The accurate segmentation and detection of defects in infrared and visible images are critical for non-destructive testing applications, however those steps are often excluded by limited annotated training data. This paper presents an innovative approach for the segmentation and detection tasks into a unified framework. The proposed method introduces and tests a novel framework tailored to the domain of infrared and visible imaging. This framework eliminates the need for annotated defect data during training, enabling models to adapt to real-world scenarios where annotations are scarce. To enhance the accuracy of segmentation and detection, it employs super-pixel segmentation, following by texture analysis.
The identification and categorization of subsurface damages in thermal images of concrete structures remain an ongoing challenge that demands expert knowledge. Consequently, creating a substantial number of annotated samples for training deep neural networks poses a significant issue. Artificial intelligence (AI) models particularly encounter the problem of false positives arising from thermal patterns on concrete surfaces that do not correspond to subsurface damages. Such false detections would be easily identifiable in visible images, underscoring the advantage of possessing additional information about the sample surface through visible imaging. In light of these challenges, this study proposes an approach that employs a few-shot learning method known as the Siamese Neural Network (SNN), to frame the problem of subsurface delamination detection in concrete structures as a multi-modal similarity region comparison problem. The proposed procedure is evaluated using a dataset comprising 500 registered pairs of infrared and visible images captured in various infrastructure scenarios. Our findings indicate that leveraging prior knowledge regarding the similarity between visible and thermal data can significantly reduce the rate of false positive detection by AI models in thermal images.
In recent years, many studies have focused on using deep-learning approaches for automatic defect detection in the thermographic inspection of industrial and construction components. Deep Convolutional Neural Networks have proven to perform remarkably on thermal defect detection. However, their convergence and accuracy are heavily associated with having a large amount of training data to avoid overfitting and ensure reliable detection. Unfortunately, the number of available labeled thermal datasets for inspection-related applications is very limited. One of the practical approaches to address this issue is data augmentation. This paper proposes a novel approach for augmenting simulated thermal defects on regions of interest using coupled thermal and visible images. The visible images are employed to extract regions of interest in both modalities using a texture segmentation method. Later, the introduced method is used to augment thermal defects on thermal images.
KEYWORDS: Inspection, Thermography, Image processing, Image registration, Signal to noise ratio, Signal processing, Data acquisition, Infrared radiation, Civil engineering
This study developed an end-to-end procedure to overcome common issues faced during the analysis of passive infrared thermography (IRT) sequences from outdoor concrete infrastructures. The processing pipeline includes the automatic pre-processing of raw thermograms, data cleaning and organization, image adjustment, and sequential image registration. One image registration method was implemented, and the results were evaluated using the Euclidean distance metric. Furthermore, the resulting sequences were processed using signal processing techniques to increase the detectability of the defects. The results from outdoor IRT surveys over two academic samples are presented, where one image per minute was taken for 24 hours on slabs and columns representative structures. By addressing the difficulties encountered during the analysis of passive IRT sequences, our contribution can broaden the spectrum of the application of IRT for the condition assessment of concrete infrastructure.
Infrared thermography (IRT) technology has evolved during the last decade extending its capabilities to the industrial and infrastructures level. Because of the necessity to perform regular inspections of in-service assets such as bridges, it becomes necessary to investigate and develop efficient inspection technologies that can adapt to the needs of the industry. So, IRT is considered an effective technology to perform NDE. However, its integration with other sensing technologies such as visible cameras still needs to be further investigated so the inspection and maintenance strategy can be more effective when inspecting large structures and assets. Hence, this project investigates fusion strategies and proposes a multi-modal processing pipeline using a deep learning-based panorama stitching method for infrared and visible images. Then, an image registration method to fuse infrared and visible images so identification of defects in visible and thermal spectra becomes more efficient.
KEYWORDS: Signal to noise ratio, Inspection, Thermography, Signal processing, Data acquisition, Sun, Solar energy, Solar radiation, Interference (communication), Nondestructive evaluation
Infrared Thermography (IRT) is a Nondestructive Testing (NDT) method that can complement the concrete infrastructure condition assessment in a fast and contactless manner. When applied to large structures in outdoor areas, the heat source is usually the Sun, which is dynamic and varies through the days, months, and year. Solar irradiation is vulnerable to changes in environmental conditions, which affects the upcoming IRT measurements. Besides that, vertical elements have multiple locations and orientations, where the solar exposure varies according to the solar cycle. Consequently, column faces can experience reduced energy flow, where low or inexistent thermal contrast restrains the detection of existing subsurface damages. In this study, three signal processing techniques, named Principal Component Thermography (PCT), Pulsed Phase Thermography (PPT), and Partial Least Square Thermography (PLST), were applied to thermograms sequences acquired from a concrete element under varying solar exposure. One reinforced concrete column was constructed with ten simulated subsurface defects positioned in the Northeast, Southeast, Northwest, and Southwest faces. This column was inspected hourly through different days of summer and winter periods. It was demonstrated the difference between the signature contrast registered in thermograms acquired from faces exposed to small and large periods of solar irradiation. The preliminary results of using thermographic signal processing techniques verified the possibility of increasing the signal-to-noise ratio and thermal contrast in elements under unfavorable solar exposure. In addition, the research explored the use of different image sequence intervals on the performance of the signal processing techniques.
Trapped humidity is a deleterious condition in several industrial sectors. Beyond reducing performance in process industry, it can create or accelerate structural damage mechanisms like corrosion, where thermally insulated equipment are the main concerning structures. Underneath the insulation, the corrosion evolves invisible at uncontrolled rates until leakage or more catastrophic failure occurs.
The aim of this work is then to establish a passive thermographic technique to be deployed in such environments that can reliably detect trapped humidity. The Multivariate Thermography, based on the Partial Least-Squares regression, can efficiently separate indications from different physical phenomena and drastically reduce the effect of surface finish on the detectability. Preliminary results are encouraging to potentially extend the applicability of thermography to considerably low levels of surface emissivity and reduce the incidence of false positives in thermographic inspection.
The conventional methods for inspection of industrial sites involve the revision of data by an experienced inspector during the acquisition process to avoid possible data missing and misinterpretation. Despite all the advantages of drone-based inspection, inspectors often do not easily have physical access to the site to check for any data ambiguity. Therefore, it is essential for autonomous or semi-autonomous systems to check for missing data or to highlight possible data ambiguity. Reflection in thermal imagery data is one of the main sources of misinterpretation, and it can be problematic when there is no physical access to the site for a secondary inspection. In this paper, we present a novel algorithm based on the analysis and stitching of consecutive aerial thermal images to detect areas with reflection effect and possibly reduce these effects. The conducted experiments have shown significant results in the detection of reflection in drone-based thermographic inspections.
This paper presents an inspection methodology for high-temperature furnace tubes by IR thermography based on the acquisition and analysis over the time of a sequence of thermographic images. With this aim, a set of IR data has been collected during a furnace inspection (operated in steady-state condition) using a high-speed IR camera manufactured by TELOPS (3.0 - 5.4 μm with filter BBP-3670-4020 nm, 320×256 pixels, 3100 Hz). The stacks of IR images have been processed using multivariate statistical analysis – more specifically, partial least squares regression (PLSR), which decomposes the thermographic data sequence into a set of latent variables. Since each latent variable is orthogonal to each other and is characterized by its variance, it is possible to separate the noise affecting the IR signatures through a careful analysis of each component. A qualitative comparison between the processed and non-processed images will be made in order to evaluate the effectiveness of the proposed inspection method.
Infrared Thermography (IRT) is a well-known Non-destructive Testing (NDT) technique. In the last decades, it has been widely applied in several fields including inspection of composite materials (CM), specially the fiber-reinforced polymer matrix ones. Consequently, it is important to develop and improve efficient NDT techniques to inspect and assess the quality of CM parts in order to warranty airworthiness and, at the same time, reduce costs of airline companies. In this paper, active IRT is used to inspect carbon fiber-reinforced polymer (CFRP) at laminate with artificial inserts (built-in sample) placed on different layers prior to the manufacture. Two optical active IRT are used. The first is pulsed thermography (PT) which is the most widely utilized IRT technique. The second is a line-scan thermography (LST) technique: a dynamic technique, which can be employed for the inspection of materials by heating a component, line-by-line, while acquiring a series of thermograms with an infrared camera. It is especially suitable for inspection of large parts as well as complex shaped parts. A computational model developed using COMSOL Multiphysics® was used in order to simulate the inspections. Sequences obtained from PT and LST were processed using principal component thermography (PCT) for comparison. Results showed that it is possible to detect insertions of different sizes at different depths using both PT and LST IRT techniques.
In this work, pulsed phase thermography (PPT), principal component thermography (PCT), and partial least squares thermography (PLST) techniques were applied in order to detect the masonry texture, as well as to map the subsurface damages formed beneath three different mural paintings. The latter were inspected after the 2009 earthquake, i.e., the seismic event that devastated L’Aquila City (Italy) and its surroundings. The mural supports explored by infrared thermography (IRT) are constituted by a single leaf, and the sides of the inspected paintings are confined by marble frames or by buried horizontal and vertical structures. Hence, the analyzed objects can be considered as monolithic structures. IRT can help to understand the masonry morphology, e.g. if there exist structural continuity between the arriccio layer (the first coat of plaster) and the support. In the present case, the heating phase was provided by lamps or propane gas and feature detection was enhanced by advanced signal processing. A comparison among the results is presented. Two of the three objects analyzed, painted by the art masters Serbucci and Avicola, are preserved inside Santa Maria della Croce di Roio Church in Roio Poggio (L’Aquila, Italy); they were executed on two masonries built in different periods. The last one was realized in Montorio al Vomano (Teramo, Italy) on the internal cloister of the Zoccolanti’s Church (undated). The villages are separated by 50 km as the crow flies. Finally, near-infrared reflectography (NIRR) technique was also used to investigate the condition of the painting layer.
This work focuses in the implementation of infrared and optical imaging techniques for the inspection of aeronautics parts. To this aim, a helicopter blade with known defects is inspected with four different techniques: long pulse thermography, pulsed thermography, digital speckle photography (DSP) and holographic interferometry (HI). The first two techniques belongs to the group of infrared imaging techniques, which are based on the analysis of the infrared thermal patterns in order to detect internal anomalies in the material; whilst the last two (DSP and HI) corresponds to the optical imaging techniques which make use of visible light to measure the material response to an applied stress. Both techniques were applied using the active approach, i.e. an external stimulation is applied in order to produce a gradient in either, the thermal and/or displacement field of the material. The results are then compared in order to evaluate the advantages and limitations of each technique.
Infrared thermography is a valuable tool for non-destructive evaluation of antique artworks. Active thermographic
techniques can be applied on-site thanks to their contactless and non-invasive nature. On-site monitoring is a challenging
task. The observed objects are often hard to reach and of unknown thermal and physical properties. Moreover there are
usually hard constraints on the availability of the site, in terms of space and time. For these reasons the acquired data are
typically inhomogeneous and need to be reorganized and post-processed, with dedicated algorithms, to enhance the
analysis.
The frescoes of the San Gottardo Church, located in Asolo, in the North-East of Italy, are showing multiple detachments
due to the ageing process. More than 60 frescoed surfaces have been selected for evaluation via an active thermography
procedure. Each area has been heated with handheld air heaters and a sequence of infrared images of the cooling process
has been recorded.
Several techniques are available for the post-processing of thermographic sequences. In this work standard algorithms,
such as correlated contrast and principal component thermography, are compared with new ones. We propose two new
algorithms, the first is based on sum and filtering, the second is an adaptation of the partial least squares method to
thermography. The obtained results allow to identify and locate the most important detachments on the surfaces.
Pulsed Thermography (PT) is one of the most widely used approaches for the inspection of composites materials, being its main attraction the deployment in transient regime. However, due to the physical phenomena involved during the inspection, the signals acquired by the infrared camera are nearly always affected by external reflections and local emissivity variations. Furthermore, non-uniform heating at the surface and thermal losses at the edges of the material also represent constraints in the detection capability. For this reason, the thermographics signals should be processed in order to improve – qualitatively and quantitatively – the quality of the thermal images. Signal processing constitutes an important step in the chain of thermal image analysis, especially when defects characterization is required. Several of the signals processing techniques employed nowadays are based on the one-dimensional solution of Fourier’s law of heat conduction. This investigation brings into discussion the three-most used techniques based on the 1D Fourier’s law: Thermographic Signal Reconstruction (TSR), Differential Absolute Contrast (DAC) and Pulsed Phase Thermography (PPT), applied on carbon fiber laminated composites. It is of special interest to determine the detection capabilities of each technique, allowing in this way more reliable results when performing an inspection by PT.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.