A bridge disease intelligent identification and wisdom decision method and system
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHAANXI TRAFFIC CONTROL KAIDA ROAD & BRIDGE ENG CONSTR CO LTD
- Filing Date
- 2026-04-16
- Publication Date
- 2026-07-24
Smart Images
Figure CN122156999B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of image and video recognition technology, and in particular to a method and system for intelligent identification and smart decision-making of bridge defects. Background Technology
[0002] During long-term service, bridge structures are susceptible to various defects such as cracks, steel corrosion, concrete spalling, and interface debonding due to repeated traffic loads, cyclical changes in environmental temperature and humidity, and material aging. In existing bridge reinforcement projects, the use of carbon fiber reinforced polymer (CFRP) to bond and reinforce key components such as beams, slabs, and piers has become a common technical approach. While carbon fiber materials offer advantages such as high strength, lightweight, and corrosion resistance, their reinforcement effectiveness is highly dependent on the bonding quality between the carbon fiber layer and the concrete matrix. If voids or debonding occur at the interface, it not only weakens the load-bearing capacity of the reinforcement layer but may also lead to premature failure of the reinforcement system, posing safety risks.
[0003] Currently, the main detection methods for interface debonding problems in carbon fiber reinforcement layers include manual tapping and infrared thermography. Manual tapping relies on human judgment based on differences in hollow sounds, resulting in low efficiency, significant susceptibility to operator experience, and operational risks in high-altitude bridge environments. Infrared thermography, on the other hand, thermally excites the structural surface and acquires temperature distribution images, identifying defects based on differences in heat conduction between defective and intact areas. Compared to manual methods, it offers advantages such as being non-contact and highly efficient, thus finding application in bridge inspection.
[0004] However, in practical engineering applications, carbon fiber reinforced surfaces are typically coated with a black resin layer, which has high emissivity and a certain degree of specular reflection, making it prone to reflective interference under natural or artificial light conditions. Local bright spots or false edges often appear in infrared images, leading to a decrease in defect identification accuracy. Furthermore, traditional thermal imaging analysis is mostly based on grayscale or temperature difference comparisons of single or a small number of frames, essentially a static analysis mode. For early, small void areas, the temperature difference response amplitude is low and easily masked by environmental noise, making reliable identification difficult.
[0005] Furthermore, carbon fiber interface debonding typically exhibits minute dynamic deformation characteristics during controlled thermal excitation. For example, localized areas may show slight bulges due to the expansion of internal air caused by heat, while surface resin wrinkles or construction-residual textures mainly present static geometric features. Existing detection methods lack the ability to continuously track and quantify the evolution of microscopic deformation on the material surface during heating, making it difficult to effectively distinguish between "static surface unevenness" and "structural debonding that evolves over time," and hindering the accurate identification of early-stage hollowing defects at carbon fiber reinforced interfaces. Summary of the Invention
[0006] To improve the accuracy of identifying early-stage hollowing defects at the interface of carbon fiber reinforced steel, this application provides a method and system for intelligent identification and smart decision-making of bridge defects.
[0007] Firstly, this application provides a method for intelligent identification and smart decision-making regarding bridge defects, employing the following technical solution: A method for intelligent identification and smart decision-making of bridge defects includes: acquiring surface images of the carbon fiber reinforced surface of a bridge during multiple controlled thermal excitation processes to form a surface image sequence; and extracting a structural skeleton image representing the image texture for any frame in the surface image sequence. Based on the skeleton image sequence composed of structural skeleton images, a dense optical flow field is constructed, and the displacement vector of each pixel at adjacent sampling times is calculated. For any pixel in the structural skeleton image, calculate the local expansion potential energy representing the spatial deformation characteristics and the thermal dynamic response value representing the temporal acceleration characteristics based on the pixel's displacement vector. The local expansion potential energy is fused with the thermal dynamic response value to obtain the comprehensive peeling risk index of each pixel; the disease level of the carbon fiber reinforcement layer is determined based on the comprehensive peeling risk index.
[0008] During controlled thermal excitation, a sequence of surface images is acquired. Under thermal action, the hollow areas at the interface undergo observable minute deformations. Compared with single-frame static thermal image analysis, the method in this application introduces time dimension information, providing a basis for identifying early and subtle defects.
[0009] Secondly, by extracting the structural skeleton image, false features caused by resin reflection and texture repetition are filtered out, ensuring that subsequent displacement analysis is based on stable structural features and improving data reliability. Thirdly, a dense optical flow field is constructed based on the skeleton image sequence, and the displacement vector of each pixel is calculated, achieving a quantitative characterization of continuous deformation and enabling the accurate capture of spatial divergence features caused by interface debonding.
[0010] Furthermore, by constructing a local expansion potential energy model to characterize spatial divergence, and simultaneously constructing a thermal dynamic response value to characterize the acceleration characteristics of displacement modulus changing over time, the evolution of defects is depicted from both spatial and temporal dimensions, effectively distinguishing between static surface unevenness and structural debonding that evolves with heat. Finally, the two are integrated to form a comprehensive debonding risk index, which is used to determine the defect level, realizing a complete decision-making chain from data acquisition to risk output. Accurate identification of early-stage delamination is achieved through dynamic deformation analysis of the carbon fiber reinforced surface texture, improving identification accuracy and reliability.
[0011] Optionally, the surface image is transformed to the frequency domain, and the image is convolved through a set of multi-scale, multi-directional filters. The phase coincidence of frequency components at the same position at different scales and directions is analyzed to obtain the phase consistency intensity. The structural skeleton image is then extracted based on the phase consistency intensity.
[0012] The image is transformed to the frequency domain and convolved using multi-scale, multi-directional filters to calculate phase consistency intensity and extract the structural skeleton. Frequency domain analysis can decompose texture information at different scales and directions, ensuring that the edges of real structures exhibit stable phase coincidence characteristics across multiple frequency components, while random reflection noise struggles to maintain consistency across different scales. This reduces the interference of illumination intensity fluctuations and specular reflections on the recognition results.
[0013] Optionally, the local expansion potential energy characterizing the spatial deformation features is calculated based on the displacement vector of the pixel, including: for any pixel, obtaining the spatial divergence of the displacement vector corresponding to the pixel, using the Sigmoid function to nonlinearly suppress the spatial divergence to obtain the positive expansion degree; and using the product of the positive expansion degree and the phase consistency intensity of the pixel as the local expansion potential energy.
[0014] Spatial divergence is calculated from the displacement vector field, and nonlinear suppression is performed using the Sigmoid function. This nonlinearity is then multiplied by the phase consistency intensity to construct the local expansion potential. Spatial divergence reflects whether a local region exhibits an outward divergence tendency and is a direct mathematical representation of the thermal expansion of hollow areas. Nonlinear mapping of the divergence using the Sigmoid function suppresses negative contraction signals, avoiding misjudgments caused by local contraction or noise fluctuations, making the model sensitive only to positive expansion. Introducing phase consistency intensity as a weight weakens the divergence contribution from low-confidence texture regions.
[0015] Optionally, the phase coherence intensity is the ratio of the sum of the local energy values of all frequency components to the sum of the amplitude values.
[0016] Optionally, the displacement vector of each pixel at adjacent sampling times is calculated, including: using the Farneback algorithm to perform polynomial expansion modeling on adjacent frame images to obtain the motion vector of the pixel, and converting the motion vector of the pixel into a displacement vector according to a preset mapping scale.
[0017] The Farneback algorithm is used to perform polynomial expansion modeling to obtain pixel motion vectors, which are then converted into physical displacement vectors by mapping a scale. Polynomial expansion modeling can estimate continuous displacement fields with sub-pixel accuracy, making it suitable for capturing minute deformations; the dense optical flow method ensures that motion information can be obtained for each pixel, enabling full-field analysis.
[0018] Optionally, a log-Gaussian filter can be used to perform a convolution operation on the image.
[0019] Logarithmic Gaussian filters have zero DC component characteristics in the frequency domain, which can avoid interference from low-frequency components on edge detection and maintain good resolution in the high-frequency region.
[0020] Optionally, the thermal dynamic response value representing the temporal acceleration characteristics is calculated based on the displacement vector of the pixel, including: obtaining the magnitude difference of the displacement vector corresponding to the same pixel at the current time and the previous time, using the ratio of the magnitude difference to the time interval between the two times as the influence coefficient, and determining the thermal dynamic response value based on the influence coefficient, wherein the influence coefficient is positively correlated with the thermal dynamic response value.
[0021] An influence coefficient is constructed by calculating the ratio of the displacement modulus difference between adjacent time points to the time interval, and this coefficient is positively correlated with the thermal dynamic response value. This design essentially standardizes the displacement change rate, ensuring consistency of results under different sampling frequencies. The actual debonding region exhibits an accelerated expansion trend during heating, and this influence coefficient reflects the acceleration characteristics of texture changes. Static texture regions do not undergo continuous changes, resulting in lower thermal dynamic response values.
[0022] Optionally, the local expansion potential energy is fused with the thermal dynamic response value to obtain a comprehensive stripping risk index for each pixel. This includes: constructing a local spatial neighborhood window centered on the target pixel; calculating the product of the local expansion potential energy and the thermal dynamic response value for any pixel within the local spatial neighborhood window; summing all products within the local spatial neighborhood window to obtain the comprehensive risk level; and performing a logarithmic transformation on the comprehensive risk level to obtain the comprehensive stripping risk index.
[0023] Summing the local spatial neighborhood window can suppress the influence of isolated noise points, giving the comprehensive stripping risk index spatial continuity; using a product-based fusion method ensures that the risk value only increases significantly when both spatial expansion and temporal acceleration characteristics are present, thus improving the accuracy of the stripping risk index calculation.
[0024] Optionally, determining the damage level of the carbon fiber reinforcement layer based on the comprehensive peeling risk index includes: comparing the comprehensive peeling risk index with at least one preset risk threshold to determine the damage level of the carbon fiber reinforcement layer in the target area.
[0025] Secondly, this application provides a bridge defect intelligent identification and smart decision-making system, which adopts the following technical solution: A bridge defect intelligent identification and smart decision-making system includes a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the system implements the bridge defect intelligent identification and smart decision-making method described above.
[0026] The aforementioned intelligent identification and smart decision-making method for bridge defects is generated into a computer program and stored in a memory for loading and execution by a processor. Thus, a system is created based on the memory and processor for convenient use.
[0027] This application has the following technical effects: A phase consistency algorithm is employed to overcome strong surface reflection and extract the true structural skeleton. A dense optical flow field is used to simultaneously quantify the spatial local expansion and divergence characteristics and temporal nonlinear acceleration characteristics of defects under heat. Through dual verification and fusion of spatial and temporal dimensions, "static surface unevenness" and "dynamic structural debonding" are accurately distinguished, achieving non-contact, interference-resistant, and high-precision identification of early-stage minor hollow defects in bridges. Attached Figure Description
[0028] Figure 1 This is a flowchart of a method for intelligent identification and smart decision-making of bridge defects according to an embodiment of this application.
[0029] Figure 2 This is a comprehensive risk quantification mapping diagram in a bridge defect intelligent identification and smart decision-making method according to an embodiment of this application. Detailed Implementation
[0030] This application discloses an intelligent identification and smart decision-making method for bridge defects. It utilizes controlled thermal excitation to induce micron-level physical deformation in the defect area, extracts stable structural features under strong reflective background using a phase consistency algorithm, and captures the dynamic displacement of feature points over time using a dense optical flow field. Furthermore, by calculating the local expansion potential energy in the spatial dimension and the thermal dynamic response in the temporal dimension, it achieves a quantitative characterization of the defect evolution law. Finally, a comprehensive stripping risk index is obtained through multiplicative fusion, providing accurate and scientific decision-making basis for bridge maintenance.
[0031] Reference Figure 1 A method for intelligent identification and smart decision-making of bridge defects, comprising: steps S1-S4.
[0032] S1: Collect surface images of the carbon fiber reinforced surface of the bridge during multiple controlled thermal excitation processes to form a surface image sequence; for any frame image in the surface image sequence, extract the structural skeleton image that represents the image texture.
[0033] In this embodiment, it is first necessary to acquire a sequence of surface images of the carbon fiber reinforced surface of the bridge during controlled thermal excitation, and then utilize phase consistency (Phase Congruency)... The algorithm processes the surface image sequence to extract the structural skeleton of the image.
[0034] Specifically, this utilizes a high-resolution industrial camera with a fixed pose (e.g., a resolution of 1000 ppm). Pixel A camera (or similar device) acquires time-series images during the application of controlled thermal excitation to the carbon fiber reinforced surface of a bridge. Controlled thermal excitation refers to the process of raising the temperature of a target object through external energy input; for example, it can utilize the natural heating process of solar radiation, or use a combination of multiple [unspecified] cameras. Artificial heating is achieved using a constant heat source consisting of halogen lamps. During the heating process, due to the air layer existing in the hollow areas between the carbon fiber cloth and the concrete matrix, the thermal conductivity of the air layer is much lower than that of the concrete, causing slight thermal expansion deformation of the carbon fiber surface at the hollow areas.
[0035] Due to carbon fiber reinforced polymer (CFRP) The surface texture is highly repetitive and coated with resin, making it prone to specular reflection under strong light. Directly processing grayscale images would introduce a large amount of false noise. Therefore, this embodiment extracts the structural skeleton of the image by calculating the phase consistency of different pixels at different frequency scales and directions. Phase consistency is a feature extraction method based on frequency domain analysis, which is insensitive to changes in illumination and can extract edges that truly represent the physical structure.
[0036] Specifically: each captured image frame Transformed to the frequency domain, it is then convolved using a set of multi-scale, multi-directional log-Gabor filters.
[0037] In this embodiment, typically selected arrive Each scale and Perform combined convolutions in each direction.
[0038] Phase coherence intensity is obtained by calculating the component energy of each pixel across all scales and directions and analyzing the phase overlap of frequency components at the same location.
[0039] For any pixel, the formula for calculating its phase coherence intensity is: ; In the formula: Represents pixels The phase coherence intensity is a dimensionless scalar value whose range is mapped to... ; Indicates the first Each frequency component is in the coordinate system The local energy value at that location; Indicates the first The amplitude of each frequency component; digital It is a structured constant, used as a smoothing term in the denominator to ensure that the formula holds logically in textureless smooth regions (i.e., when the sum of amplitudes is zero), and to prevent division by zero errors.
[0040] When the frequency components of a pixel are highly consistent in phase, the sum of the local energies of the numerators approaches the sum of the amplitudes of the denominators, causing the phase consistency strength to tend towards... This indicates that the point has a real texture structure; if the point is a random noise point formed by reflection, the phase distribution of each frequency component is disordered, and the energy will cancel each other out, resulting in the numerator being much smaller than the denominator, and the phase consistency intensity tending to be... .
[0041] For example, a pixel with a clearly defined carbon fiber texture. Place, passing through Filter processing at scale n yields the nth... Local energy of each frequency component ,amplitude ;No. Frequency components , ;No. Frequency components , Then the numerator term denominator The phase coherence intensity of this pixel is calculated. This value is close to This indicates that the point is a reliable structural feature point.
[0042] After obtaining the phase consistency intensity of each pixel, the structural skeleton of the image is extracted based on the phase consistency intensity of each pixel. Specifically, an intensity threshold can be set, and pixels with phase consistency greater than the intensity threshold can be used as the structural skeleton of the image.
[0043] S2: Based on the skeleton image sequence composed of the structural skeleton image, a dense optical flow field is constructed, and the displacement vector of each pixel at adjacent sampling times is calculated. The displacement vector includes horizontal and vertical components.
[0044] A dense optical flow field is constructed based on the structural skeleton, and the displacement vector of the pixel is calculated at adjacent sampling times.
[0045] The displacement vector includes a horizontal component. and vertical components In this embodiment, the dense optical flow algorithm is used to calculate the motion vector of pixels at adjacent sampling times. As a preferred embodiment, the following method is adopted: The algorithm processes the skeleton image by performing polynomial expansion modeling on adjacent frames using a multi-scale Gaussian pyramid, thereby obtaining the motion vector of each pixel in the image. Based on the horizontal and vertical pixel offsets in the pixel's motion vector, the horizontal and vertical components of the displacement vector are calculated.
[0046] Since the optical flow algorithm directly outputs pixel offsets in the image coordinate system, it must be converted to displacement values in the physical coordinate system to correspond to the actual degree of damage. The specific process is as follows: First, an image coordinate system is established using a calibration algorithm (unit: ...). ) and physical coordinate system (unit: ) mapping scale For example, Zhang's calibration method can be used to calculate the camera's intrinsic and extrinsic parameters using a known checkerboard calibration board.
[0047] The formula for calculating displacement in the physical coordinate system can be expressed as: ; ; In the formula: This represents the horizontal pixel offset of a texture feature in an image, in units of... ; This represents the vertical pixel offset of a texture feature in an image, in units of... ; Indicates the mapping scale, in units of ; and These represent the physical displacement values of a pixel in the horizontal and vertical components, respectively, in units of... .
[0048] In this embodiment, the mapping scale The preferred range of values is When the scale is larger than When the physical size represented by a single pixel is too large, early-stage deformations at the micrometer level can be overwhelmed by sampling errors; when the scale is smaller than... While the accuracy is extremely high, the camera's field of view is too small, making it difficult to perform large-area detection. Therefore, the scale is controlled at... It can ensure both accuracy and detection efficiency.
[0049] Assuming it is determined through calibration At that moment The optical flow algorithm calculates the relative position of a pixel. It has been moving horizontally at all times. The pixel moved vertically. Pixel. Then the physical displacement value of that point is: , .
[0050] S3: For any pixel in the structural skeleton image, calculate the local expansion potential energy representing the spatial deformation characteristics and the thermal dynamic response value representing the temporal acceleration characteristics based on the pixel's displacement vector.
[0051] In areas with hollow defects, the internal air expands due to heat, generating back pressure that causes tiny outward bulges in the surface carbon fibers. This physical phenomenon exhibits a clear divergent characteristic in a two-dimensional displacement field.
[0052] Therefore, the local expansion potential energy characterizing spatial deformation can be calculated based on the displacement vector. For any pixel, the spatial divergence of the displacement vector corresponding to that pixel is obtained, and the positive expansion degree is obtained by nonlinearly suppressing the spatial divergence using the Sigmoid function. The product of the positive expansion degree and the phase consistency intensity of the pixel is taken as the local expansion potential energy.
[0053] The formula for calculating local expansion potential energy can be expressed as: In the formula: Represents pixels The local expansion potential energy is a dimensionless value; Represents pixels Phase coherence intensity; Represents pixels The spatial gradient of the horizontal displacement component along the horizontal direction reflects the texture in the neighborhood of that point. The rate of tensile deformation on the shaft; Represents pixels The spatial gradient of the vertical displacement component along the vertical direction; This represents the natural exponential function.
[0054] This represents the spatial divergence of the displacement vector field. When the divergence is positive and the value is large, it indicates that there is a significant outward expansion trend in the region, which is a typical spatial characteristic of hollowing disease. This section is for introduction. A type-type logic function term is used to suppress spatial divergence nonlinearly, ensuring that only positive expansion is significantly activated. Simultaneously, phase consistency strength is introduced as a weighting coefficient, ensuring that the expansion signal is only accepted in regions with clear texture and high data confidence, effectively filtering out spurious deformations caused by noise.
[0055] Because the deformation rate of real structural debonding defects will exhibit a nonlinear acceleration characteristic as heat accumulates during the heating process, while static surface unevenness will not accelerate over time.
[0056] Therefore, the thermal dynamic response value, which characterizes the time-domain acceleration, is calculated based on the rate of change of the magnitude of the displacement vector. For any pixel, the difference in magnitude of the displacement vector corresponding to the same pixel at the current time and the previous time is taken. The ratio of the magnitude difference to the time interval between the two time points is used as an influence coefficient. The thermal dynamic response value is determined based on the influence coefficient, which is positively correlated with the thermal dynamic response value. The formula for calculating the thermal dynamic response value can be expressed as: ; In the formula: Represents pixels The thermal dynamic response value reflects the acceleration characteristics of deformation, and its magnitude corresponds to the change in velocity. Represents pixels The displacement modulus at the current sampling moment is calculated as follows: ; Represents pixels The displacement magnitude at the previous sampling time; Indicates the time step between two samplings (unit: ); This represents the benchmark correction factor, with values as follows: ; This indicates the preset reference speed, mainly used for... Partial standardization is performed to eliminate dimensions; it can be set to 1 mm / s.
[0057] Numerator This reflects the increment of the displacement modulus over time, i.e., the change in modulus. The larger this increment, the greater the thermal dynamic response value, indicating that the affected area is undergoing drastic dynamic evolution. The thermal dynamic response value can effectively distinguish between "static geometric features" and "dynamic debonding defects."
[0058] For example, setting , If a certain pixel is in displacement modulus at time ,exist displacement modulus at time t Then its modulus change is The final calculated thermal dynamic response value is If it is a static surface, the displacement does not change with time. Then its modulus change The result is 0, and therefore the final calculated thermal dynamic response value is 0.
[0059] S4: The local expansion potential energy is fused with the thermal dynamic response value to obtain a comprehensive peeling risk index; the disease level of the carbon fiber reinforcement layer is determined based on the comprehensive peeling risk index.
[0060] A local spatial neighborhood window centered on the target pixel is constructed. For any pixel within the local spatial neighborhood window, the product of its local expansion potential energy and thermal dynamic response value is calculated. The summation of all products within the local spatial neighborhood window yields the comprehensive risk level. A logarithmic transformation is then applied to the comprehensive risk level to obtain the comprehensive stripping risk index. The formula for calculating the comprehensive stripping risk index can be expressed as: In the formula, Represents pixels The overall divestiture risk index at the location; Represented by pixels A local spatial neighborhood window centered on the center (e.g.) or The number of pixels in the window; Represents the pixel coordinates within the window; Represents the first in the local spatial neighborhood window Local dilation potential energy of each pixel; Represents the first in the local spatial neighborhood window Thermal dynamic response value of each pixel; digital This is a logarithmic field safety correction term to ensure that the input for logarithmic operations is always greater than 1. .
[0061] Only when a region simultaneously possesses significant spatial expansion characteristics (high local expansion potential energy) and significant temporal acceleration characteristics (high thermal dynamic response value) can this part be considered valid. This will lead to a significant increase. This "double verification" mechanism reduces false alarms caused by environmental temperature drift or camera shake. By accumulating the product terms of all pixels in the neighborhood and taking the logarithm, isolated noise points can be smoothed, resulting in a stable risk assessment value.
[0062] The comprehensive stripping risk index is compared with at least one preset risk threshold to determine the severity of damage to the carbon fiber reinforcement layer in the target area.
[0063] In this embodiment, two risk thresholds are preset, namely a preset first-level risk threshold. and secondary risk threshold .
[0064] when At this point, it is judged as "well bonded," meaning the carbon fiber layer is tightly bonded to the concrete, posing no safety hazard. When the system determines that there is a "minor risk of hollowness," it recommends adding the area to the key observation list and increasing the monitoring frequency. When the condition is determined to be "severely peeled off," the damage has already affected the structural safety. The system automatically generates reinforcement suggestions, such as "high-pressure injection repair is recommended."
[0065] In addition, combined Figure 2 As a preferred approach, after generating decision recommendations, the system can also generate a comprehensive risk stripping quantification mapping map based on the comprehensive risk stripping index. This cloud map maps the risk index onto a two-dimensional coordinate grid and presents it intuitively using pseudo-color.
[0066] This application also discloses a bridge defect intelligent identification and smart decision-making system, including a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, a bridge defect intelligent identification and smart decision-making method according to this application is implemented.
[0067] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.
[0068] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.
Claims
1. A method for intelligent identification and smart decision-making regarding bridge defects, characterized in that, Surface images of the carbon fiber reinforced surface of the bridge were acquired during multiple controlled thermal excitation processes to form a surface image sequence; For any frame of a surface image sequence, extract the structural skeleton image representing the image texture, including: transforming the surface image to the frequency domain, performing a convolution operation on the image through a set of multi-scale, multi-directional filters, analyzing the phase coincidence of frequency components at the same position at different scales and directions to obtain the phase consistency intensity, and extracting the structural skeleton image based on the phase consistency intensity. Based on the skeleton image sequence composed of structural skeleton images, a dense optical flow field is constructed, and the displacement vector of each pixel at adjacent sampling times is calculated. For any pixel in the structural skeleton image, the local dilatation potential energy representing the spatial deformation characteristics is calculated based on the pixel's displacement vector, including: for any pixel, obtaining the spatial divergence of the corresponding displacement vector, using the Sigmoid function to nonlinearly suppress the spatial divergence to obtain the positive dilatation degree; and using the product of the positive dilatation degree and the phase consistency intensity of the pixel as the local dilatation potential energy. The thermal dynamic response value, which characterizes the temporal acceleration feature, is calculated based on the displacement vector of the pixel. This includes: obtaining the magnitude difference of the displacement vector of the same pixel at the current time and the previous time; using the ratio of the magnitude difference to the time interval between the two times as the influence coefficient; and determining the thermal dynamic response value based on the influence coefficient. The influence coefficient is positively correlated with the thermal dynamic response value. The local expansion potential energy and thermal dynamic response value are fused to obtain the comprehensive peeling risk index of each pixel; the disease level of the carbon fiber reinforcement layer is determined based on the comprehensive peeling risk index.
2. The intelligent identification and smart decision-making method for bridge defects according to claim 1, characterized in that, Phase coherence intensity is the ratio of the sum of the local energy values of all frequency components to the sum of the amplitude values.
3. The intelligent identification and smart decision-making method for bridge defects according to claim 1, characterized in that, Calculating the displacement vector of each pixel at adjacent sampling times includes: using the Farneback algorithm to perform polynomial expansion modeling on adjacent frame images to obtain the motion vector of the pixel, and converting the motion vector of the pixel into a displacement vector according to a preset mapping scale.
4. The intelligent identification and smart decision-making method for bridge defects according to claim 1, characterized in that, A log-Gaussian filter is used to perform a convolution operation on the image.
5. The intelligent identification and smart decision-making method for bridge defects according to claim 1, characterized in that, The local expansion potential energy is fused with the thermal dynamic response value to obtain the comprehensive stripping risk index for each pixel. This includes: constructing a local spatial neighborhood window centered on the target pixel; for any pixel within the local spatial neighborhood window, calculating the product of its local expansion potential energy and thermal dynamic response value; summing all products within the local spatial neighborhood window to obtain the comprehensive risk level; and performing a logarithmic transformation on the comprehensive risk level to obtain the comprehensive stripping risk index.
6. The intelligent identification and smart decision-making method for bridge defects according to claim 1, characterized in that, Determining the damage level of the carbon fiber reinforcement layer based on the comprehensive stripping risk index includes: comparing the comprehensive stripping risk index with at least one preset risk threshold to determine the damage level of the carbon fiber reinforcement layer in the target area.
7. A bridge defect intelligent identification and smart decision-making system, characterized in that, include: A processor and a memory, wherein the memory stores computer program instructions that, when executed by the processor, implement a bridge defect intelligent identification and smart decision-making method according to any one of claims 1-6.