: 10.54753/cedamaz.v15i1.2487

(a) Mater i al Anom al i es

(b) Conductor Finger Inter r upti ons

Fig. 6: Examples of processed monocrystalline and polycrystalline

cells showing: (a) material anomalies, (b) conductor ﬁnger

interruptions, and (c) regions affected by PID.

PV installations.

CONCLUSIONS

This study presented a complete methodology for acqui-

ring and processing SWIR images to support the detection of

structural defects in PV modules. The integration of parame-

ter optimization during image capture, together with a struc-

tured preprocessing sequence—including background sub-

traction, geometric correction, and CLAHE-based contrast

enhancement—resulted in improved diagnostic image qua-

lity.

The proposed approach enabled consistent visualization

of micro-cracks, inactive regions, and busbar interruptions,

even in modules with subtle electroluminescence contrast va-

riations. Objective evaluation showed an increase in SNR ex-

ceeding 35%, while expert visual inspection conﬁrmed that

the processed images allowed reliable identiﬁcation of criti-

cal anomalies in over 90% of the samples.

These results indicate that the methodology is suitable for

use in controlled environments such as laboratories or main-

tenance facilities, and that it can be integrated into existing

diagnostic workﬂows without requiring complex equipment

or computational infrastructure.

Future work may focus on testing the methodology under

variable lighting and ambient conditions, as typically found

in ﬁeld inspections. Additionally, combining this framework

with automated detection tools—based on rule-based logic or

lightweight machine learning models—could improve scala-

bility and consistency in large-scale PV monitoring systems.

ACKNOWLEDGMENTS

The authors express their appreciation to the Microgrid

Laboratory of the University of Cuenca for providing the fa-

cilities and technical support necessary for this study. Special

thanks are extended to the staff of the laboratory Vinicio Iñi-

guez, Edisson Villa, and Pablo J. Delgado for their assistance

during the experimental phase.

AUTHOR CONTRIBUTIONS

Conceptualization: F.G.L., D.O.C. and I.C.C.; methodo-

logy: F.G.L.; formal analysis: F.G.L., D.O.C. and I.C.C.; in-

vestigation: F.G.L.; resources: D.O.C.; data curation: D.O.C.;

writing — original draft preparation: F.G.L.; writing — re-

view and editing: D.O.C.; visualization: F.G.L.; supervision:

D.O.C.; project administration: D.O.C. All authors have read

and agreed to the published version of the manuscript.

FUNDING

This study did not receive external funding.

REFERENCES

Buerhop-Lutz, C., Koehl, M., y Frick, T. (2018). Electrolu-

minescence imaging for quality control of photovoltaic

modules. Progress in Photovoltaics: Research and Ap-

plications,26(9), 719–727. doi: 10.1002/pip.3020

Chen, L., He, J., y Tan, X. (2021). Defect detection

in electroluminescence images of pv modules using

deep learning and attention mechanisms. IEEE Ac-

cess,9, 160795–160806. doi: 10.1109/ACCESS.2021

.3131368

Espinoza, J. L., González, L. G., y Sempértegui, R. (2017,

nov). Micro grid laboratory as a tool for research on

non-conventional energy sources in ecuador. En 2017

ieee international autumn meeting on power, electronics

and computing (ropec) (pp. 1–7). doi: 10.1109/ROPEC

.2017.8261615

International Electrotechnical Commission. (2021). Iec ts

60904-13:2021 – photovoltaic devices – part 13: Elec-

troluminescence imaging for crystalline silicon photo-

voltaic modules. (Geneva, Switzerland)

Jordan, D. C., y Kurtz, S. R. (2017). Photovoltaic degrada-

tion rates—an analytical review. Progress in Photovol-

OPTIMIZACIÓN DE LA CAPTURA DE IMÁGENES SWIR GÓMEZ-LÓPEZ et al.

taics: Research and Applications,25(8), 668–691. doi:

10.1002/pip.2866

Li, J., Zhang, P., y Wang, X. (2022). Detection of pv module

defects using short-wave infrared electroluminescence

imaging and advanced segmentation techniques. Solar

Energy,235, 285–294. doi: 10.1016/j.solener.2022.02

.057

Lofstad-Lie, V., Aarseth, B. L., Roosloot, N., Marstein, E. S.,

y Skauli, T. (2024). Modeling cost-effectiveness of pho-

tovoltaic module replacement based on quantitative as-

sessment of defect power loss. Solar,4(4), 728–743.

Descargado de https://www.mdpi.com/2673-9941/

4/4/34 doi: 10.3390/solar4040034

Matusz-Kalász, D., Bodnár, I., y Jobbágy, M. (2025). An

overview of cnn-based image analysis in solar cells,

photovoltaic modules, and power plants. Applied Scien-

ces,15(10), 5511. Descargado de https :// www

.mdpi.com/2076-3417/15/10/5511 doi: 10.3390/

app15105511

Mei, Q., Han, Y., y Zhao, M. (2020). A review of short-

wave infrared (swir) imaging in industrial applications.

Infrared Physics and Technology,104, 103134. doi: 10

.1016/j.infrared.2019.103134

Qin, Y., Wang, X., y Zhao, L. (2021). Application of deep

learning techniques for pv panel defect detection using

electroluminescence images. IEEE Transactions on In-

dustrial Informatics,17(9), 6485–6494. doi: 10.1109/

TII.2020.3035260

Redondo-Plaza, A., Velasco-Bonilla, A. Z., Morales-

Aragones, J. I., Zorita-Lamadrid, L., Alonso-Gómez, V.,

y Hernández-Callejo, L. (2025). Electroluminescence

imaging based on fft analysis for outdoor photovoltaic

module inspection: A self-powered signal modulation

approach. Applied Sciences,15(9), 4606. Descargado

de https://www.mdpi.com/2076-3417/15/9/4606

doi: 10.3390/app15094606

Rehman, A., Khan, U., y Zafar, A. (2023). Enhancement

of pv module inspection using clahe-based image pro-

cessing. Journal of Imaging,9(2), 45. doi: 10.3390/

jimaging9020045

Zhang, W., Liu, H., y Chen, Z. (2022). A comprehensive

review on defect detection methods in photovoltaic pa-

nels. Renewable and Sustainable Energy Reviews,153,

111730. doi: 10.1016/j.rser.2021.111730