Artificial Intelligence-Based Method and System for Bridge Crack Monitoring

Abstract

This invention provides an artificial intelligence-based method and system for monitoring bridge cracks. The method includes the following steps: identifying principal stress regions based on a bridge finite element model and deploying an orthogonal fiber optic mesh; acquiring images of regions corresponding to light intensity attenuation signals; segmenting the acquired images for cracks using an adaptive adversarial network (ADAN) to generate a binarized mask and connected component coordinates; dividing the connected components into micro-element meshes based on the binarized mask and connected component coordinates, and calculating the radius of curvature and risk weight of each micro-element; and fusing fiber optic strain data with visual parameters to calculate a risk index and output an alarm signal. By filtering images of regions with stress changes based on light intensity attenuation signals, and using an ADNA to achieve environmentally robust crack segmentation, this invention solves the failure problem of traditional visual algorithms in complex environments. By fusing visual parameters and fiber optic strain data to output a dynamic risk index, high-precision risk warning is achieved.

Description

Technical Field

[0001] This invention relates to the field of bridge monitoring technology, and in particular to a bridge crack monitoring method and system based on artificial intelligence. Background Technology

[0002] Artificial intelligence-based bridge crack monitoring refers to the process of using technical means to capture, quantify, and evaluate the geometric characteristics and expansion dynamics of cracks on the surface of bridge structures in real time, in order to provide early warning of structural damage.

[0003] Existing technologies include manual inspection and point sensor monitoring methods, which are inefficient and have limited coverage. While visual algorithms are highly efficient for crack monitoring, they are prone to false detections or missed detections due to the influence of ambient light, resulting in poor monitoring performance and delayed maintenance decisions. Summary of the Invention

[0004] In view of the shortcomings of the existing technology, the purpose of this invention is to provide a bridge crack monitoring method and system based on artificial intelligence, which aims to solve the technical problems of poor monitoring effect and delayed maintenance decision-making in the existing technology.

[0005] To achieve the above objectives, in a first aspect, the present invention provides: a bridge crack monitoring method based on artificial intelligence, comprising the following steps:

[0006] Based on the bridge finite element model, the principal stress region is identified. Orthogonal fiber mesh is laid out with the mid-span and support connection points as the reference. The mesh density is positively correlated with the stress gradient to collect light intensity attenuation signals in real time. Vibration sensors for collecting vibration signals are installed at the nodes of the orthogonal fiber mesh.

[0007] When the light intensity attenuation signal exceeds the threshold and the energy ratio of the vibration signal in the preset frequency band is greater than the preset value, image acquisition is performed on the area corresponding to the light intensity attenuation signal.

[0008] The acquired images are segmented based on an adaptive adversarial network to generate a binarized mask and connected component coordinates.

[0009] Based on the binarized mask and connected component coordinates, the connected component is divided into a micro-element mesh. The radius of curvature and risk weight of each micro-element are calculated. Visual parameters including crack width, length and propagation rate are calculated by weighted formula.

[0010] The risk index is calculated by integrating fiber optic strain data and visual parameters. When the risk index exceeds the alarm threshold, an alarm signal is output.

[0011] According to one aspect of the above technical solution, the steps for image acquisition of the region corresponding to the light intensity attenuation signal specifically include:

[0012] The ambient intensity is detected by a light sensor. When the intensity gradient is greater than the preset gradient value, short exposure frames are collected to obtain crack texture features, and long exposure frames are collected to obtain shadow area features.

[0013] Output image generated based on fusion formula:

[0014] ;

[0015] ;

[0016] In the formula, For short exposure frames, For long exposure frames, For the fusion weights, C is the standard deviation of illumination contrast, j is the basic weight coefficient, b is the decay rate coefficient, and e is the natural constant.

[0017] According to one aspect of the above technical solution, the steps of performing crack segmentation on the acquired image based on an adaptive adversarial network and generating a binarized mask and connected component coordinates specifically include:

[0018] The output image of the region triggered by the light intensity attenuation signal is loaded, and a noise map is generated based on environmental parameters:

[0019] ;

[0020] In the formula, For the output image, K is the environmental scattering convolution kernel fitted based on historical data. For convolution operations, The noise intensity coefficient, It is a random noise matrix;

[0021] The generator takes the noise map as input and outputs a probability mask. The discriminator takes the probability mask and the ground truth mask as input, updates the network parameters by minimizing the loss function, and finally obtains the binarized mask and the coordinates of the connected components by binarizing the probability mask.

[0022] According to one aspect of the above technical solution, the steps of dividing the connected domain into a micro-element mesh based on a binarized mask and connected domain coordinates, and calculating the radius of curvature and risk weight of each micro-element, specifically include:

[0023] Extract the boundary polygons defined by the coordinates of the connected components in the mask, perform pixel meshing on the polygons, and use linear interpolation to complete the infinitesimal elements that cross the boundary.

[0024] The radius of curvature of the micro-element is calculated based on the sub-pixel edge points according to the following formula:

[0025] ;

[0026] In the formula, Let be the radius of curvature. These are the coordinates of the edge points in the micro-mesh. The angle of the tangent direction at the edge point. The rate of change of edge curvature;

[0027] The degree of local bending of the crack is quantified based on the radius of curvature, and the risk weight is calculated based on the material stress sensitivity coefficient.

[0028] ;

[0029] In the formula, Let k be the risk weight of the j-th micro-element, and k be the material stress sensitivity coefficient. The radius of curvature is used to quantify the degree of local bending of the crack.

[0030] According to one aspect of the above technical solution, the calculation expressions for the crack width, length, and propagation rate are as follows:

[0031] ;

[0032] ;

[0033] ;

[0034] In the formula, The width of the crack. Where n is the basic width of the infinitesimal element, and n is the number of infinitesimal elements. The length of the crack. The length of the skeletonized connected component. The physical size of a pixel. Let t be the crack propagation rate, and t be the sliding window size. For the first The width of the crack in the sky, For time intervals.

[0035] According to one aspect of the above technical solution, the calculation expression for the risk index is as follows:

[0036] ;

[0037] ;

[0038] ;

[0039] In the formula, As a risk index, The visual crack risk factor. The fiber strain risk coefficient. For environmental change rate, , , , For contribution weighting coefficients, This is the width weighting coefficient. This is the length weighting coefficient. To expand the weighting coefficients, , The maximum strain value monitored by the optical fiber and its weighting coefficient. , This represents the spatial gradient of strain and its weighting coefficients.

[0040] Secondly, the present invention provides an artificial intelligence-based bridge crack monitoring system, comprising:

[0041] The acquisition module is used to identify the principal stress region based on the bridge finite element model. With the mid-span and support connection points as the reference, an orthogonal fiber optic mesh is laid out. The mesh density is positively correlated with the stress gradient to acquire the light intensity attenuation signal in real time. Vibration sensors for acquiring vibration signals are installed at the nodes of the orthogonal fiber optic mesh.

[0042] The image module is used to acquire an image of the area corresponding to the light intensity attenuation signal when the light intensity attenuation signal exceeds a threshold and the energy proportion of the vibration signal in the preset frequency band is greater than a preset value.

[0043] The segmentation module is used to perform crack segmentation on the acquired image based on an adaptive adversarial network, and generate a binarized mask and connected component coordinates.

[0044] The risk module is used to divide the connected domain into micro-element meshes based on the binarized mask and connected domain coordinates, calculate the radius of curvature and risk weight of each micro-element, and calculate visual parameters including crack width, length and propagation rate through a weighted formula.

[0045] The alarm module is used to calculate the risk index by integrating fiber optic strain data and visual parameters. When the risk index is greater than the alarm threshold, an alarm signal is output.

[0046] According to one aspect of the above technical solution, the acquisition module is specifically used for:

[0047] The ambient intensity is detected by a light sensor. When the intensity gradient is greater than the preset gradient value, short exposure frames are collected to obtain crack texture features, and long exposure frames are collected to obtain shadow area features.

[0048] Output image generated based on fusion formula:

[0049] ;

[0050] ;

[0051] In the formula, For short exposure frames, For long exposure frames, For the fusion weights, C is the standard deviation of illumination contrast, j is the basic weight coefficient, and b is the decay rate coefficient.

[0052] According to one aspect of the above technical solution, the risk module is specifically used for:

[0053] Extract the boundary polygons defined by the coordinates of the connected components in the mask, perform pixel meshing on the polygons, and use linear interpolation to complete the infinitesimal elements that cross the boundary.

[0054] The radius of curvature of the micro-element is calculated based on the sub-pixel edge points according to the following formula:

[0055] ;

[0056] In the formula, Let be the radius of curvature. These are the coordinates of the edge points in the micro-mesh. The angle of the tangent direction at the edge point. The rate of change of edge curvature;

[0057] The degree of local bending of the crack is quantified based on the radius of curvature, and the risk weight is calculated based on the material stress sensitivity coefficient.

[0058] ;

[0059] In the formula, Let k be the risk weight of the j-th micro-element, and k be the material stress sensitivity coefficient. The radius of curvature is used to quantify the degree of local bending of the crack.

[0060] Compared with the prior art, the beneficial effects of the present invention are as follows: by deploying an optical fiber mesh to collect light intensity attenuation signals, and using the light intensity attenuation signals to select areas with stress changes for image acquisition, an environmentally robust crack segmentation is achieved through an adaptive adversarial network, solving the failure problem of traditional visual algorithms in complex environments. Based on a binary mask, a connected domain micro-element mesh is divided, and the curvature radius and risk weight of each micro-element are calculated. By fusing visual parameters and optical fiber strain data, a dynamic risk index is output, achieving high-precision risk warning. Attached Figure Description

[0061] Figure 1 This is a flowchart illustrating the bridge crack monitoring method based on artificial intelligence in the first embodiment of the present invention.

[0062] Figure 2 This is a structural block diagram of the bridge crack monitoring system based on artificial intelligence in the second embodiment of the present invention;

[0063] The following detailed description, in conjunction with the accompanying drawings, will further illustrate the present invention. Detailed Implementation

[0064] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Several embodiments of the invention are illustrated in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

[0065] It should be noted that when a component is said to be "fixed to" another component, it can be directly on the other component or there may be an intervening component. When a component is said to be "connected to" another component, it can be directly connected to the other component or there may be an intervening component. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this document are for illustrative purposes only.

[0066] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0067] Example 1

[0068] Please see Figure 1 The figure shows a flowchart of the bridge crack monitoring method based on artificial intelligence in the first embodiment of the present invention. As shown in the figure, the method includes the following steps:

[0069] Step S100: Based on the bridge finite element model, identify the principal stress regions. Using the mid-span and support connection points as references, lay out an orthogonal fiber optic mesh. The mesh density is positively correlated with the stress gradient to collect light intensity attenuation signals in real time. Vibration sensors for collecting vibration signals are installed at the nodes of the orthogonal fiber optic mesh. By inputting bridge design parameters including span, cross-section, and materials, a three-dimensional finite element model is generated, and loads are applied to calculate the principal stress distribution diagram. The mid-span region is the maximum bending moment region, prone to bottom tensile cracks, while the support connection point region is the maximum shear force region, prone to 45° diagonal cracks.

[0070] Step S200: When the light intensity attenuation signal exceeds a threshold, and the energy proportion of the vibration signal in the preset frequency band is greater than a preset value, image acquisition is performed on the region corresponding to the light intensity attenuation signal. Orthogonal fiber optic mesh analyzes the spatial distribution of strain through optical time-domain reflection. The principle is that cracks cause micro-bending of the fiber, leading to an increase in light intensity attenuation. The aforementioned threshold is preferably 0.5 dB, the aforementioned preset frequency band is 0.1-10 Hz, and the preset value is preferably 0.65. When the proportion is greater than 0.65, it is determined to be crack propagation vibration. Combining dual-threshold filtering of the light intensity attenuation signal and the vibration signal can reduce the false alarm rate.

[0071] Specifically, in this embodiment, the step of image acquisition of the region corresponding to the light intensity attenuation signal includes:

[0072] The ambient intensity is detected by a light sensor. When the intensity gradient is greater than the preset gradient value, short exposure frames are collected to obtain crack texture features, and long exposure frames are collected to obtain shadow area features.

[0073] Output image generated based on fusion formula:

[0074] ;

[0075] ;

[0076] In the formula, For short exposure frames, For long exposure frames, Here, C represents the standard deviation of illumination contrast, j is the base weight coefficient, b is the decay rate coefficient, and e is the natural constant. Adaptive optimization of illumination is achieved using images generated through dual-mode exposure fusion.

[0077] Step S300: Based on the adaptive adversarial network, the acquired image is segmented to generate a binarized mask and connected component coordinates.

[0078] The specific steps for performing crack segmentation on acquired images based on an adaptive adversarial network, and generating a binarized mask and connected component coordinates, include:

[0079] The output image of the region triggered by the light intensity attenuation signal is loaded, and a noise map is generated based on environmental parameters:

[0080] ;

[0081] In the formula, For the output image, K is the environmental scattering convolution kernel fitted based on historical data. For convolution operations, The noise intensity coefficient, It is a random noise matrix;

[0082] The weight distribution of the convolution kernel K corresponds to the scattering intensity in different directions. A high value at the center and decreasing value at the edges indicate that light scattering is strongest in the central region and attenuates towards the edges. When the humidity at the bridge site increases, the center value of K is increased to enhance the blurring effect.

[0083] The generator takes the noise map as input and outputs a probability mask. The discriminator takes the probability mask and the ground truth mask as input, updates the network parameters by minimizing the loss function, and finally obtains the binarized mask and the coordinates of the connected components by binarizing the probability mask.

[0084] Specifically, a noise map is generated based on the environmental scattering matrix to simulate the degradation caused by interference such as rain, fog, and dust, thereby improving the model's generalization ability. The generator includes a CBAM dual-path attention module. Channel attention is used to weight crack feature channels, increasing the weight of crack texture and decreasing the weight of the background. The spatial attention module is used to focus on crack edge regions, increasing the weight of edge regions and decreasing the weight of the center region. The decoding region skip connections are embedded in the CBAM dual-path attention module to fuse shallow texture and deep semantic features, solving the breakpoint problem. The discriminator distinguishes between the probability mask output by the input generator and the manually labeled ground truth mask, identifying the local realism of the generated mask and outputting the probability. The calculation expression of the above loss function is as follows:

[0085] ;

[0086] ;

[0087] ;

[0088] ;

[0089] In the formula, The Dice loss function is used to optimize pixel-level overlap in crack regions and address the issue of imbalanced class outcomes. This is the probability mask output by the generator (each pixel represents the probability value of the crack, ranging from [0-1]). This is a manually labeled true binary mask. To counter the loss function, The probability of authenticity output by the discriminator. This is a probability mask for the generator's output based on the noise map. This is a curvature regularization term used to avoid generating unreasonable jagged edges or burrs, ensuring that the radius of curvature conforms to the laws of material mechanics, and filtering noise interference. Let be the radius of curvature. The spatial gradient of the radius of curvature is the change in curvature between adjacent infinitesimal elements.

[0090] Step S400: Based on the binarized mask and connected component coordinates, the connected component is divided into a micro-element mesh, the radius of curvature and risk weight of each micro-element are calculated, and visual parameters including crack width, length and propagation rate are calculated by weighted formula.

[0091] Specifically, in this embodiment, the steps of dividing the connected domain into a micro-element mesh based on the binarized mask and connected domain coordinates, and calculating the radius of curvature and risk weight of each micro-element, specifically include:

[0092] The boundary polygons defined by the coordinates of the connected components in the mask are extracted, and the polygons are divided into pixel grids. Linear interpolation is used to complete the infinitesimal elements that cross the boundary. The binarized mask is scanned in 8 neighborhoods to mark the independent crack connected components. The coordinate set is output, and the crack edges are fitted by bicubic spline interpolation. Subpixel optimization is performed to reduce coordinate errors.

[0093] The radius of curvature of the micro-element is calculated based on the sub-pixel edge points according to the following formula:

[0094] ;

[0095] In the formula, Let be the radius of curvature. These are the coordinates of the edge points in the micro-mesh. The angle of the tangent direction at the edge point. The rate of change of edge curvature;

[0096] The degree of local bending of the crack is quantified based on the radius of curvature, and the risk weight is calculated based on the material stress sensitivity coefficient.

[0097] ;

[0098] In the formula, Let k be the risk weight of the j-th micro-element, and k be the material stress sensitivity coefficient. The radius of curvature is used to quantify the degree of local bending of the crack.

[0099] Preferably, in this embodiment, the calculation expressions for the crack width, length, and propagation rate are as follows:

[0100] ;

[0101] ;

[0102] ;

[0103] In the formula, The width of the crack. Where n is the basic width of the infinitesimal element, and n is the number of infinitesimal elements. The length of the crack. The length of the skeletonized connected component. The physical size of a pixel. Let t be the crack propagation rate, and t be the sliding window size. For the first The width of the crack in the sky, For time intervals.

[0104] Step S500: Calculate the risk index by integrating fiber optic strain data and visual parameters. When the risk index is greater than the alarm threshold, output an alarm signal.

[0105] Preferably, in this embodiment, the calculation expression for the risk index is:

[0106] ;

[0107] ;

[0108] ;

[0109] In the formula, As a risk index, The visual crack risk factor. The fiber strain risk coefficient. For environmental change rate, , , , For contribution weighting coefficients, This is the width weighting coefficient. This is the length weighting coefficient. To expand the weighting coefficients, , The maximum strain value monitored by the optical fiber and its weighting coefficient. , This represents the strain space gradient and its weighting coefficients. Wherein, , , , The sum of is 1. , , The sum of is 1. , The sum of is 1. Reflecting localized damage mutations, Used to quantify the coupling effect between cracks and strain, where abrupt strain changes induce new cracks.

[0110] In summary, the artificial intelligence-based bridge crack monitoring method in the above embodiments of the present invention collects light intensity attenuation signals by deploying an optical fiber mesh, and selects areas with stress changes based on the light intensity attenuation signals to collect images. Through an adaptive adversarial network, it achieves environmentally robust crack segmentation, solving the failure problem of traditional visual algorithms in complex environments. Based on a binary mask, it divides the connected domain into micro-element meshes, calculates the curvature radius and risk weight of each micro-element, and outputs a dynamic risk index by fusing visual parameters and optical fiber strain data, thereby achieving high-precision risk warning.

[0111] Example 2

[0112] The second embodiment of this application also provides an artificial intelligence-based bridge crack monitoring system, which is used to implement the embodiments and preferred embodiments described herein, and will not be repeated hereafter. As used below, the terms "module," "unit," "subunit," etc., can refer to a combination of software and / or hardware that performs a predetermined function. Although the system described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0113] like Figure 2 As shown, the system includes: acquisition module 100, image module 200, segmentation module 300, risk module 400, and alarm module 500.

[0114] The acquisition module 100 is used to identify the principal stress region based on the bridge finite element model. With the mid-span and support connection points as the reference, an orthogonal fiber optic mesh is laid out. The mesh density is positively correlated with the stress gradient to acquire the light intensity attenuation signal in real time. Vibration sensors for acquiring vibration signals are provided at the nodes of the orthogonal fiber optic mesh.

[0115] The image module 200 is used to acquire an image of the area corresponding to the light intensity attenuation signal when the light intensity attenuation signal exceeds a threshold and the energy ratio of the vibration signal in the preset frequency band is greater than a preset value.

[0116] The segmentation module 300 is used to perform crack segmentation on the acquired image based on an adaptive adversarial network, and generate a binarized mask and connected component coordinates.

[0117] The risk module 400 is used to divide the connected domain into micro-element meshes based on the binarized mask and connected domain coordinates, calculate the radius of curvature and risk weight of each micro-element, and calculate visual parameters including crack width, length and propagation rate through a weighted formula.

[0118] The alarm module 500 is used to calculate the risk index by integrating fiber optic strain data and visual parameters. When the risk index is greater than the alarm threshold, an alarm signal is output.

[0119] Preferably, in this embodiment, the acquisition module 100 is specifically used for:

[0120] The ambient intensity is detected by a light sensor. When the intensity gradient is greater than the preset gradient value, short exposure frames are collected to obtain crack texture features, and long exposure frames are collected to obtain shadow area features.

[0121] Output image generated based on fusion formula:

[0122] ;

[0123] ;

[0124] In the formula, For short exposure frames, For long exposure frames, For the fusion weights, C is the standard deviation of illumination contrast, j is the basic weight coefficient, and b is the decay rate coefficient.

[0125] Preferably, in this embodiment, the risk module 400 is specifically used for:

[0126] Extract the boundary polygons defined by the coordinates of the connected components in the mask, perform pixel meshing on the polygons, and use linear interpolation to complete the infinitesimal elements that cross the boundary.

[0127] The radius of curvature of the micro-element is calculated based on the sub-pixel edge points according to the following formula:

[0128] ;

[0129] In the formula, Let be the radius of curvature. These are the coordinates of the edge points in the micro-mesh. The angle of the tangent direction at the edge point. The rate of change of edge curvature;

[0130] The degree of local bending of the crack is quantified based on the radius of curvature, and the risk weight is calculated based on the material stress sensitivity coefficient.

[0131] ;

[0132] In the formula, Let k be the risk weight of the j-th micro-element, and k be the material stress sensitivity coefficient. The radius of curvature is used to quantify the degree of local bending of the crack.

[0133] It should be noted that the modules can be functional modules or program modules, and can be implemented in software or hardware. For modules implemented in hardware, the modules can reside in the same processor; or the modules can be located in different processors in any combination.

Claims

1. A bridge crack monitoring method based on artificial intelligence, characterized by, The method comprises the following steps: Based on the bridge finite element model, the principal stress region is identified, and the orthogonal optical fiber grid is laid based on the midspan and support connection point as the benchmark, and the grid density is positively correlated with the stress gradient, and the light intensity attenuation signal is collected in real time, wherein the nodes of the orthogonal optical fiber grid are provided with vibration sensors for collecting vibration signals; When the light intensity attenuation signal exceeds the threshold value, and the energy proportion of the vibration signal in the preset frequency band is greater than the preset value, the region corresponding to the light intensity attenuation signal is imaged; Based on the adaptive adversarial network, the collected images are cracked, and a binary mask and a connected domain coordinate are generated; Based on the binary mask and the connected domain coordinate, the connected domain is divided into micro-element grids, the curvature radius and risk weight of each micro-element are calculated, and the visual parameters including crack width, length and expansion rate are calculated through a weighted formula; Fusion of fiber strain data and visual parameter calculation risk index, when the risk index is greater than the alarm threshold, output alarm signal; The step of imaging the region corresponding to the light intensity attenuation signal specifically comprises: Through the light sensor, the environmental intensity is detected, when the intensity gradient is greater than the preset gradient value, the short exposure frame is collected to obtain the crack texture feature, and the long exposure frame is collected to obtain the shadow area feature; Based on the fusion formula, the output image is generated: ； ； wherein, is a short exposure frame, is a long exposure frame, is a fusion weight, C is a standard deviation of illumination contrast, j is a base weight coefficient, b is a decay rate coefficient, and e is a natural constant. Based on the adaptive adversarial network, the collected images are cracked, and a binary mask and a connected domain coordinate are generated; The output image of the light intensity attenuation signal trigger region is loaded, and a noise map is generated based on the environmental parameters: ； wherein, K is an environmental scatter convolution kernel fitted based on historical data, is a convolution operation, is a noise intensity coefficient, is a random noise matrix; The generator outputs a probability mask with the noise map as input, the discriminator inputs the probability mask and the true value mask, updates the network parameters by minimizing the loss function, and finally obtains the binary mask and the connected domain coordinate by binarizing the probability mask; Based on the binary mask and the connected domain coordinate, the connected domain is divided into micro-element grids, the curvature radius and risk weight of each micro-element are calculated, and the visual parameters including crack width, length and expansion rate are calculated through a weighted formula; The boundary polygon defined by the connected domain coordinate in the mask is extracted, the polygon is pixel grid dissected, and the micro-elements across the boundary are completed by linear interpolation; Based on the following calculation formula, the curvature radius of the micro-element is calculated based on the sub-pixel edge point: ； wherein is the radius of curvature, is the edge point coordinate in the infinitesimal mesh, is the edge point tangent direction angle, is the edge bending rate of change; Based on the curvature radius, the local bending degree of the crack is quantified, and the risk weight is calculated based on the material stress sensitivity coefficient: ； wherein is the risk weight of the jth infinitesimal element, is the material stress sensitivity coefficient.

2. The bridge crack monitoring method based on artificial intelligence according to claim 1, characterized in that, The calculation expression of the crack width, length and expansion rate is: ； ； ； wherein, is the crack width, is the microelement base width, n is the microelement number, is the crack length, is the connected domain skeleton length, is the pixel physical size, is the crack propagation rate, t is the sliding window size, is the crack width of the day, is the time interval. 3.The bridge crack monitoring method based on artificial intelligence according to claim 2, characterized in that, The calculation expression of the risk index is: ； ； ； wherein, is the risk index, is the visual crack risk coefficient, is the fiber strain risk coefficient, is the environmental change rate, , , , are the visual crack risk coefficient, the fiber strain risk coefficient, the environmental change rate, and the contribution weight coefficient of the visual crack risk coefficient and the fiber strain risk coefficient coupling each other, respectively, is the width weight coefficient, is the length weight coefficient, is the expansion rate weight coefficient, , is the maximum strain value of the fiber monitoring and its weight coefficient, , represents the strain space gradient and its weight coefficient.

4. An artificial intelligence-based bridge crack monitoring system implementing the artificial intelligence-based bridge crack monitoring method of any one of claims 1 to 3, characterized by, It comprises: The acquisition module is used for identifying the principal stress region based on the bridge finite element model, laying the orthogonal optical fiber grid based on the midspan and support connection point as the benchmark, and the grid density is positively correlated with the stress gradient, and the light intensity attenuation signal is collected in real time, wherein the nodes of the orthogonal optical fiber grid are provided with vibration sensors for collecting vibration signals; The image module is used for imaging the region corresponding to the light intensity attenuation signal when the light intensity attenuation signal exceeds the threshold value, and the energy proportion of the vibration signal in the preset frequency band is greater than the preset value; The segmentation module is used for cracking the collected images based on the adaptive adversarial network, and generating a binary mask and a connected domain coordinate; The risk module is configured to divide the connected domain into a micro-element grid based on the binary mask and the connected domain coordinates, calculate the curvature radius and risk weight of each micro-element, and calculate visual parameters including the crack width, length and expansion rate through a weighting formula. The alarm module is configured to fuse the fiber strain data and the visual parameters to calculate a risk index, and output an alarm signal when the risk index is greater than an alarm threshold. 5.The bridge crack monitoring system based on artificial intelligence according to claim 4, wherein, The acquisition module is specifically configured to: detect the environmental intensity through a light sensor, and acquire crack texture features through a short exposure frame and shadow area features through a long exposure frame when the intensity gradient is greater than a preset gradient value; generate an output image based on a fusion formula: ； ； wherein is a short exposure frame, is a long exposure frame, is a fusion weight, C is a standard deviation of illumination contrast, j is a base weight coefficient, and b is a decay rate coefficient.

Images