A bridge health monitoring method and system based on image recognition

Abstract

The application relates to the technical field of bridge health monitoring, and discloses a bridge health monitoring method and system based on image recognition, which first identifies the image of a bridge to be monitored, extracts sub-pixel level displacement time history response of the slight vibration of the bridge from continuous image frames by using a discrete centroid search algorithm, obtains the modal parameters of the slight vibration of the bridge by using a Hankel dynamic modal decomposition method, establishes a modal vibration mode calculation model according to the modal parameters of the bridge, and identifies the damage position and damage degree of the bridge according to the modal vibration mode calculation model. In this way, the damage position and damage degree of the bridge can be accurately identified in a mode based on image recognition, and the safety hazards of the bridge can be found in time.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of bridge health monitoring, and in particular to a bridge health monitoring method and system based on image recognition. BACKGROUND

[0002] Bridge structure vibration measurement is the key of bridge structure health monitoring (SHM). Modal parameters (natural frequency, mode shape and damping ratio, etc.) are important indicators reflecting the health condition of the structure, which can identify the bridge structure damage and state change through the change of modal parameters, and evaluate the service performance of the bridge structure. During the operation of the bridge, vibration phenomenon will inevitably occur. How to quickly measure the small vibration response of the bridge is the premise to ensure the safe operation of the bridge, and is also a powerful guarantee to realize the real-time monitoring of the health condition of the bridge.

[0003] At present, the commonly used instrument for measuring the vibration characteristics of the bridge is the acceleration sensor, but the acceleration sensor has the disadvantages of high cost, difficult installation, limited measurement points, low measurement precision and poor real-time performance, which is difficult to meet the demand of real-time monitoring of the dynamic response of the bridge. Other conventional measurement methods such as level gauge, dial gauge and total station are difficult to perform dynamic measurement. Although GPS (Global Positioning System) can realize dynamic measurement, the debugging and installation are very tedious, and the complex bridge environment, satellite and weather factors will affect the measurement accuracy and measurement time. Most of the existing non-contact vibration measurement methods based on computer vision technology are only suitable for scenes with large structural vibration displacement amplitude, and are difficult to be applied to the measurement of small structural vibration. However, the small vibration signal of the bridge contains important information. Therefore, the existing bridge monitoring method has low precision and cannot timely discover the safety hazards existing in the bridge. SUMMARY

[0004] The present application provides a bridge health monitoring method and system based on image recognition to solve the problem of low precision of the existing bridge monitoring method and the inability to timely discover the safety hazards existing in the bridge.

[0005] In order to achieve the above-mentioned purpose, the present application realizes the technical scheme as follows:

[0006] In the first aspect, the present application provides a bridge health monitoring method based on image recognition, comprising:

[0007] obtaining a first image of a bridge to be monitored, and pre-processing the first image to obtain a second image;

[0008] spatially decomposing the second image to obtain a corresponding target image sequence;

[0009] calculate the sub-pixel level coordinates of the center of mass positions of the continuous image frames in the target image sequence, and determine the real displacement time history response of the bridge to be monitored according to the sub-pixel level coordinates of the center of mass positions;

[0010] extract bridge modal parameters including natural frequencies, mode shapes and damping ratios from the real displacement time history response by using a Hankel dynamic modal decomposition method;

[0011] establish a modal shape calculation model according to the bridge modal parameters, and identify the bridge damage location and the bridge damage degree according to the modal shape calculation model.

[0012] In a second aspect, the present application provides a bridge health monitoring system based on image recognition, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method of the first aspect when executing the computer program.

[0013] Advantages:

[0014] The bridge health monitoring method based on image recognition provided by the present application first identifies the image of the bridge to be monitored, extracts the sub-pixel level displacement time history response of the bridge micro-vibration from the continuous image frames by using a discrete center of mass search algorithm, obtains the modal parameters of the bridge micro-vibration by using a Hankel dynamic modal decomposition method, establishes a modal shape calculation model according to the bridge modal parameters, and identifies the bridge damage location and the bridge damage degree according to the modal shape calculation model. In this way, the damage location and the damage degree of the bridge can be accurately identified based on the image recognition method, and the safety hazards of the bridge can be found in time.

[0015] In a preferred scheme, the final target image sequence is obtained by denoising the initial target image sequence, which enhances the measurement accuracy and facilitates obtaining accurate real displacement time history responses and modal parameters.

[0016] In a preferred scheme, the center of mass of the discretized object in each grid is calculated by dividing each image frame in the amplified target image sequence into a grid of multiple cropping regions and using an Otsu threshold segmentation algorithm, and the sub-pixel level coordinates of the center of mass positions of all continuous image frames are calculated, so that a more accurate displacement time history response can be obtained. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 One of the flowcharts of a bridge monitoring method based on image recognition according to a preferred embodiment of the present application;

[0018] Figure 2 One of the flowcharts of a bridge monitoring method based on image recognition according to a preferred embodiment of the present application;

[0019] Figure 3 The schematic diagram of the centroid position of two continuous image frames of a preferred embodiment of the present application;

[0020] Figure 4 The schematic diagram of the principle of obtaining the real dynamic displacement time history response of the bridge micro-vibration of a preferred embodiment of the present application. DETAILED DESCRIPTION

[0021] The technical solutions of the present application will be described clearly and completely below. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all the other embodiments obtained by those skilled in the art without any creative work belong to the protection scope of the present application.

[0022] Unless otherwise defined, the technical terms or scientific terms used in the present application should be understood as the common meanings understood by those skilled in the art to which the present application belongs. The terms "first", "second" and similar terms used in the present application do not represent any order, number or importance, but are only used to distinguish different components. Similarly, the terms "one" or "a" and similar terms do not represent a quantity limitation, but represent the existence of at least one. The terms "connected" or "connected" and similar terms are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. The terms "up", "down", "left", "right" and the like are only used to represent relative positional relationships, and when the absolute positions of the described objects are changed, the relative positional relationships are also changed accordingly.

[0023] Please refer to Figures 1-2 The present application provides a bridge health monitoring method based on image recognition, comprising:

[0024] Obtaining a first image of a bridge to be monitored, and pre-processing the first image to obtain a second image;

[0025] Performing spatial decomposition on the second image to obtain a corresponding target image sequence;

[0026] Calculating the centroid position sub-pixel level coordinates of continuous image frames in the target image sequence, and determining the real displacement time history response of the bridge to be monitored according to the centroid position sub-pixel level coordinates;

[0027] Extracting bridge modal parameters from the real displacement time history response by using Hankel dynamic modal decomposition method, wherein the bridge modal parameters include natural frequency, mode shape and damping ratio;

[0028] Establishing a modal shape calculation model according to the bridge modal parameters, and identifying the bridge damage position and the bridge damage degree according to the modal shape calculation model.

[0029] In this embodiment, the first image of the bridge to be monitored can be obtained by a high-speed camera.

[0030] The bridge health monitoring method based on image recognition described above first identifies the image of the bridge to be monitored, extracts the sub-pixel level displacement time history response of the micro-vibration of the bridge from the continuous image frames by using the discrete centroid search algorithm, obtains the modal parameters of the micro-vibration of the bridge by using the Hankel dynamic modal decomposition method, establishes a modal vibration mode calculation model according to the modal parameters of the bridge, and identifies the damage location and damage degree of the bridge according to the modal vibration mode calculation model. In this way, the damage location and damage degree of the bridge can be accurately identified based on the image recognition method, and the safety hazards of the bridge can be found in time.

[0031] Optionally, the first image is pre-processed to obtain a second image, including:

[0032] The first image is rotated, cropped and scaled to obtain the second image.

[0033] In this optional embodiment, the digital image processing software can be used to rotate, crop and scale the first image for pre-processing. In this way, the noise in the image sequence can be preliminarily removed, avoiding the noise being amplified by the same proportion when the wideband phase motion amplification algorithm is used for amplification processing, and excluding the influence of noise on the image amplification result.

[0034] Optionally, the second image is spatially decomposed to obtain a corresponding target image sequence, including:

[0035] The second image is spatially decomposed to obtain an initial target image sequence, and the initial target image sequence includes a first residual part, different frequency baseband phase information and a second residual part.

[0036] The texture features of the initial target image sequence are extracted by using a two-dimensional Gabor wavelet filter, and the image information in the target image sequence other than the texture features is translated to complete the noise reduction processing of the initial target image sequence, and obtain the final target image sequence.

[0037] In this optional embodiment, the second image is spatially decomposed by using a complex steerable pyramid (CSP) to obtain an image sequence of different scales, different directions and different positions, so that the image sequence realizes the expression of local phase to local motion in the micro-vibration video of the bridge structure to be tested. It should be noted that the initial target image sequence obtained after decomposition includes a first residual part, different frequency baseband phase information and a second residual part, wherein the first residual part refers to a high-pass residual part, the second residual part refers to a low-pass residual part, and the different frequency baseband phase information refers to intermediate different frequency baseband phase information.

[0038] In the embodiment of the present application, after the initial target image sequence is obtained, the initial target image sequence needs to be further processed to reduce noise to obtain a to-be-measured image sequence. Specifically, a two-dimensional Gabor wavelet filter is used to extract the texture features of the initial target image sequence, and the image information of the initial target image sequence other than the texture features is processed by translation to complete the noise reduction processing of the initial target image sequence to obtain a final target image sequence. It should be noted that the two-dimensional Gabor wavelet filter is sensitive to the edge information of the image, can provide good direction selection and scale selection characteristics, and in addition, the two-dimensional Gabor wavelet filter can be used for spatial filtering. It can be understood that using the two-dimensional Gabor wavelet filter for noise reduction processing effectively avoids the artifacts generated by noise in motion amplification.

[0039] In the embodiment, the two-dimensional Gabor wavelet filter is a sinusoidal function modulated by a Gaussian function, and its complex expression is as follows:

[0040] (1)

[0041] The real part is as follows:

[0042] (2)

[0043] The imaginary part is as follows:

[0044] (3)

[0045] In the formula, represents the wavelength of the sinusoidal function; represents the phase shift of the tuning function; determines the space aspect ratio of the two-dimensional Gabor function; is the standard deviation of the Gaussian function, which determines the size of the acceptable region of the two-dimensional Gabor filter kernel; represents the direction of the two-dimensional Gabor wavelet filter kernel, and ; and are spatial position variables; exp represents exponential calculation; i represents the imaginary unit, .

[0046] It should be noted that and represent the direction information and spatial information of the video image sequence respectively, and satisfy the following relationship:

[0047] (4)

[0048] Optionally, before calculating the sub-pixel level coordinates of the center of mass positions of the continuous image frames in the target image sequence, the above method further comprises:

[0049] amplifying the target image sequence based on the different frequency baseband phase information and a preset amplification factor;

[0050] calculating the sub-pixel level coordinates of the center of mass positions of the continuous image frames in the target image sequence, and determining the real displacement time history response of the bridge to be monitored according to the sub-pixel level coordinates of the center of mass positions, comprising:

[0051] dividing each image frame in the amplified target image sequence into a grid of multiple clipping regions, calculating the center of mass of the discretized object in each grid using the Otsu threshold segmentation algorithm, calculating the sub-pixel level coordinates of the center of mass positions of all continuous image frames, subtracting the center of mass coordinates of the subsequent frame images starting from the second frame image from the center of mass coordinates of the first frame image to obtain the sub-pixel level displacement time history response of the bridge to be monitored, and converting the sub-pixel level displacement time history response of the bridge to be monitored into a real displacement time history response according to the scale factor.

[0052] In this embodiment, the specific steps of the amplification process are as follows:

[0053] selecting a wideband frequency baseband of a region of interest (ROI) in the intermediate different frequency baseband phase information , and setting a suitable amplification factor As a preset amplification factor, the selected wideband frequency baseband is amplified by the preset amplification factor to achieve amplification processing of the small vibration amplitude of the frequency baseband of interest. The wideband frequency baseband after amplification processing is added back to the high-pass residual part and the low-pass residual part image sequence, and the amplified image sequence is reconstructed and output using a complex direction controllable pyramid to output the amplified video. Among them, is the low frequency cutoff frequency of the selected wideband frequency baseband; is the high frequency cutoff frequency of the selected wideband frequency baseband.

[0054] The principle of the wideband phase motion amplification algorithm is introduced as follows. Assuming that the spatial signal of bridge vibration is , is the small vibration displacement occurring in the time domain, and the preset amplification factor is used to obtain the amplified spatial signal of bridge vertical vibration . For wideband phase motion amplification, the spatial signal of bridge vibration is first expressed as a superposition of a series of complex sinusoidal signals, as shown in equation (5).

[0055] ; (5)​

[0056] where, is the circular frequency of a certain sub-sine signal; is the amplitude of a certain sub-sine signal; is the spatial position variable.

[0057] For a certain complex sine signal with circular frequency , we have:

[0058] ; (6)

[0059] denotes a complex sine signal with phase , and we perform time-domain filtering on this signal to filter out , we obtain:

[0060] ; (7)

[0061] denotes the phase difference.

[0062] Using the wideband phase motion amplification algorithm, we amplify times and add back , obtaining the result of equation (8)

[0063] ; (8)

[0064] denotes the amplified complex sine signal, and we add the amplified signal back to the original signal, obtaining the wideband phase motion amplification result under the wideband frequency baseband . Thus, we obtain the amplified target image sequence. Further, we use the discrete centroid search algorithm to extract the dynamic displacement time history response of the bridge structure micro-vibration from the amplified target image sequence, as follows:

[0065] Divide the amplified bridge vibration image frame into a grid of multiple cropping areas, and use the Otsu threshold segmentation algorithm to calculate the centroid of the discretized object in each grid. Calculate the sub-pixel level coordinates of the centroid positions of all consecutive image frames, and subtract the centroid coordinates of the subsequent frame images starting from the second frame image from the centroid coordinates of the first frame image, i.e., we obtain the sub-pixel level displacement time history response of the bridge vibration. The centroid positions of two consecutive image frames are shown in

[0066] Figure 3

[0067] ​​​It is worth pointing out that the discrete centroid search algorithm in the present application has the advantages of small interference from image background noise, strong real-time performance and high displacement accuracy compared with Digital Image Correlation (DIC), optical flow method, edge detection algorithm and target tracking algorithm based on deep learning. In the present embodiment, the discrete centroid search algorithm can obtain more accurate displacement time history response.

[0068] Further, after obtaining the sub-pixel level dynamic displacement time history of the bridge micro-vibration, the sub-pixel level dynamic displacement time history can be converted into a physical displacement time history response according to the scale factor.

[0069] Specifically, the calculation method of the scale factor is different in different cases. When the camera optical axis is perpendicular to the bridge structure plane, i.e. the optical axis is collinear with the normal of the structure plane, s The scale factor is shown in formula (9) or formula (10).

[0070] ; (9)

[0071] Or:

[0072] ; (10)

[0073] In the formula, D is the size of the selected object in the structure plane; d is the corresponding pixel number of the object in the image plane; f is the focal length of the lens; Z is the distance from the camera to the structure plane; is the pixel size.

[0074] When the camera optical axis is not perpendicular to the bridge structure plane, i.e. the optical axis has an angle with the normal of the structure plane , the scale factor s is shown in formula (11).

[0075] ; (11)

[0076] When the wideband phase motion amplification algorithm is used to amplify the micro-vibration of the bridge structure, the micro-vibration displacement amplitude of the bridge is amplified by times, and the physical displacement time history response extracted by the discrete centroid search algorithm is not the true displacement time history response of the bridge micro-vibration. Assuming is the true displacement of the bridge micro-vibration, is the displacement amplitude of the bridge micro-vibration, is the displacement recognition error caused by video illumination change noise. Then the displacement of the structure without amplification processing is The displacement after amplifying the minute vibrations of the bridge is For the magnified displacement The true dynamic displacement time history response of a bridge's minor vibrations can be obtained by performing motion normalization. The principle for obtaining the true dynamic displacement time history response of a bridge's minor vibrations is as follows: Figure 4 As shown in equation (12), it can be seen that broadband phase motion amplification processing can effectively reduce the impact of noise in the video on the identification of small vibration displacements of the bridge.

[0077] ;(12)

[0078] Furthermore, the Hankel dynamic modal decomposition method is used to extract bridge modal parameters (natural frequencies, mode shapes, and damping ratios) from the bridge's actual dynamic displacement time history response. Specifically, the Hankel dynamic modal decomposition method can be used to extract the bridge's modal parameters (natural frequencies, mode shapes, and damping ratios) from the bridge's actual dynamic displacement time history response. Constructing the Hankel matrix As shown in equation (13).

[0079] ; (13)

[0080] In the formula, m , n , p Both M and M are positive integers. Indicates the first k The point is at The dynamic displacement at any given moment.

[0081] The specific steps for extracting the modal parameters of a bridge using the Hankel dynamic modal decomposition method are as follows:

[0082] (1) Calculate the scaling factor Its calculation formula satisfies the following relationship:

[0083] ;(14)

[0084] In the formula, where It is a matrix The last column, yes The first sub-block of the last column.

[0085] (2) The Hankel matrix is ​​composed of equation (15).

[0086] (15)

[0087] In the formula, For the same Hankel matrix, move it forward in time; and This represents two adjacent time data matrices.

[0088] (3) X The truncated singular value decomposition (SVD) is calculated as shown in Equation (16).

[0089] ; (16)

[0090] In the formula, and These are two types of unitary matrices obtained from singular value decomposition. for The adjoint matrix; It is a diagonal matrix with diagonal elements as follows: A singular value, It is a positive integer.

[0091] (4) Operator matrix As shown in equation (17).

[0092] ; (17)

[0093] In the formula, for The adjoint matrix; for The conjugate matrix; for The inverse matrix.

[0094] (5) The eigenvalues ​​and eigenvectors are and ,and .

[0095] (6) The natural frequency and damping ratio of the bridge are calculated by equation (18).

[0096] (18)

[0097] In the formula, The sampling time represents the time history response of the bridge's dynamic displacement. Represents the natural frequencies of the bridge; Indicates the damping ratios of different orders of a bridge; This represents the modal order and is a positive integer.

[0098] (7) Then, the vibration mode of the bridge can be identified by equation (19).

[0099] ; (19)

[0100] Optionally, the bridge damage position and the bridge damage degree are identified according to the modal shape calculation model, and the method comprises the following steps of:

[0101] The modal shape normalization processing is performed on each order modal shape of the bridge, and the absolute values are superimposed, the average modal shape energy of the bridge without damage and the average modal shape energy of the bridge with damage are respectively calculated, the average energy difference of each point on the bridge is calculated according to the average modal shape energy, and the bridge damage position and the bridge damage degree are identified according to the average energy.

[0102] Optionally, the bridge damage position and the bridge damage degree are identified according to the average energy, and the method comprises the following steps of:

[0103] A fitting curve is generated according to the average energy.

[0104] The position of the mutation of the fitting curve is determined as the damage position of the bridge, and the damage degree of the bridge structure is determined by judging the amplitude of the singular point of the fitting curve.

[0105] In the optional embodiment, the modal shape superposition energy algorithm is used to identify the damage position and the damage degree of the bridge. First, the modal shape normalization processing is performed on each order modal shape of the bridge, and the absolute values are superimposed, as shown in formula (20).

[0106] ; (20)

[0107] In the formula, is the superposition of the first order modal shape of the bridge without damage. N is the superposition of the first order modal shape of the bridge without damage.

[0108] ; (21)

[0109] In the formula, is the superposition of the first order modal shape of the bridge without damage. N is the superposition of the first order modal shape of the bridge without damage. is the superposition of the first order modal shape of the bridge without damage.

[0110] Secondly, the average modal shape energy of the bridge without damage and the average modal shape energy of the bridge with damage are respectively calculated, as shown in formula (22) and (23).

[0111] ; (22)

[0112] ; (23)

[0113] In the formula, is the superposition of the first order modal shape of the bridge without damage. naverage modal shape energy value of the point, for the bridge when there is damage n average modal shape energy value of the point.

[0114] Finally, the average energy difference of each point on the bridge is calculated, as shown in equation (24).

[0115] (24)

[0116] By observing the mutation position of the fitting curve is the damage position of the bridge. The damage degree of the bridge structure is determined by judging the amplitude of the singular point. In this way, the damage position and damage degree of the bridge can be quickly identified through the fitting curve.

[0117] Optionally, after identifying the bridge damage position and bridge damage degree according to the modal shape calculation model, the above method further comprises:

[0118] According to the bridge damage position and bridge damage degree, the safety of the bridge to be monitored is evaluated, and in the case that the damage position of the bridge is determined, or in the case that the damage degree of the bridge exceeds the damage threshold, an alarm information is generated.

[0119] In the optional embodiment, the bridge damage position and damage degree obtained by the above image processing technology are combined with the historical data of the health condition of the bridge to perform real-time evaluation on the operation safety of the bridge. Based on the broadband phase motion amplification algorithm and the discrete centroid search algorithm, the micro-vibration of the bridge can be monitored in real time, and through the real-time monitoring data, the operation safety, degradation behavior and remaining life of all bridges in the entire region can be evaluated, which can provide data support for the health monitoring, maintenance and daily maintenance of the bridge.

[0120] The embodiment of the present application also provides a bridge monitoring system based on image recognition, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the computer program. The bridge monitoring system based on image recognition can implement each embodiment of the above method and achieve the same beneficial effects, which will not be described here.

[0121] The embodiment of the present application also provides a computer readable storage medium having a computer program stored thereon, wherein the program is executed by a processor to implement the steps of the above method. The readable storage medium can implement each embodiment of the above method and achieve the same beneficial effects, which will not be described here.

[0122] The preferred embodiments of the present application have been described above in detail. It should be understood that modifications and variations to the present application can be affected by those skilled in the art without departing from the scope of the application. Accordingly, it is intended that all of the subject matter of the above description and the claims be interpreted to encompass all such modifications and changes.

Claims

1. An image recognition-based bridge health monitoring method, characterized by, The method comprises the following steps: obtaining a first image of a bridge to be monitored, and pre-processing the first image to obtain a second image; spatially decomposing the second image to obtain a corresponding target image sequence; calculating the sub-pixel level coordinates of the center of mass of continuous image frames in the target image sequence, and determining the real displacement time history response of the bridge to be monitored according to the sub-pixel level coordinates of the center of mass; extracting bridge modal parameters from the real displacement time history response by using a Hankel dynamic modal decomposition method, wherein the bridge modal parameters include natural frequency, mode shape and damping ratio; establishing a modal mode shape calculation model according to the bridge modal parameters, and identifying the damage location and damage degree of the bridge according to the modal mode shape calculation model; the spatial decomposition of the second image to obtain the corresponding target image sequence comprises: spatially decomposing the second image to obtain an initial target image sequence, wherein the initial target image sequence comprises a first residual part, different frequency baseband phase information and a second residual part; extracting the texture features of the initial target image sequence by using a two-dimensional Gabor wavelet filter, and performing translation processing on the image information in the target image sequence except the texture features to complete the noise reduction processing of the initial target image sequence, thereby obtaining the final target image sequence; before the calculation of the sub-pixel level coordinates of the center of mass of continuous image frames in the target image sequence, the method further comprises: performing amplification processing on the target image sequence based on the different frequency baseband phase information and a preset amplification factor; the calculation of the sub-pixel level coordinates of the center of mass of continuous image frames in the target image sequence, and the determination of the real displacement time history response of the bridge to be monitored according to the sub-pixel level coordinates of the center of mass comprises: dividing each image frame in the amplified target image sequence into a grid of multiple clipping regions, calculating the center of mass of the discretized object in each grid by using an Otsu threshold segmentation algorithm, calculating the sub-pixel level coordinates of the center of mass of all continuous image frames, subtracting the center of mass coordinates of the subsequent frame images starting from the second frame image from the center of mass coordinates of the first frame image to obtain the sub-pixel level displacement time history response of the bridge to be monitored, and converting the sub-pixel level displacement time history response of the bridge to be monitored into a real displacement time history response according to a scale factor; the identification of the damage location and damage degree of the bridge according to the modal mode shape calculation model comprises: performing mode shape normalization processing on each modal mode shape of the bridge, superimposing the absolute values thereof, calculating the average modal mode shape energy of the bridge when the bridge is undamaged and damaged respectively, calculating the average energy difference of each point on the bridge according to the average modal mode shape energy, and identifying the damage location and damage degree of the bridge according to the average energy; the identification of the damage location and damage degree of the bridge according to the average energy comprises: generating a fitting curve according to the average energy; determining the position of the mutation of the fitting curve as the damage location of the bridge, and determining the damage degree of the bridge structure by judging the amplitude of the singular point of the fitting curve.

2. The image recognition based bridge health monitoring method of claim 1, wherein, the pre-processing of the first image to obtain the second image comprises: The first image is rotated, cropped and scaled to obtain the second image.

3. The image recognition based bridge health monitoring method of claim 1, wherein, After the bridge damage position and the bridge damage degree are identified according to the modal vibration mode calculation model, the method further comprises: According to the bridge damage position and the bridge damage degree, a safety evaluation is performed on a bridge to be monitored, and an alarm information is generated when the damage position of the bridge is determined or when the bridge damage degree exceeds a damage threshold.

4. An image recognition based bridge health monitoring system comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor implements the steps of the method of any one of claims 1 to 3 when executing the computer program.

Images