Image processing-based intelligent defect diagnosis method and device for deformed steel bars

Abstract

The present application relates to the field of defect diagnosis, and more particularly to a threaded steel intelligent defect diagnosis method and device based on image processing, which comprises the following steps: continuously collecting thermal images of the production line to obtain a thermal filter sequence; calculating the temperature distribution between frames in real time for the thermal filter sequence to output the temperature value of each grid cell; analyzing the anisotropic heat conduction path according to the temperature value to obtain the heat conduction path; analyzing the potential thermal defects based on the heat conduction path to mark the potential defect position; tracking the potential defect position across frames in time sequence, and performing comprehensive diagnosis of the threaded steel to generate a defect diagnosis report. The present application improves the overall recognition ability of threaded steel defects, reduces the false positive and false negative rates, and enhances the real-time application performance of the diagnosis results.

Description

Technical Field

[0001] This invention relates to the field of defect diagnosis, and in particular to an intelligent defect diagnosis method and apparatus for rebar based on image processing. Background Technology

[0002] As a key structural material in building construction and infrastructure development, the quality stability and internal integrity of rebar directly affect the safety and service life of projects. While continuous production models such as high-speed continuous casting, rolling, and online cooling have significantly increased rebar output, they also place higher demands on the real-time performance, accuracy, and intelligence of online quality inspection.

[0003] During the rolling and slow cooling process of rebar, defects such as surface cracks, folds, inclusions, porosity, and localized structural abnormalities are easily generated due to factors such as uneven metal flow, roll wear, unstable temperature control, or fluctuations in raw material composition. These defects are often accompanied by localized abnormal heat conduction or uneven temperature distribution, but are difficult to identify directly using visible light detection methods at high temperatures. Furthermore, the vibration, rotation, and speed changes during rebar transportation make it difficult for traditional contact-based detection methods to achieve continuous and stable measurements, and also pose safety hazards. Therefore, how to achieve non-contact, real-time, and comprehensive defect detection under high-temperature and dynamic environments has become a key technical problem that urgently needs to be solved in the field of rebar quality control. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention proposes an intelligent defect diagnosis method and apparatus for rebar based on image processing, thereby resolving at least one of the aforementioned technical problems.

[0005] To achieve the above objectives, the present invention provides an intelligent defect diagnosis method for rebar based on image processing, comprising the following steps: Step S1: Perform continuous frame thermal imaging acquisition on the production line to obtain a thermal filtering sequence; Step S2: Calculate the inter-frame temperature distribution of the thermally filtered sequence in real time and output the temperature value of each grid cell; Step S3: Analyze the anisotropic heat conduction path based on the temperature value to obtain the heat conduction path; Step S4: Analyze potential thermal defects based on the heat conduction path and mark the locations of potential defects; Step S5: Perform cross-frame time-series tracing of potential defect locations, conduct comprehensive diagnosis of rebar, and generate a defect diagnosis report.

[0006] This specification provides an image processing-based intelligent defect diagnosis device for rebar, used to perform the image processing-based intelligent defect diagnosis method for rebar as described above, including: The imaging module is used to perform continuous frame thermal imaging acquisition of the production line to obtain a thermally filtered sequence. The calculation module is used to perform real-time inter-frame temperature distribution calculation on the thermally filtered sequence and output the temperature value of each grid cell. The heat conduction analysis module is used to analyze the anisotropic heat conduction path based on the temperature value to obtain the heat conduction path. The defect analysis module is used to analyze potential thermal defects based on the heat conduction path and mark the locations of potential defects. The diagnostic module is used to perform cross-frame time-series tracing of potential defect locations, conduct comprehensive diagnostics on rebar, and generate a defect diagnostic report.

[0007] The specific benefits of this invention are as follows: By acquiring stable temporal thermal field data through continuous frame thermal imaging, the dynamic temperature changes of rebar during rolling and cooling can be fully reflected. Combined with thermal filtering, industrial environmental noise and infrared interference are effectively suppressed, improving the signal-to-noise ratio and stability of temperature data. Through real-time inter-frame temperature distribution calculation, the thermal image is transformed into a structured grid temperature matrix, achieving precise quantification of temperature gradient and rate of change. This improves spatial resolution and positioning accuracy, transforming defect identification from visual judgment to numerical analysis, enhancing the real-time performance and objectivity of online detection, and providing accurate physical parameters for heat conduction path analysis. By analyzing anisotropic heat conduction paths and introducing the material's thermal conductivity direction differences, abnormal heat flow propagation behavior can be accurately identified. Compared to simple temperature threshold judgment, it can distinguish between normal heat diffusion and heat path deviations or blockages caused by structural anomalies, improving the ability to identify hidden defects such as internal cracks and inclusions, and enhancing the physical interpretability of diagnostic results.

[0008] Analysis of potential defects based on heat conduction path characteristics can effectively eliminate transient interference and surface reflection errors, improving defect identification accuracy. Determining defect type and severity through abnormal heat flow patterns enables defect grading, labeling, and precise location, facilitating early warning of potential quality risks and reducing false alarms and missed detections. Cross-frame time-series tracking verifies defect stability and evolution trends, distinguishing between persistent defects and transient anomalies, thus improving diagnostic reliability. Generating comprehensive diagnostic reports from time-series data allows for automatic output of defect location, level, and development trend, providing data support for production control and quality traceability, and enhancing the intelligence level of production lines. Attached Figure Description

[0009] Figure 1 This is a schematic diagram of the steps of the intelligent defect diagnosis method for rebar based on image processing according to the present invention; Figure 2 This is a detailed flowchart illustrating the implementation steps of step S1. Figure 3 This is a detailed flowchart illustrating the implementation steps of step S2; Figure 4 This is a schematic diagram of the temperature distribution field in step S2; Figure 5 This is a schematic diagram of the microgrid distribution field in step S2; Figure 6 A schematic diagram of the marked thermal imaging of the potential defect location in step S4; Figure 7 This is a schematic diagram of the thermal deviation at different locations obtained from the thermal response deviation calculation in step S4; Figure 8 This is the temperature curve in the vertical direction of the defect location in step S5. Detailed Implementation

[0010] It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0011] This application provides a method and apparatus for intelligent defect diagnosis of rebar based on image processing. The executing entities of the method and apparatus include, but are not limited to, mechanical equipment, data processing platforms, cloud server nodes, and network upload devices that can be considered general computing nodes in this application. The data processing platform includes, but is not limited to, at least one of an audio-visual management system, an information management system, and a cloud-based data management system.

[0012] Please see Figures 1 to 8 This invention provides an intelligent defect diagnosis method for rebar based on image processing, comprising the following steps: Step S1: Perform continuous frame thermal imaging acquisition on the production line to obtain a thermal filtering sequence; Step S2: Calculate the inter-frame temperature distribution of the thermally filtered sequence in real time and output the temperature value of each grid cell; Step S3: Analyze the anisotropic heat conduction path based on the temperature value to obtain the heat conduction path; Step S4: Analyze potential thermal defects based on the heat conduction path and mark the locations of potential defects; Step S5: Perform cross-frame time-series tracing of potential defect locations, conduct comprehensive diagnosis of rebar, and generate a defect diagnosis report.

[0013] In the embodiments of the present invention, see Figure 1 The diagram below illustrates the steps of an image-processing-based intelligent defect diagnosis method for rebar. In this example, the steps of the image-processing-based intelligent defect diagnosis method for rebar include: Step S1: Perform continuous frame thermal imaging acquisition on the production line to obtain a thermal filtering sequence; In this embodiment, an infrared thermal imaging device is deployed along the conveying direction of the rebar to ensure that the field of view covers the full width of the steel surface and that the angle between the viewing angle and the steel axis is controlled within 5° to reduce reflection errors. The acquisition frame rate is typically set to 60–120 fps to adapt to the conveying speed of the production line (0.5–1.5 m / s) and to ensure sufficient spatial overlap (≥85%) between adjacent frames. The thermal imaging band is selected in the 8–14 μm long-wave infrared range, with an initial emissivity value set at 0.82–0.88, dynamically corrected according to the degree of oxidation on the steel surface. The acquired raw thermal images are first subjected to non-uniformity correction, and a radiation response curve is established using a two-point calibration method (e.g., 25 °C and 150 °C blackbody reference sources) to eliminate detector response differences. Subsequently, time-domain thermal noise suppression processing is performed, using a combination of 3-frame moving average and median filtering to remove random thermal noise and transient interference, keeping the standard deviation of temperature fluctuation within ±0.3 °C. To address the potential saturation phenomenon in high-temperature regions, an adaptive dynamic range compression algorithm is used to recalibrate the grayscale mapping, concentrating the effective temperature range within 200–650 °C.

[0014] Step S2: Calculate the inter-frame temperature distribution of the thermally filtered sequence in real time and output the temperature value of each grid cell; In this embodiment, each frame of thermal image in the thermal filtering sequence is converted into a temperature matrix, and a physical size mapping relationship is established based on spatial resolution (e.g., 0.5 mm / pixel). Inter-frame registration is performed, and sub-pixel displacement estimation based on phase correlation is used to control the spatial offset between adjacent frames within 0.1 pixels, ensuring spatial consistency in temperature distribution calculation. Subsequently, adaptive grid partitioning is applied to the temperature matrix. The initial grid size is set to 8×8 pixels, and dynamically refined to 4×4 pixels based on a local temperature gradient threshold (e.g., ≥5 °C / mm) to improve the resolution of defect-sensitive areas. For each grid cell, a weighted average method is used to calculate the representative temperature value. The weights are determined by the pixel signal-to-noise ratio and edge confidence, thereby reducing the impact of edge blurring on the results. To ensure real-time performance, the temperature calculation delay is controlled within 20 ms, and an exponential smoothing algorithm (smoothing coefficient 0.3–0.5) is used to suppress inter-frame fluctuations, making the temperature time series more stable.

[0015] Step S3: Analyze the anisotropic heat conduction path based on the temperature value to obtain the heat conduction path; In this embodiment, the heat conduction trend inside and on the surface of the rebar is identified by analyzing the spatial gradient and directional characteristics of the grid temperature field. The temperature gradient vector field is calculated using the finite difference method, and the temperature change rates in the x and y directions are obtained, constructing the gradient magnitude and direction matrix. When the local gradient magnitude exceeds a set threshold (e.g., ≥8 °C / mm), significant heat flow is determined to exist in that region. Subsequently, a structural tensor analysis method is introduced to extract the principal direction of the gradient, obtaining the dominant direction angle of heat conduction, and calculating the anisotropy ratio (principal direction gradient / vertical direction gradient). When this ratio is greater than 1.5, it indicates that heat conduction has a significant directionality, which may be related to uneven internal structure or microcracks in the material. To enhance path continuity, a heat flow line tracing algorithm based on minimum energy path search is used to connect adjacent high-gradient grids into continuous heat conduction paths, and the path length, curvature, and bifurcation are quantified. Typical parameters include a minimum path length of 10 mm and a curvature threshold of 0.2 mm. -1 wait.

[0016] Step S4: Analyze potential thermal defects based on the heat conduction path and mark the locations of potential defects; In this embodiment, potential thermal defect regions are identified by comprehensively analyzing the continuity, stability, and abnormal characteristics of the heat conduction path. The rate of change of heat flux density along each path is statistically analyzed. When a sudden change in heat flux density exceeding 30% occurs within a 5 mm range along a path, an abnormal thermal resistance is identified, potentially corresponding to internal inclusions or microcracks. Next, the angle of abrupt changes along the path direction is analyzed. If the angle between adjacent path segments is greater than 45°, it is considered a heat flux deflection phenomenon, indicating the presence of a discontinuous thermal conduction structure in that region. Combined with temperature residual analysis, the actual temperature value is compared with the predicted value from the local heat conduction model. Anomalies are marked when the residual exceeds ±12 °C. To reduce false positives, morphological connected component analysis is used to cluster adjacent anomalies. Only regions with an area exceeding 20 mm² and a duration of ≥5 frames are marked as potential defect locations. The marking results are output in the form of spatial coordinates and confidence scores. The confidence score is calculated based on the anomaly amplitude and duration, ranging from 0 to 1.

[0017] Step S5: Perform cross-frame time-series tracing of potential defect locations, conduct comprehensive diagnosis of rebar, and generate a defect diagnosis report.

[0018] In this embodiment, a Kalman filter-based target tracking method is used to predict and update the center coordinates of the defect region across frames, keeping the position tracking error within ±1 mm. The changes in temperature peak, area, and shape factor over time are recorded to form a thermal evolution time series. Subsequently, trend analysis is performed on the time series. The temperature change rate and area expansion rate are calculated using linear regression and exponential fitting. For example, when the temperature increase rate exceeds 2 °C / s or the area expansion rate exceeds 5 mm² / s, the defect is determined to have a development trend. Combined with the defect morphological stability index (shape similarity ≥ 0.85 indicates stability), anomalies are classified into sporadic thermal disturbances or persistent structural hazards. Furthermore, a risk assessment model is constructed based on the temperature anomaly amplitude, duration, and expansion rate, classifying defect risks into low, medium, and high levels. For example, a temperature difference > 25 °C and a duration > 10 seconds is considered high risk.

[0019] In this embodiment, see Figure 2 The diagram below illustrates the detailed implementation steps of step S1. In this embodiment, the detailed implementation steps of step S1 include: Deploy a high-speed infrared thermal imager in the slow cooling zone of the production line, set a high-speed acquisition frame rate, and perform continuous frame thermal imaging acquisition of the production line to obtain a thermal imaging sequence. Calculate the inter-frame displacement vector and exposure time parameters of the thermal imaging sequence; Thermal imaging motion blur analysis is performed based on the inter-frame displacement vector and exposure time parameters to obtain thermal motion blur image features in the direction of motion. Based on the thermal trailing image features, the thermal imaging sequence is deconvolved frame by frame to eliminate thermal trailing, resulting in a de-thermal trailing frame sequence. Thermal radiation filtering is applied to the de-thermal ghosting frame sequence to obtain the thermally filtered sequence.

[0020] In this embodiment, a high-speed infrared thermal imager is deployed to acquire thermal imaging sequences. A mid-wave (3–5 μm) or long-wave (8–14 μm) high-speed infrared thermal imager is installed 1.5–2 m above the slow cooling zone of the rebar production line (where the steel temperature typically decreases gradually from approximately 850 °C to 400 °C), ensuring the field of view covers more than 1.2 times the width of the steel to avoid missing edge detection. Based on the production line operating speed of 2–6 m / s, the thermal imager's acquisition frame rate is set to 300–1000 fps, with typical parameters being 500 fps, exposure time 0.5–1 ms, and resolution 640×512 pixels, ensuring that the displacement of moving targets within a single frame does not exceed 20 pixels, thereby reducing severe blurring. Before acquisition, two-point or three-point calibration is performed using a 600 °C standard blackbody, and the emissivity of the rebar oxide surface is set to ε = 0.82–0.88. Temperature matrix data and timestamps are transmitted in real time via an industrial Ethernet or Camera Link interface to form a continuous thermal imaging sequence T(x,y,t). Simultaneously, the real-time linear velocity of the steel is acquired from the production line encoder for subsequent motion analysis and trailing image modeling. For continuous thermal image sequences, the phase correlation method or the pyramid Lucas-Kanade optical flow method is used to calculate the sub-pixel displacement vector between adjacent frames to characterize the trajectory of the steel in the image plane. Specifically, Fourier transform is performed on adjacent frames and the cross-power spectrum is calculated. After inverse transform, the peak position is obtained to obtain the displacement (Δx, Δy). Under typical parameter conditions, when the steel velocity is 4 m / s, the spatial resolution is 0.5 mm / pixel, and the frame rate is 500 fps, the theoretical inter-frame displacement is approximately 16 pixels. By fusing and correcting the continuous frame statistics with encoder velocity data, the error can be controlled within ±0.5 pixels. At the same time, the exposure time parameter (typically 0.8 ms) is read from the thermal imager synchronization signal and combined with the displacement data to calculate the motion blur length L_blur=v×t_exp, for example, 4 m / s×0.8 ms≈3.2 mm (approximately 6–7 pixels).

[0021] Based on the calculated displacement direction and fuzzy length, a Motion Point Spread Function (PSF) is constructed to describe the thermal trailing shadow formation mechanism. The trailing shadow direction is determined by θ = arctan(Δy / Δx), which is typically close to 0° in the production line, indicating that the trailing shadow is distributed along the rolling direction. Subsequently, a linear or exponential decay kernel model is established based on a fuzzy length of 5–10 pixels to simulate the integral effect of thermal radiation along the motion direction during the exposure time. The energy attenuation characteristics of the trailing shadow direction are analyzed in the frequency domain, and the gradient energy ratio between the motion direction and the vertical direction is calculated using a directional gradient operator. A significant trailing shadow is determined to exist when the gradient energy in the motion direction is more than 30% lower than that in the vertical direction. In areas with large temperature gradients in the thread ribs, the trailing shadow causes a decrease in peak temperature of approximately 8–15 °C and stretches the length of the high-temperature region by 20–30%, thereby masking the true defect boundary.

[0022] Using the aforementioned established motion PSF, Wiener deconvolution or constrained least squares deconvolution is performed on each frame of the thermal image to restore the true temperature distribution and eliminate motion blur. A two-dimensional PSF kernel is generated based on the trail length (e.g., 7 pixels) and direction, and the thermal noise variance (typically ±1.5 °C) in the background region is statistically analyzed to estimate the signal-to-noise ratio. The Wiener filter parameter SNR is set to 20–30 dB to balance the sharpening effect with noise amplification. Subsequently, deconvolution is performed in the frequency domain to restore the high-frequency details of the image, and radiometric consistency correction is performed to ensure that the temperature error is controlled within ±2 °C. Deconvolution processing can reduce the trail width by 60–75%, restore the temperature peak of the thread rib by about 10 °C, and significantly improve the boundary clarity of defect areas such as cracks, folds, and inclusions. To eliminate environmental radiation interference, transient hot spots caused by oxide scale peeling, and sensor random noise, spatiotemporal joint filtering is applied to the deconvolutioned sequence. Spatially, bilateral filtering (σ_s = 2–3 pixels, σ_r = 5–8 °C) is used to smooth temperature fluctuations while preserving edge gradients, reducing random noise by approximately 40%. Temporally, exponential moving average (α = 0.4–0.6) is used to smooth consecutive frames, suppressing short-term thermal disturbances. Simultaneously, an ambient temperature sensor (approximately 120–180 °C in the slow-cooling zone) is used to establish and compensate for background radiation, reducing the impact of reflected heat. Finally, 3×3 median filtering is used to remove isolated high-temperature noise points, avoiding misidentification as point defects. After processing, the standard deviation of temperature fluctuations in the thermal image is reduced from ±4 °C to ±1.2 °C, and the signal-to-noise ratio is improved by approximately 2.5 times.

[0023] In this embodiment, the specific steps for performing thermal radiation filtering on the de-thermal ghosting frame sequence to obtain the thermally filtered sequence are as follows: Multi-frame feature point labeling is performed on the heat-removed ghosting frame sequence to extract image feature points from different frames; Based on the image feature points, the offset between adjacent frames is calculated to obtain the global offset index of adjacent frames; The global offset index is compared based on a preset image deformation coefficient. When the global offset index is greater than the preset image deformation coefficient, it is marked as an abnormal frame. Spatial consistency interpolation is performed on abnormal frames to repair them, and an optimized imaging sequence is output. Thermal radiation filtering is applied to the optimized imaging sequence to obtain a thermally filtered sequence.

[0024] In this embodiment, after thermal trailing removal and thermal radiation filtering, a thermal imaging sequence with clear boundaries and controlled noise is obtained. To analyze inter-frame deformation and anomalous disturbances, representative feature points need to be stably extracted from multiple frames. Temperature gradient calculation and contrast enhancement are performed on each frame's thermal image, and regions with temperature gradients greater than 8–12 °C / mm are selected as candidate feature regions. These regions are typically located at thread rib edges, rolling texture boundaries, and locations of abrupt temperature changes. Subsequently, Scale Invariant Feature Transform (SIFT) or Accelerated Robust Feature Transform (SURF) methods are used for feature point detection, combined with FAST corner detection to improve detection efficiency in high-speed scenes. Typical parameters are set as follows: minimum feature response threshold of 0.01, number of scale layers of 3–5, and minimum spacing between feature points of 5 pixels to avoid overly dense distribution leading to matching confusion. At a resolution of 640×512, 200–400 feature points can be stably extracted per frame, and their coordinates, principal direction, and local temperature texture descriptors are recorded. To improve stability, transient feature points with temperature fluctuations exceeding ±15 °C are removed to avoid false features caused by oxide scale peeling. The offset between adjacent frames is calculated based on image feature points to obtain the global offset index for adjacent frames. After extracting feature points from multiple frames, the overall displacement between adjacent frames is calculated using a feature matching method. Nearest neighbor matching and a ratio test (Lowween ratio 0.7–0.8) are used to screen reliable matching point pairs, and the Random Sample Consensus (RANSAC) algorithm is used to remove mismatched points to improve matching robustness. Under typical conditions, 80–150 pairs of valid matching points can be obtained between adjacent frames. The local displacement vector is calculated based on the coordinate difference of the matching point pairs, and a weighted average of all vectors is taken to obtain the global displacement vector (ΔX_g, ΔY_g). The global offset index GI is then defined as: GI = √(ΔX_g² + ΔY_g²) / W; Where W is the effective width (in pixels) of the steel in the image. For example, when the average displacement is 12 pixels and the effective width is 400 pixels, GI≈0.03. This index reflects the overall image stability and degree of deformation. To improve accuracy, it can be corrected by combining production line speed data to keep the calculation error within ±1 pixel. By quantifying the overall inter-frame offset, a unified evaluation index is provided for identifying abnormal deformation.

[0025] To identify abnormal images caused by mechanical vibration, steel movement, or local deformation, an image deformation coefficient threshold is set as the judgment criterion. Based on statistical results under stable production line operating conditions, the global offset index (GI) typically stabilizes within the range of 0.01–0.05; therefore, the deformation coefficient threshold can be set to 0.08–0.12. When the GI exceeds this threshold, it is judged as an abnormal frame. In specific implementation, a sliding window (window length 5–10 frames) is used to statistically analyze the GI of consecutive frames. If the GI of a single frame suddenly increases by more than twice the average value, or if three consecutive frames exceed the threshold, it is marked as a deformation anomaly. Common causes of anomalies include changes in roller vibration frequency (5–15 Hz), increased steel sway amplitude (>3 mm), or thermal expansion differences caused by uneven local temperature. After marking abnormal frames, their timestamps and deformation directions are recorded. For the marked abnormal frames, a spatial consistency interpolation method is used to repair them to maintain sequence continuity and temperature distribution consistency. Two to three frames before and after the anomalous frame are selected as reference frames. A spatial mapping relationship is established using feature point matching, and the reference frames are geometrically aligned using affine transformation or a thin plate spline (TPS) model. Subsequently, the temperature matrix in the anomalous frame is reconstructed using weighted interpolation, with weights determined based on temporal distance and spatial consistency (e.g., temporal weight 0.4, spatial consistency weight 0.6). For locally missing regions (typically less than 5% of the image area), bilinear interpolation or anisotropic diffusion based on neighborhood temperature gradients is used for compensation to maintain smooth temperature transitions and prevent blurring of defect boundaries. The temperature deviation of the repaired frame can be controlled within ±2 °C, and the geometric offset error is less than 1 pixel, thus ensuring that the spatial location and temperature characteristics of the defect area are not distorted.

[0026] In this embodiment, the specific steps for performing thermal radiation filtering on the optimized imaging sequence to obtain the thermally filtered sequence are as follows: Background segmentation is performed on the optimized imaging sequence to extract the background region image; Background thermal radiation is identified in the background region image to determine the background radiation brightness value; Convert the background radiance value into the equivalent ambient radiance temperature; The disturbance coefficient is obtained by calculating the thermal imaging disturbance based on the equivalent radiation temperature of the environment. Adaptive thermal noise filtering is applied to the first imaging sequence based on the perturbation coefficient to obtain the thermally filtered sequence.

[0027] In this embodiment, after spatial consistency restoration, the steel region and background region in the optimized imaging sequence exhibit significant temperature and texture differences. Background segmentation can be performed using temperature thresholding and morphological methods. A temperature threshold range is set based on the slow-cooling zone conditions. For example, the surface temperature of steel is typically between 400–850 °C, while the background equipment and ambient temperature are mostly in the range of 80–200 °C. An initial segmentation threshold of 250 °C can be set, marking regions below this temperature as candidate background regions. Subsequently, edge detection (Canny operator thresholds set to low threshold 30 and high threshold 80) is used to extract the steel contour, and morphological closing operations (structural element 5×5) are used to fill in edge breakage areas, ensuring the integrity of the steel region. Connectivity analysis is performed on the candidate background regions to remove isolated regions with an area less than 1% of the total image area, avoiding misjudgments caused by oxide scale fragments or high-temperature splashes. After extracting the background region, quantitative analysis of the background thermal radiation characteristics is required to assess the impact of environmental radiation on thermal imaging. Statistical analysis is performed on the temperature matrix of the background region to calculate the average temperature, standard deviation, and radiance distribution. Considering that the background may contain different materials such as roller conveyors, supports, and insulation materials, a partitioned clustering method (such as K-means clustering, K=2–3) should be used to radiate the background region into groups, and the average radiance value of each group should be calculated separately. Radiance can be approximately expressed by the Stefan-Boltzmann law as L=εσT 4 Where ε is the emissivity of the background material (typically 0.6–0.9), and σ is a constant. To improve stability, outliers with temperatures deviating from the mean by ±2 standard deviations are removed to avoid interference from reflected heat or transient high temperatures. Under typical conditions, the average background temperature is approximately 120–180 °C, corresponding to a radiance range of approximately 2.5 × 10⁻⁶. 4 –6.0×10 4 W / m²·sr.

[0028] After obtaining the background radiance value, it needs to be converted into the ambient equivalent radiant temperature for comparison with thermal imaging temperature data. The equivalent radiant temperature T_eq = (L / (εσ))^(1 / 4). In practical applications, to avoid errors caused by emissivity variations, a weighted average emissivity ε_avg = 0.75 can be used as the background uniformity parameter. For example, when the background radiance is 4.0 × 10⁻⁶... 4When it is [[ID=]], the equivalent radiation temperature can be calculated to be approximately 150 °C. To improve the temporal stability, the background temperature of 10 consecutive frames is processed by moving average to reduce the influence of short-term fluctuations and control the temperature fluctuation within ±3 °C. The equivalent temperature can be regarded as the unified influence amount of environmental radiation on thermal imaging, which is used to evaluate the deviation degree of the background reflected heat on the measurement of the steel surface temperature. After determining the environmental equivalent radiation temperature, it is necessary to evaluate its disturbance degree on the thermal imaging measurement of steel. The disturbance mainly comes from the background radiation reflection and air medium scattering, and a disturbance coefficient model can be established through the temperature difference ratio. Suppose the average surface temperature of the steel Ts is approximately 600 °C, and the environmental equivalent radiation temperature Teq is approximately 150 °C, then the temperature difference ratio R = Teq / Ts ≈ 0.25. Further combining the surface emissivity of the steel εs = 0.85 and the environmental reflection coefficient ρ = 1 - εs ≈ 0.15, the disturbance coefficient Kd = ρ × R is established, and the typical disturbance coefficient is about 0.037 - 0.05. When the environmental temperature in the slow cooling zone rises to 180 °C, the disturbance coefficient can rise to about 0.06. To improve the accuracy, the local temperature difference distribution can be calculated frame by frame to generate a spatial distribution map of the disturbance coefficient, which is used to identify the areas with strong background radiation influence.

[0029] An adaptive thermal noise filter is implemented on the original imaging sequence to eliminate the temperature offset and random noise caused by environmental radiation. The filtering strategy adopts an adaptive weight model driven by the disturbance coefficient: when Kd ≤ 0.04, a weak filtering mode is adopted, and the bilateral filtering temperature kernel σr = 4 °C is set; when 0.04 < Kd ≤ 0.06, a medium filtering mode is adopted, and σr is increased to 6 °C; when Kd > 0.06, an enhanced filtering mode is adopted, and σr is set to 8 °C, and a time moving average (α = 0.5) is superimposed to suppress the persistent background interference. The size of the spatial filtering kernel remains at 3×3 or 5×5 to avoid blurring the defect boundary due to excessive smoothing. At the same time, the background reflection component is temperature compensated according to the disturbance coefficient, and the steel surface temperature is subtracted by Kd × (Ts - Teq) to correct the temperature measurement deviation caused by environmental radiation. After the adaptive filtering process, the standard deviation of the thermal map temperature noise can be reduced from ±3.5 °C to ±1.0 - 1.3 °C, and the temperature difference contrast in the defect area is increased by about 30 - 45%.

[0030] In this embodiment, refer to Figure 3 , which is a schematic diagram of the detailed implementation steps of step S2. In this embodiment, the detailed implementation steps of the said step S2 include: Perform real-time inter-frame temperature distribution calculation on the thermal filtering sequence to obtain the temperature distribution field; Set the adaptive size of the grid unit based on the temperature distribution field; Perform micro-grid division on the temperature distribution field according to the adaptive size of the grid unit to obtain the micro-grid distribution field; Based on the microgrid distribution field, the numerical space of the grid cells is transformed, and the temperature value of each grid cell is output.

[0031] In this embodiment, after adaptive thermal noise filtering, each frame of the thermal image corresponds to stable temperature matrix data, which can be directly used to construct the temperature distribution field. The radiance values ​​output by the thermal imager are converted into a temperature matrix T(x,y), and combined with timestamps to form a continuous frame temperature sequence T(x,y,t). To enhance spatial continuity, temperature difference calculation ΔT(x,y,t)=T(x,y,t)–T(x,y,t–1) is performed on adjacent frames to identify temperature change trends and local abnormal regions. For the typical temperature gradient of the slow cooling process of rebar (approximately 5–15 °C / cm along the length direction and approximately 2–6 °C / cm along the lateral direction), bilinear interpolation is used to resample the temperature matrix at the sub-pixel level, improving the spatial resolution from 0.5 mm / pixel to 0.25 mm / pixel, thereby improving the defect boundary resolution capability. At the same time, a weighted average (weights 0.5, 0.3, 0.2) is applied to 3–5 consecutive frames to reduce the impact of transient fluctuations, keeping the temperature distribution fluctuation within ±1 °C.

[0032] To balance computational efficiency and defect detection accuracy, the mesh element size needs to be dynamically adjusted based on the temperature gradient characteristics. The gradient magnitude of the temperature distribution field needs to be calculated. The gradient is then normalized. Based on the surface temperature variation characteristics of the rebar, a gradient threshold range can be set: when G < 3 °C / mm, it is considered a region with gradual temperature change; when G – 8 °C / mm, it is a transition region; when G > 8 °C / mm, it is a high gradient region (usually corresponding to defects or thread edges). The mesh size is adaptively set according to the gradient level: a larger mesh (e.g., 8×8 pixels, corresponding to approximately 2×2 mm) is used for gradual regions, a medium mesh (4×4 pixels) is used for transition regions, and a fine mesh (2×2 pixels) is used for high gradient regions. To avoid boundary discontinuities caused by abrupt changes in mesh size, a neighborhood smoothing strategy is adopted to perform transition adjustments for regions where the size difference between adjacent meshes exceeds two levels.

[0033] Using the effective area of ​​the steel as the dividing range, a mesh generation matrix is ​​established based on the gradient partitioning results, and each region is divided into blocks according to the corresponding size. During the partitioning process, the mesh boundaries are kept aligned with the steel outline to avoid data aliasing caused by crossing the steel edge. For a temperature field with a resolution of 640×512, approximately 8,000–15,000 micro-mesh units can be generated under typical conditions, with high-gradient regions accounting for about 30% of the mesh number, but covering more than 80% of the critical defect areas. To ensure mesh continuity, transition meshes (such as 2×4 or 4×8 pixels) are used to stitch together the boundaries of meshes of different sizes to ensure the continuity of the spatial representation of the temperature field. After partitioning, a micro-mesh distribution field is formed, and each mesh unit contains its spatial location, size, and internal pixel set information.

[0034] Statistical calculations are performed on the pixels within each grid cell, including the average temperature T_avg, maximum temperature T_max, minimum temperature T_min, and temperature standard deviation σ_T, to characterize the local temperature distribution. For smaller grids (2×2 pixels), a weighted averaging method is used (center pixel weight 0.4, other pixels equally weighted) to enhance spatial representativeness; for larger grids (8×8 pixels), a combination of median filtering and mean averaging is used to reduce the impact of local outliers. The grid coordinates are then converted from pixel coordinates to actual physical coordinates (unit: mm). For example, at a resolution of 0.25 mm / pixel, a 4×4 pixel grid corresponds to an actual size of 1×1 mm. Finally, the temperature value and spatial location of each grid cell are output, forming a structured temperature data field. Through this conversion process, the dimensionality of the temperature data can be reduced by approximately 60%, while maintaining the temperature error in key defect areas within ±1.5 °C.

[0035] In this embodiment, step S3 includes the following steps: The temperature values ​​are averaged globally to obtain the average temperature value; Perform temperature difference analysis on a grid cell-by-grid basis based on the average temperature value, and output the temperature difference value of each grid cell; Gradient analysis is performed on the temperature difference value to obtain temperature difference gradient information; Anisotropic heat conduction paths are analyzed based on temperature gradient information to obtain the heat conduction paths.

[0036] In this embodiment, background meshes are removed based on a mask of the steel region, retaining only effective mesh cells (typically 60–75% of the total mesh). The average temperature T_avg_i of all mesh cells is then weighted and summed, with weights determined by the mesh area to avoid bias from different mesh sizes. For example, the weight ratio for a 2×2 pixel fine mesh and an 8×8 pixel coarse mesh is 1:16. The global average temperature T_mean can be expressed as the sum of the products of each mesh temperature and its area weight, divided by the total area. During a typical slow cooling phase, the average temperature of the steel surface is typically in the range of 450–650 °C, with a standard deviation of approximately 20–35 °C. To enhance stability, a moving average of the average temperature over five consecutive frames can be applied to keep fluctuations within ±5 °C. After obtaining the global average temperature, a temperature difference calculation is performed on each mesh cell to quantify the degree of local temperature deviation. Specifically, ΔT_i = T_avg_i – T_mean, where ΔT_i represents the temperature difference value of the i-th mesh. To facilitate defect identification, temperature differences can be divided into multiple levels, for example: |ΔT| < 10 °C is the normal fluctuation zone, 10–25 °C is the slight anomaly zone, and > 25 °C is the significant anomaly zone. Based on the cooling characteristics of rebar, cracked or inclusion areas often exhibit localized heat dissipation anomalies, with temperature differences reaching 30–60 °C; while folded or oxide scale-accumulated areas may form localized high-temperature zones, with temperature differences reaching 20–40 °C. To avoid isolated noise points interfering with the analysis, a 3×3 neighborhood median filter can be applied to the temperature difference field to remove anomalies with an area smaller than two grid cells.

[0037] The temperature gradient components are calculated using the central difference method: and And further calculate the gradient magnitude. Under typical cooling conditions, the gradient amplitude in the normal region is usually less than 5 °C / mm, while the gradient near the defect can reach 10–25 °C / mm. To improve stability, the gradient field is Gaussian smoothed (σ=1 pixel) to reduce the influence of discrete noise. Subsequently, the gradient direction is... Determine the direction of heat diffusion and statistically analyze the spatial consistency of gradient directions. When more than 70% of the grid gradient directions are consistent within a certain region and the gradient magnitude continues to increase, it can be identified as an abnormal heat conduction region.

[0038] Further analysis of heat conduction paths on the steel surface was conducted to identify anisotropic thermal conductivity caused by material defects or microstructure inhomogeneity. A heat flow vector field was constructed with the gradient direction as the reverse vector of heat flow direction (heat is transferred from high temperature to low temperature). A path tracing algorithm was then used to connect adjacent grids along the gradient direction, starting from the high-temperature anomaly grid, to form a continuous heat conduction path. To avoid noise interference, a minimum gradient threshold of 8 °C / mm was set, and path connections were only made for grids exceeding this threshold. Normally, heat conduction paths on the steel surface are uniformly parallel, but when cracks or inclusions are present, the paths will bend, branch, or be interrupted, with path deflection angles reaching 15–40°. The degree of thermal anomaly can be quantified by statistically analyzing path length, curvature, and the number of branches. For example, when the path curvature exceeds 0.3 mm... -1 If a continuous interruption with a length greater than 5 mm occurs, it can be identified as a potential defect area.

[0039] In this embodiment, the specific steps of step S4 are as follows: Regional thermal stability is assessed based on heat conduction paths, and the thermal inertia consistency index is obtained. Identify the specifications and models of the rebar currently being produced on the production line; Standard thermal response index is extracted from a pre-set database based on the specifications and models of rebar. Based on the standard thermal response index, the thermal inertia consistency index is calculated to obtain an abnormal thermal behavior index. Potential thermal defects are analyzed based on abnormal thermal behavior indicators, and the locations of potential defects are marked.

[0040] In this embodiment, the heat conduction path is divided into several continuous region units, each containing multiple microgrid units. Thermal stability is measured by statistically analyzing the temperature difference amplitude, gradient change rate, and heat flow direction change in each region. The Thermal Inertia Consistency Index (TIC) can be defined as the ratio of the standardized root mean square deviations of the temperature change rates in each region. σ_region represents the standard deviation of the regional temperature change rate, and σ_global represents the standard deviation of the temperature change rate of the entire effective region of the steel. Under typical slow cooling conditions for rebar, the normal region TIC is usually between 0.85 and 0.95, while in defective regions, due to localized heat diffusion obstruction or uneven thermal conduction, the TIC may drop to 0.5–0.7. To enhance stability, the regional TIC can be calculated for 5–10 consecutive frames and then averaged, reducing the impact of short-term temperature fluctuations on the index to within ±0.02. The rebar specification (e.g., HRB400, HRB500, diameter 12–32 mm) is determined through the production line management interface or by automatically recognizing image features (such as steel diameter, thread rib height, and spacing). The image recognition method can use Hough transform to detect the cylindrical profile and combine it with thread rib template matching to calculate the thread spacing and rib height, with a typical error controlled within ±0.5 mm. After confirming the specification, the standard thermal performance parameters of the steel in the preset database can be used, including thermal conductivity, specific heat capacity, and typical cooling thermal response curves.

[0041] Standard thermal response indices (STRIs) are extracted from a pre-set database based on rebar specifications. After identifying the steel type, corresponding STRIs are extracted from the database, including typical cooling curves, temperature gradient thresholds, thermal inertia reference values, and thermal decay rates at various locations. The temperature changes of each steel specification under slow cooling conditions are pre-measured and normalized in the database to form a Standard Thermal Response Index (STRI), which typically includes the temperature drop rate along the length, transverse temperature uniformity index, and thermal inertia differences between thread ribs. Typical values, such as the cooling rate along the length of HRB500 φ20mm steel, are approximately 12–18 °C / s, with a transverse temperature difference not exceeding 8 °C / mm. These standard indices serve as a reference for normal thermal behavior, and are compared and analyzed using the calculated thermal inertia consistency index (TIC) and the STRI. Deviation can be defined as... Alternatively, the Abnormal Thermal Behavior Index (ATBI) can be obtained through normalization, such as ATBI = |ΔTIC| / TIC_standard. Under typical conditions, the ATBI of normal steel is usually below 0.15, while the ATBI of areas with local defects may exceed 0.3–0.5, reflecting a significant deviation of the thermal inertia of the region from the standard cooling characteristics. To enhance robustness, the ATBI of consecutive frames can be time-smoothed (a moving average window of 5 frames) to eliminate misjudgments caused by short-term temperature disturbances. A threshold is set, for example, ATBI > 0.25 is used to identify potential defect areas, and isolated points are screened based on spatial continuity (abnormal points with an area less than 2 microgrid units are removed). Subsequently, topological analysis is performed on the abnormal areas based on the direction and gradient information of the heat conduction path to mark the locations of possible cracks, folds, inclusions, or local porosity. Typical defect characteristics include continuous distribution of high local ATBI values, interruption of heat conduction paths or abnormal curvature, and abnormally enhanced temperature gradients. The final output is a heat map with potential defect markings, which can achieve millimeter-level positioning accuracy on the surface of threaded steel, and at the same time provide input for subsequent intelligent classification or alarm systems.

[0042] In this embodiment, the specific steps of step S5 are as follows: Cross-frame temporal tracking is performed on potential defect locations to extract the thermal evolution time series of the defect locations; The thermal morphological changes of defects are calculated from the thermal evolution time series to obtain the morphological change trajectory; Calculate the defect propagation direction and propagation rate of the morphological change trajectory; Based on the defect propagation direction and propagation rate, a trend prediction is made to obtain the defect development trend; The defect development trend is classified into different defect types to obtain the classification results; the classification results include occasional thermal disturbance anomalies and persistent structural hazards; A risk assessment is conducted on the development trend of defects to obtain the defect risk level; A comprehensive diagnosis of rebar is performed based on the defect risk level and differentiation results, generating a defect diagnosis report.

[0043] In this embodiment, after marking potential thermal defects, each defect region needs to be time-series tracked across consecutive frames to obtain the changes in its temperature and spatial characteristics over time. A defect region location index is established using microgrid spatial coordinates, and its location is matched across consecutive frames. Matching methods can employ feature-point-based optical flow tracing (such as the Lucas-Kanade optical flow method) or template matching based on normalized cross-correlation, ensuring consistent identification of the same defect across different frames. Under typical acquisition conditions, the frame rate is 200–500 frames per second, ensuring continuous defect tracking without frame loss when the rebar cooling rate is 2–5 m / min. The extracted thermal evolution time series includes the average temperature, maximum temperature, temperature difference, gradient, and area change of the defect region in each frame. Using the thermal evolution data from consecutive frames, the spatial morphology of the defect region is analyzed over time. Contour extraction is performed on the defect region in each frame, using a combination of temperature threshold segmentation and edge detection to obtain the defect boundary coordinates. Subsequently, based on the spatial mapping relationship of the defect contours across consecutive frames, the defect area, aspect ratio, principal axis direction, and contour curvature changes are calculated to form a thermal morphology change trajectory. Under typical observation conditions, the rate of defect propagation along the length direction caused by cracks or folds is approximately 0.1–0.3 mm / s, while thermal anomalies caused by localized oxide scale or spatter often exhibit rapid changes in area but random locations.

[0044] After obtaining the thermal morphology trajectory of the defect, its propagation direction and rate need to be quantified. The propagation direction can be calculated by connecting the coordinates of the center points of the defect contour in adjacent frames, such as θ = arctan(Δy / Δx), representing the main propagation direction of the defect in the image coordinate system. The propagation rate can be calculated by the displacement of the center points of consecutive frames and the time interval, such as v = √(Δx² + Δy²) / Δt, and the actual propagation rate is corrected by combining the defect area change rate. The propagation speed of steel cracks along the length direction is 0.1–0.3 mm / s, and the lateral propagation speed is less than 0.05 mm / s. Using the propagation direction and rate information, the possible evolution of the defect during the subsequent slow cooling process can be predicted. Linear or nonlinear regression methods are used to extrapolate the propagation trajectory, and multi-parameter weighted prediction is performed by combining thermal gradient changes, local thermal inertia consistency, and historical defect propagation patterns. For example, for crack-type defects, it can be assumed that they propagate at a constant rate along the length direction, and the propagation probability is increased where the boundary temperature gradient exceeds 10 °C / mm. The prediction results can generate the trend of defect location, shape and area changes in the next few to tens of seconds.

[0045] Defect development trends are analyzed to differentiate between types, resulting in distinctions between sporadic thermal disturbances and persistent structural hazards. Based on the defect's thermal evolution time series, morphological change trajectory, and expansion rate, the defect type can be determined. Sporadic thermal disturbances typically manifest as short-term abnormal temperature fluctuations, random area changes, and no clear pattern in expansion direction, with a duration generally less than 5–10 seconds. Persistent structural hazards (such as cracks, folds, or inclusions) are characterized by persistent temperature anomalies, continuous area increases over time, and a stable expansion direction exhibiting a linear or curvilinear trend. The determination method combines threshold judgment and pattern recognition: defects with a sustained temperature difference above 20–30 °C, an expansion rate greater than 0.1 mm / s, and a duration exceeding 15 seconds are marked as persistent structural hazards; otherwise, they are classified as sporadic thermal disturbances.

[0046] Based on the defect type determination, it is necessary to assess its potential risks to the quality of rebar or production safety. Risk assessment can comprehensively consider parameters such as defect propagation rate, area, location, and duration, using a multi-factor weighted scoring method to calculate the Defect Risk Index (DRI), and classifying it into three risk levels: low, medium, and high according to a preset classification. For example, DRI < 0.3 is low risk, 0.3 ≤ DRI < 0.6 is medium risk, and DRI ≥ 0.6 is high risk. Under typical conditions, the DRI for persistent crack defects can reach 0.65–0.8, with high-risk areas often located in the middle of the steel or at the threaded rib connection; the DRI for occasional thermal disturbances is generally below 0.25.

[0047] The spatial location, propagation trajectory, type determination, risk level, and development trend of defects are integrated to form a visual diagnostic report. The report includes heat map annotations, microgrid temperature distribution, heat conduction paths, defect development trend arrows, and risk level markings, along with textual descriptions such as "A persistent crack exists in area #12, propagating along its length, with a high risk level, requiring close attention."

[0048] In this embodiment, an image processing-based intelligent defect diagnosis device for rebar is provided, used to execute the image processing-based intelligent defect diagnosis method for rebar as described above, including: The imaging module is used to perform continuous frame thermal imaging acquisition of the production line to obtain a thermally filtered sequence. The calculation module is used to perform real-time inter-frame temperature distribution calculation on the thermally filtered sequence and output the temperature value of each grid cell. The heat conduction analysis module is used to analyze the anisotropic heat conduction path based on the temperature value to obtain the heat conduction path. The defect analysis module is used to analyze potential thermal defects based on the heat conduction path and mark the locations of potential defects. The diagnostic module is used to perform cross-frame time-series tracing of potential defect locations, conduct comprehensive diagnostics on rebar, and generate a defect diagnostic report.

[0049] Therefore, the embodiments should be considered as exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of the equivalents of the application are intended to be included within the invention.

[0050] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement it. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein are implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A method for intelligent defect diagnosis of rebar based on image processing, characterized in that, Includes the following steps: Step S1: Perform continuous frame thermal imaging acquisition on the production line to obtain a thermal filtering sequence; Step S2: Calculate the inter-frame temperature distribution of the thermally filtered sequence in real time and output the temperature value of each grid cell; Step S3: Analyze the anisotropic heat conduction path based on the temperature value to obtain the heat conduction path; Step S4: Analyze potential thermal defects based on the heat conduction path and mark the locations of potential defects; Step S5: Perform cross-frame time-series tracing of potential defect locations, conduct comprehensive diagnosis of rebar, and generate a defect diagnosis report.

2. The intelligent defect diagnosis method for rebar based on image processing according to claim 1, characterized in that, The specific steps of step S1 are as follows: Deploy a high-speed infrared thermal imager in the slow cooling zone of the production line, set a high-speed acquisition frame rate, and perform continuous frame thermal imaging acquisition of the production line to obtain a thermal imaging sequence. Calculate the inter-frame displacement vector and exposure time parameters of the thermal imaging sequence; Thermal imaging motion blur analysis is performed based on the inter-frame displacement vector and exposure time parameters to obtain thermal motion blur image features in the direction of motion. Based on the thermal trailing image features, the thermal imaging sequence is deconvolved frame by frame to eliminate thermal trailing, resulting in a de-thermal trailing frame sequence. Thermal radiation filtering is applied to the de-thermal ghosting frame sequence to obtain the thermally filtered sequence.

3. The intelligent defect diagnosis method for rebar based on image processing according to claim 2, characterized in that, The specific steps for performing thermal radiation filtering on the de-thermal ghosting frame sequence to obtain the thermally filtered sequence are as follows: Multi-frame feature point labeling is performed on the heat-removed ghosting frame sequence to extract image feature points from different frames; Based on the image feature points, the offset between adjacent frames is calculated to obtain the global offset index of adjacent frames; The global offset index is compared based on a preset image deformation coefficient. When the global offset index is greater than the preset image deformation coefficient, it is marked as an abnormal frame. Spatial consistency interpolation is performed on abnormal frames to repair them, and an optimized imaging sequence is output. Thermal radiation filtering is applied to the optimized imaging sequence to obtain a thermally filtered sequence.

4. The intelligent defect diagnosis method for rebar based on image processing according to claim 3, characterized in that, The specific steps for performing thermal radiation filtering on the optimized imaging sequence to obtain the thermally filtered sequence are as follows: Background segmentation is performed on the optimized imaging sequence to extract the background region image; Background thermal radiation is identified in the background region image to determine the background radiation brightness value; Convert the background radiance value into the equivalent ambient radiance temperature; The disturbance coefficient is obtained by calculating the thermal imaging disturbance based on the equivalent radiation temperature of the environment. Adaptive thermal noise filtering is applied to the first imaging sequence based on the perturbation coefficient to obtain the thermally filtered sequence.

5. The intelligent defect diagnosis method for rebar based on image processing according to claim 1, characterized in that, The specific steps of step S2 are as follows: Real-time inter-frame temperature distribution calculations are performed on the thermally filtered sequence to obtain the temperature distribution field; The mesh element size is set adaptively based on the temperature distribution field; The temperature distribution field is divided into microgrids according to the adaptive size of the grid cells to obtain the microgrid distribution field. Based on the microgrid distribution field, the numerical space of the grid cells is transformed, and the temperature value of each grid cell is output.

6. The intelligent defect diagnosis method for rebar based on image processing according to claim 1, characterized in that, The specific steps of step S3 are as follows: The temperature values ​​are averaged globally to obtain the average temperature value; Perform temperature difference analysis on a grid cell-by-grid basis based on the average temperature value, and output the temperature difference value of each grid cell; Gradient analysis is performed on the temperature difference value to obtain temperature difference gradient information; Anisotropic heat conduction paths are analyzed based on temperature gradient information to obtain the heat conduction paths.

7. The intelligent defect diagnosis method for rebar based on image processing according to claim 1, characterized in that, The specific steps of step S4 are as follows: Regional thermal stability is assessed based on heat conduction paths, and the thermal inertia consistency index is obtained. Identify the specifications and models of the rebar currently being produced on the production line; Standard thermal response index is extracted from a pre-set database based on the specifications and models of rebar. Based on the standard thermal response index, the thermal inertia consistency index is calculated to obtain an abnormal thermal behavior index. Potential thermal defects are analyzed based on abnormal thermal behavior indicators, and the locations of potential defects are marked.

8. The intelligent defect diagnosis method for rebar based on image processing according to claim 1, characterized in that, The specific steps of step S5 are as follows: Cross-frame temporal tracking is performed on potential defect locations to extract the thermal evolution time series of the defect locations; The thermal morphological changes of defects are calculated from the thermal evolution time series to obtain the morphological change trajectory; Calculate the defect propagation direction and propagation rate of the morphological change trajectory; Based on the defect propagation direction and propagation rate, a trend prediction is made to obtain the defect development trend; The defect development trend is classified into different defect types to obtain the classification results; the classification results include occasional thermal disturbance anomalies and persistent structural hazards; A risk assessment is conducted on the development trend of defects to obtain the defect risk level; A comprehensive diagnosis of rebar is performed based on the defect risk level and differentiation results, generating a defect diagnosis report.

9. A smart defect diagnosis device for rebar based on image processing, characterized in that, The method for performing intelligent defect diagnosis of rebar based on image processing as described in claim 1 includes: The imaging module is used to acquire continuous frame thermal images of the production line to obtain a thermally filtered sequence. The calculation module is used to perform real-time inter-frame temperature distribution calculation on the thermally filtered sequence and output the temperature value of each grid cell. The heat conduction analysis module is used to analyze the anisotropic heat conduction path based on the temperature value to obtain the heat conduction path. The defect analysis module is used to perform potential thermal defect analysis based on the heat conduction path and mark the location of potential defects. The diagnostic module is used to perform cross-frame time-series tracing of potential defect locations, conduct comprehensive diagnostics on rebar, and generate a defect diagnostic report.

Images