Shooting distance estimation from a gunshot residue pattern (GSR) is traditionally performed using chemical and optical methods. Infrared thermography offers a non-destructive alternative for visualising GSR patterns on textile substrates. This work proposes a data-driven approach to estimating shooting distance by analysing thermographic images with a convolutional neural network (CNN). Using kurtosis images were selected as a compact, high-contrast representation of thermographic sequences. A large experimental dataset covering a wide range of shooting distances, firearms, and ammunition types was analyzed. Several CNN architectures were evaluated using transfer learning and fine-tuning. A multimodal architecture combining image features with categorical ballistic information was also investigated. A VGG-based model achieved the most reliable performance, while the multimodal model showed overfitting, suggesting potential for larger, more diverse datasets. Model explainability was addressed through error analysis and Grad-CAM visualisations, confirming that the networks focus on physically meaningful regions. The proposed approach demonstrates the potential of combining infrared thermography with deep learning for reliable, robust forensic estimation of shooting distance.
A quantifier of the thermal response to a thermal challenge in the lower extremities can reveal peripheral impairment in patients with type 2 Diabetes Mellitus (DM2), aiding timely diagnosis and treatment. This study introduces the parameter V-ri as a feasible quantifier of metabolic performance in a temperature regulation process of the lower limbs of individuals with DM2. This index is derived from the analysis of an infrared image sequence through the rate of recovery of metabolic heat of the lower limbs after an instantaneous thermal stimulus, termed the slope value index. The same thermal challenge is applied to a group of patients with DM2 and control volunteers, also clinically supervised. Evaluation of V-ri showed reduced recovery performance in patients with DM2 compared to the performance exhibited by the control group. According to ROC analysis, this parameter V-ri yielded sensitivity and specificity above 0.8, reaching 92% and 83%, respectively. An additional result is that distance V-ri shows that its correlation with clinical variables is statistically significant: glucose (p < 0.01) and HbA1c (p < 0.001). This method offers a non-radiative and non-invasive approach for assessing metabolic deterioration in the lower extremities, potentially suitable for unlimited use both in diagnosis and during treatment.
Ensuring railway safety is a paramount concern, particularly in detecting overheated axle bearings, commonly referred to as hot boxes. In an attempt to improve upon existing operational solutions, the integration of fast infrared (IR) thermal vision technology in the wayside with computer vision has provided new insights for automatically detecting hot-bearing boxes. However, this method involves labour-intensive frame-by-frame processing that can delay critical decision-making. Instead, the present research work shifts focus to the processing of panoramic IR thermal images. Our approach begins with the study and development of a pre-processing technique that employs coarse-to-fine optical flow estimation and image stitching to generate high-resolution panoramic views from sequential IR frames. To enable automatic detection and counting of the hot box, two distinct detection approaches are implemented: custom image processing approach and deep-learning based computer vision approach. The performance of both approaches is evaluated across diverse experimental scenarios, involving both freight and passenger trains and using two distinct IR camera models, cooled and uncooled. Preliminary results are encouraging, demonstrating a clear reduction in the time required for hot box detection and counting, thereby enhancing overall train monitoring efficiency.
Thermal deformation caused by uneven heat accumulation in linear motor feed systems is a major contributor to positioning inaccuracies in precision machinery. Conventional models rely on discrete temperature sensors, failing to fully capture temperature field distribution and highly nonlinear thermal-mechanical coupling, resulting in insufficient robustness under complex conditions. This study proposes an infrared thermal image-driven thermal error modelling method, systematically validating its advantages over discrete sensor-based schemes. Infrared thermal images and thermal error data under multiple load conditions were collected to construct a temperature-field-aware dataset. An enhanced ResNet18 architecture was designed, reconstructing the traditional cascaded CBAM into a parallel channel-spatial attention mechanism to synchronously select multi-dimensional features, improving sensitivity to local thermal gradients and model stability. A regression layer was added for continuous error estimation, combined with Dropout to mitigate overfitting. Experiments show the model achieves over 0.95 prediction accuracy (R2), with MAE and RMSE both below 0.5 mu m. Compared with discrete sensor models and baseline CNNs, it reduces RMSE by 18.3% and improves stability by 12.5%, breaking through the bottlenecks of discrete sensors. It provides reliable support for real-time thermal error compensation in high-precision feed systems, with industrial deployment potential.
Accurate online detection and quantitative evaluation of subsurface defects are essential for improving the reliability of wind energy systems. To overcome the limitations of conventional non-destructive testing techniques, which are typically confined to laboratory or manufacturing settings, this study proposes an inverse problem-solving framework that integrates drone-mounted active infrared thermography with Physics-Informed Neural Networks (PINNs). Transient surface temperature data of defect-containing glass fiber reinforced polymer (GFRP) specimens are obtained via an infrared imaging system deployed on an Unmanned Aerial Vehicle (UAV) platform. A PINN-based forward model is established to simulate heat conduction within the material, and Bayesian Optimization (BO) is integrated into the PINN framework (BO-PINN) to construct an inverse solver that simultaneously estimates defect depth and thickness from surface temperature data. Experimental results demonstrate that the proposed model achieves a depth estimation error below 2% and a relative thickness error below 10%, significantly outperforming conventional methods. This work provides a novel predictive framework and technical route for the non-destructive evaluation and service life prediction of GFRP composite structures.
Inductive thermography provides a non-destructive approach for detecting and characterising cracks in metallic components. This study introduces a method to assess crack geometry - depth and inclination angle - by combining inductive thermography with machine learning. Thermographic sequences from inductively heated cracked specimens were processed using various techniques, including the Fourier transform, to generate phase images. A comparative analysis revealed that the fast Fourier transform (FFT) outperformed other methods, achieving the highest contrast-to-noise ratio (CNR) and effectively suppressing non-uniform heating effects. Phase profiles perpendicular to the crack, extracted at its midpoint, were used as input features. Two machine learning models were developed: one trained on simulated phase profiles to predict crack inclination angle, and a second to estimate crack depth based on the known angle and phase data. Validated against simulated datasets, the models demonstrated high accuracy, advancing the quantitative evaluation of crack geometry for structural integrity and predictive maintenance applications.
Wicking analysis is a widely used method used to evaluate the wetting properties of fabrics. However, accurate measurement poses various challenges. In particular, in multilayer fabrics, inaccurate results may occur due to the different wetting behaviours of the front and back surfaces. Additionally, the inability to clearly observe the liquid on very light or very dark-coloured fabrics is another factor that increases the likelihood of errors. In this study, the wicking behaviour of a three-layer laminated fabric, commonly employed in automotive seat upholstery, was investigated, and a new measurement technique was developed. The method was validated by comparing thermography-derived wicking distance and wetted-area values with gravimetric mass uptake measurements obtained using conventional procedures. A strong correlation was observed between thermal and mass-based data (R = 0.98), confirming the accuracy of the proposed method. Thermal imaging also enabled reliable detection of wetting in internal layers and provided consistent results independent of fabric colour or environmental lighting conditions. Overall, the technique provides an objective and reproducible alternative for evaluating complex multilayer fabrics.
Periodic structures that may exist within the neutron imaging detector can introduce periodic noise into the imaging results, directly degrading image quality and further affecting the performance of deconvolution. This periodic noise appears as four-pointed star-shaped peaks in the amplitude spectrum of the frequency domain. However, the distribution of honeycomb-like noise structures in neutron imaging results makes it difficult to detect using conventional thresholding methods. We propose a method that applies a dilation operation before threshold detection to enhance the contrast between peaks and the surrounding areas. Then, a notch filter is used to smooth the peaks containing noise information, thereby removing the periodic noise structure. This approach effectively eliminates honeycomb structures of approximately 40 micrometers and improves the image quality after deconvolution processing.
To investigate the in situ irradiation effects of gallium nitride at varying temperatures, we combined ion beam-induced luminescence spectroscopy with variable-temperature irradiation using a home-built IBIL system and a GIC4117 2 × 1.7 MV tandem accelerator. Unlike previous static studies—limited to post-irradiation or single-temperature luminescence—we in situ tracked dynamic luminescence changes throughout irradiation, directly capturing the real-time responses of luminescent centers to coupled temperature-dose variations—a rare capability in prior work. To clarify how irradiation and temperature affect the luminescent centers of GaN, we integrated density functional theory (DFT) calculations with literature analysis, then resolved the yellow luminescence band into three emission centers via Gaussian deconvolution: 1.78 eV associated with C/O impurities, 1.94 eV linked to VGa, and 2.2 eV corresponding to CN defects. Using a single-exponential decay model, we further quantified the temperature- and dose-dependent decay rates of these centers under dual-variable temperature and dose conditions. Experimental results show that low-temperature irradiation such as at 100 K suppresses the migration and recombination of VGa/CN point defects, significantly enhancing the radiation tolerance of the 1.94 eV and 2.2 eV emission centers; meanwhile, it reduces non-radiative recombination center density, stabilizing free excitons and donor-bound excitons, thereby improving near-band-edge emission center resistance. Notably, the 1.94 eV emission center linked to gallium vacancies exhibits superior cryogenic radiation tolerance due to slower defect migration and more stable free exciton/donor-bound exciton states. Collectively, these findings reveal a synergistic regulation mechanism of temperature and radiation fluence on defect stability, addressing a key gap in static studies, providing a basis for understanding degradation mechanisms of gallium nitride-based devices under actual operating conditions (coexisting temperature fluctuations and continuous radiation), and offering theoretical/experimental support for optimizing radiation-hardened gallium nitride devices for extreme environments such as space or nuclear applications.
This paper presents a thermal management solution for a Ka-band gyrotron traveling wave tube (gyro-TWT) with non-superconducting magnets. At present, the miniaturization and non-superconductivity of gyro-TWT have become a trend, but miniaturization leads to a significant increase in power density and a severe limitation in heat sink volume, which critically limits power capacity. To address this challenge, a joint microwave–thermal management evaluation model is used to investigate the heat transfer process and identify the crucial factors constraining the power capacity. A cylindrical heat sink with narrow rectangular grooves is introduced. Based on this, the cooling efficiency has been enhanced through structural optimization. The beam–wave interaction, electrothermal conversion, and heat conduction processes of the interaction circuit are analyzed. The compact heat sink achieves a 1.2-fold increase in coolant utilization and reduces the overall volume by 27.4%. Meanwhile, this heat sink improves the cooling performance and power capability of the gyro-TWT effectively. At 29 GHz, the gyro-TWT achieves a pulse power of 150 kW. Simulation results show that the maximum temperature is 348 °C at a 45% duty cycle, reduced by 159 °C. The power capacity of the Ka-band gyro-TWT increases by 40.6%.
Infrared thermography enables the contactless visualisation of laminar-turbulent transition on airfoils. Although the technique has already been used in many wind tunnel and field experiments, for example to study the flow behaviour around the rotor blades of helicopters and wind turbines, a description of the fundamentally achievable minimal measurement uncertainty concerning the position of the flow transition is pending. To this end, the Cram & eacute;r-Rao bound for an approximate signal model is analytically derived and numerically verified. The signal model studied is a Gaussian error function superposed by additive white Gaussian noise. In addition, the effect of image pixelation on the Cram & eacute;r-Rao bound is quantified by a numerical analysis, and the critical image resolution is determined for which the uncertainty limit due to pixelation becomes larger than the uncertainty limit due to the contrast-to-noise ratio. It is proven by Monte Carlo simulations that a classical nonlinear least-squares estimator attains the Cram & eacute;r-Rao bound for a negligible pixelation or a sufficiently high signal-to-noise ratio, respectively. Thus, the minimal achievable measurement uncertainty for the thermographic measurement of laminar-turbulent transition position is clarified, and the validity of the findings is shown for an experiment on wind turbine rotor blades.
Conventional region-of-interest (ROI) methods for analysing facial thermography to assess physiological states often struggle with inter-individual variability. This paper proposes a novel method that mitigates this issue by modifying thermal images based on a standardised facial arterial structure. The method's effectiveness was validated using a long-term dataset of thermal images labelled with self-reported health conditions. Results showed that the proposed method, combined with Gaussian smoothing, extracts features that are highly generalisable across participants. Notably, higher-order statistics of the thermal distribution along the vasculature - specifically skewness and kurtosis - were identified as robust indicators of health state, outperforming conventional approaches. This anatomically informed approach provides a powerful and interpretable alternative to traditional ROI and Black-box deep learning methods for human health monitoring.
Monochromator and analyzer systems that rely on bent single crystals are in use throughout the neutron scattering community. An adequate component for the simulation of such crystals was missing in the widely used neutron simulation software package McStas. The newly developed component Monochromator_bent, which fills this gap, is introduced. It can serve as a model for crystal monochromators and analyzers of various kinds, including the bent perfect crystals, mosaic crystals, and crystals combining mosaicity with bending. The performance of the component is tested at several configurations and compared with the results of another simulation program, SIMRES. Validation is carried out using analytical calculations and the McStas NCrystal_sample component for the case of unbent crystals. Excellent agreement in all tests and good performance in terms of computing speed has been found. The component has been included in the present distribution of McStas 3.5.
At a spallation neutron source, neutron pulses of varying energies are generated, and the detection of neutrons by instrument detectors is recorded as time-of-flight from the emission of the neutron pulse to its arrival at specific detector pixels with high time resolution. The flight path of neutrons from the moderator to the sample and then to the detector must be precisely calibrated at the detector-pixel level using standard powders, so the neutron events from all pixels can be time-focused to produce high-resolution diffraction patterns. Modern time-of-flight neutron diffractometers at spallation neutron sources are equipped with two-dimensional detectors with millimeter-scale pixelations. The number of pixels in a diffraction instrument can reach millions, which makes a single-pixel-level calibration process time-consuming or even impossible with conventional refinement or fitting approaches. Here we present a machine-learning-aided calibration process using a train-and-predict approach, in which machine learning models are trained on the relationship between an individual pixel time-of-flight diffraction pattern and its diffraction constant. These models use a portion of the available pixels for training, and a good model then predicts the diffraction constants precisely and rapidly for large sets of pixel diffraction patterns.
To address the challenge of efficiently detecting surface defects in power battery pole piece electrodes during production, this paper proposes an infrared image processing method based on continuous laser line scanning thermography (CLST). This method integrates multi-column entropy weight fusion and saliency detection methods to enhance the visibility and detection accuracy of micro-defects in infrared images. First, a frame selection resampling strategy is designed to correct the nonlinear geometric distortion in thermal images caused by the mismatch between the specimen's moving speed and the infrared camera's sampling frequency. Second, by extracting multi-column data from the images and performing entropy weight fusion (EWF), the contrast and texture details of local thermal anomaly regions are enhanced. Finally, a saliency enhancement (SE) detection algorithm is introduced to highlight defect regions, improving the separation capability and recognition robustness for micro-defects compared to traditional threshold segmentation methods. The experimental results show that this method can stably detect 0.1 mm scratches, pinholes and shadow marks at a scanning speed of 0.8-1.6 mm/s. For dark spot defects with weak thermal response, the SE method needs to be used for processing.
Early diagnosis of lower extremity injuries in professional football players is crucial for maintaining performance and minimising long-term risks. Despite the growing use of thermographic imaging as a non-invasive tool for detecting musculoskeletal disorders, its integration into automated injury detection systems remains limited, particularly under data-scarce conditions. Given the need for effective early detection methods and the potential of thermography in sports medicine, this study investigates the applicability of deep learning models for classifying lower extremity injuries. Specifically, it evaluates the performance of Prototypical Network and Siamese Network models using thermographic data collected from professional athletes. The original dataset consists of images from 16 healthy and 9 injured individuals, and through augmentation it was expanded to 360 healthy and 180 injured samples. The Prototypical Network achieved an accuracy of 97.78%, while the Siamese Network attained 94%. These findings indicate that both models are capable of accurate injury detection, despite challenges posed by class imbalance and limited data availability. In conclusion, the study highlights the effectiveness of thermographic imaging combined with deep metric learning in identifying injuries in professional football players and suggests that reliable results can be achieved even in constrained data environments.
Active infrared thermography (AIRT) combined with various machine learning methods plays an important role in the detection of internal defects in carbon fiber reinforced polymer (CFRP). However, the extraction of spatio-temporal information has not been thoroughly investigated compared to existing methods. To address this gap, this research proposes a feature extraction method called three-dimensional convolutional autoencoder thermography (3DCAT). The method fuses spatio-temporal information through 3D convolutional operations and combines the encoder's feature extraction capability to effectively capture the potential changes and defect information in composite materials over time. Subsequently, principal component thermography was applied to the encoded data to extract the most representative components.To validate the effectiveness of three-dimensional convolution for feature extraction in thermal image data, this study visualizes and analyzes the features extracted during the training process of 3DCAT network using gradient-weighted class activation mapping. The role of 3D convolution in temporal feature extraction is revealed in depth by analyzing the degree of attention given to the temporal dimension. Ultimately, experimental results obtained from a CFRP specimen with six defects demonstrate the proposed methodology's effectiveness.
A novel scanning multi-directional induction thermography system for non-destructive testing (NDT) is presented. Inhomogeneous Joule heating from Eddy currents circumventing defects is one of the main factors enabling induction thermography as an NDT technique. Eddy current density around defects depends on magnetic field orientation which is the basis of the proposed system to segment defects from sound area on a continuous seamless scan. The method consists of linear scan with a system capable of sequentially inducing Eddy currents on multiple orientations on a continuous basis, enabled by the selective activation of 4 coils wounded on a quadratic ferrite core inductor. The registered scan thermography is temporally splitted for each magnetic field orientation, yielding 4 directional thermographies with spatio-temporal discontinuities. A subsequent temporal pulse synchronization and thermal drift compensation is applied, decomposing the thermogram's oscillating component for each orientation. This enables the usage of conventional spatio-temporal reductions, such as PCA and FFT, for each direction. Additionally, this paper fuses these directional reductions to segment defects based on the variance of the thermal signal associated with defects, compared to sound area. The method is demonstrated on a steel billet with delamination defect, and forged steel bolt with longitudinal surface cracking.
Current radiometric calibration procedures for thermal imaging cameras are typically performed for a single calibration geometry. Typical measurement scenarios do not replicate this geometrical condition and the values provided by the devices deviate from the calibration, which is known as the size-of-source effect (SSE). This study presents SSE measurements on four thermal imaging cameras: two operating in the long-wavelength infrared (LWIR) range and two in the mid-wavelength infrared (MWIR) range. The results indicate that the SSE increases with temperature as the size of the object deviates from the calibration geometry. The deviations observed in the MWIR cameras were smaller than those in the LWIR cameras. The horizontal modulation transfer function (MTF) was measured and utilised to simulate the response of the optical systems and to compare predictions from an analytical approach with the SSE measurements. Although similar trends are observed between the measured and predicted SSE, the model still fails to reproduce the magnitude of the deviations, possibly due to effects not accounted for by the one-dimensional model, along with additional factors such as scattering. Future work will focus on formulating a 2D model and incorporating additional terms into the system transfer function to account for effects not covered by the measured MTF and to enhance the theoretical prediction of the SSE.
Breast reconstruction following mastectomy is increasingly performed, with Deep Inferior Epigastric artery Perforator (DIEP) flap surgery considered the gold standard. Accurate preoperative perforator selection is vital to minimize complications and operative time. While computed tomography angiography (CTA) remains the clinical reference, drawbacks including radiation, contrast use, and cost motivate exploration of non-invasive alternatives. Dynamic Infrared Thermography (DIRT) offers a low-cost, radiation-free method but still lacks automation. This study evaluates deep learning for automated perforator detection in DIRT. A dataset of 50 time-lapse thermograms from five patients was acquired using various cooling methods and validated through leave-one-out cross-validation (LOOCV). Two neural network architectures were compared: a standard U-Net and a modified U-Net (mU-Net) from prior work. U-Net consistently outperformed mU-Net. Across LOOCV folds, U-Net achieved a mean weighted Dice loss of 0.42 +/- 0.09, sensitivity of 0.87 +/- 0.08, and precision of 0.82 +/- 0.14. On an independent test patient, sensitivity remained high (0.88) but precision decreased (0.58). The mU-Net failed to converge (validation loss 0.87 +/- 0.02), producing uniform segmentations. These findings demonstrate that U-Net is a robust tool for automated perforator detection in DIRT, though false positives highlight the need for larger datasets and further optimisation before clinical use.
