Bridge underwater safety detection method based on opto-acoustic dual mode

Abstract

The invention discloses a bridge underwater safety detection method based on opto-acoustic dual modes, and relates to the technical field of underwater engineering detection, and the method comprises the following steps: obtaining water area particle size distribution and concentration data by using a multi-frequency acoustic suspended sediment concentration profiler, establishing an environmental parameter database, configuring a shipborne detection device according to the initial environmental parameter database, and detecting the underwater safety of a bridge according to the shipborne detection device. The method comprises the following steps: setting a three-axis stabilized platform carrying a polarization imaging camera and a multispectral LED array, deploying a multi-beam detection module and ADCP equipment, and establishing a unified coordinate system; multi-angle original intensity images are collected through a polarized optical imaging module, and multi-exposure fusion is carried out after processing is carried out through a mutual information minimization algorithm. An optical image and an acoustic point cloud are deeply fused by adopting a graph neural network, the detection precision is improved through dynamic weight optimization of an attention mechanism and a space-time registration algorithm, a multi-index fusion model is constructed based on fuzzy logic and a D-S evidence theory for adaptive early warning decision, upgrading from single-parameter alarm to multi-dimensional decision is realized, and the false alarm rate is reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of underwater engineering detection, and particularly relates to a bridge underwater safety detection method based on a photoacoustic bimodule. BACKGROUND

[0002] In recent years, underwater bridge safety detection technology has made significant progress. Among them, the acoustic detection method is widely used because of its characteristics of not being affected by water quality. High-frequency sound waves are used for underwater scanning, three-dimensional point cloud data of underwater structures are obtained through beam forming algorithm, and point cloud processing technology is used to identify abnormal deformation of the structure surface. This technology can achieve centimeter-level resolution, and the detection accuracy of bridge pile foundation scour pits in clear water environment can reach ±2cm. In the aspect of optical imaging technology, the target is irradiated at different angles by controlling the polarized light source, and after a plurality of groups of polarized images are collected, image enhancement is carried out by using Stokes vector calculation, which effectively improves the imaging quality in turbid water. These two technologies have achieved good application effect in the fields of acoustic and optical detection respectively.

[0003] Although the existing technology has achieved certain results in underwater detection, there is still room for optimization in the collaborative processing of multi-modal data, especially in complex hydrological environment. The detection results of a single sensing mode are often difficult to fully reflect the real state of the underwater structure. For example, although the multi-beam sonar detection can accurately obtain the geometric deformation of the structure surface, it has limited ability to identify early fine cracks. Although the optical imaging technology is sensitive to surface defects, the imaging quality will decrease significantly in high turbidity water. The existing technology lacks a deep fusion mechanism for acoustic and optical data, which leads to the fact that the environmental parameters such as flow field and sediment concentration cannot establish an effective correlation analysis model with the detection results of structure defects, thereby affecting the accuracy of the overall detection results and the timeliness of the early warning. SUMMARY

[0004] In view of the above existing problems, the present application is proposed.

[0005] Therefore, the present application provides a bridge underwater safety detection method based on a photoacoustic bimodule, which solves the technical problems of insufficient fusion of acoustic and optical data and poor correlation between environmental parameters and defect detection in the prior art.

[0006] To solve the above technical problems, the present application provides the following technical solutions: In a first aspect, the present application provides a bridge underwater safety detection method based on a photoacoustic bimodule, which comprises: using a multi-frequency acoustic suspended sediment concentration profiler to obtain particle size distribution and concentration data of the water area, establishing an environmental parameter database, configuring a shipborne detection device according to the initial environmental parameter database, setting a three-axis stabilization platform carrying a polarization imaging camera, a multi-spectral LED array, deploying a multi-beam detection module and an ADCP device, and establishing a unified coordinate system. The multi-angle original intensity image is collected by the polarized optical imaging module, the mutual information minimization algorithm is used for processing, and then multi-exposure fusion is carried out, the enhanced image is output through feature pyramid network fusion and intelligent tone mapping; Based on the control multi-beam detection module, the time delay sum beam forming algorithm is used for coarse scanning, the MVDR fine scanning is started in the scouring area, and the ADCP device is used to measure the flow velocity profile based on the Doppler frequency offset; The optical image and the acoustic point cloud are fused by a graph neural network to construct a three-dimensional model, surface defects are identified and scour pit characteristics are calculated, and a detection result is obtained; The detection result is converted into engineering visualization, a realistic three-dimensional model is generated and superimposed with a flow velocity vector field and a sediment concentration distribution, a multi-index fusion early warning decision is executed, a fuzzy logic decision model is used to process detection data, a dynamic early warning threshold is set, and an early warning report is generated.

[0007] As a preferred scheme of the bridge underwater safety detection method based on the photoacoustic dual mode, wherein: a multi-frequency acoustic suspended sediment concentration profiler is used to obtain the particle size distribution and concentration data of the water area, and an environmental parameter database is established, including the following steps, The multi-frequency acoustic suspended sediment concentration profiler is started, and the electronic element is preheated for thirty minutes for stabilization, four frequency bands are configured, the pulse width, interval and transmission power are set, the reference echo signals of each frequency band are collected in the sediment-free distilled water, and the noise floor is recorded; The built-in CTD sensor is used to measure the water temperature, salinity and depth, and the sound velocity profile is calculated; According to the spiral scanning path, control instructions are generated to guide the motion trajectory of the transducer array, long-short pulse combinations are transmitted in order of frequency priority, acoustic detection signals are generated, 16-bit high-precision ADC sampling rate is used to receive echoes of each frequency band, and time gain compensation is applied to amplify weak signals; Based on the modified mixed scattering model, the scattering intensity of each frequency band at different depth units is obtained, an intensity-depth matrix is established, the multi-frequency simultaneous equations are solved by Tikhonov regularization, the concentration distribution of fine, medium and coarse particle sizes is output, bubble interference is identified using high-frequency data, abnormal measurement points are removed by SVM classifier and sliding window average is performed, and an environmental parameter database is established.

[0008] As a preferred scheme of the bridge underwater safety detection method based on the photoacoustic dual mode, wherein: according to the initial environmental parameter database, a shipborne detection device is configured, a three-axis stabilization platform carrying a polarization imaging camera and a multi-spectral LED array are set, a multi-beam detection module and an ADCP device are deployed, and a unified coordinate system is established, including the following steps, Read turbidity, water depth and bottom material indicators from the initial environmental parameter database, generate device configuration basic parameter table, calculate light attenuation coefficient according to turbidity data, get each waveband penetration ability evaluation report, calculate platform compensation angle based on ship IMU real-time attitude data; According to the turbidity grade, the working mode is automatically selected, the polarization angle sequence is generated, the multi-spectral LED output proportion is set according to the optical evaluation report, and the illumination scheme is obtained; According to the water depth data, the working frequency is automatically selected, the frequency configuration instruction is obtained, the pulse length is calculated based on the flow velocity range and the depth, and the profile measurement scheme is output; Through the PTPv2 protocol, all devices are synchronized to the microsecond level, unified timestamp data is obtained, and the ENU geodetic coordinate system is established.

[0009] As a preferred scheme of the bridge underwater safety detection method based on the photoacoustic dual mode, wherein: the multi-angle original intensity image is collected by the polarized optical imaging module, the mutual information minimization algorithm is used for processing, and then multi-exposure fusion is carried out, the enhanced image is output through feature pyramid network fusion and intelligent tone mapping, including the following steps, Based on the proportion of the control multi-spectral LED array, a stable illumination environment is provided for image acquisition, the polarizer is rotated to 0°, 45°, 90° and 135° in turn, a polarization angle control sequence is generated, and the camera is triggered to collect images at three exposure times at each polarization angle, to obtain the original intensity image; The mutual information minimization algorithm is used to calculate the separation degree of target information light and background scattered light in each polarization image; The mutual information values of different polarization angle combinations are compared to obtain the de-scattering intermediate image, the local contrast of each exposure image is calculated, and a quality score map is generated; Based on the quality score, the softmax function is used to calculate the pixel-level weight of each exposure image, and a weight distribution matrix is output; A 5-layer Laplacian pyramid is constructed, different exposure images are fused on each pyramid layer according to the weight map, and an HDR fusion image is output.

[0010] As a preferred scheme of the bridge underwater safety detection method based on the photoacoustic dual mode, wherein: based on the control multi-beam detection module, the time delay sum beam forming algorithm is used for coarse scanning, the MVDR fine scanning is started in the scouring area, and the ADCP device is used to measure the flow velocity profile based on the Doppler frequency offset, including the following steps, According to the water depth measurement data, the working frequency is automatically selected, the pulse length and the beam width parameters are set, the phase consistency detection of the transducer array is performed, the time delay deviation of each array element is corrected, the array element time delay is obtained based on each beam direction, the output beam signal is obtained by using the Hanning window weighting, the initial terrain grid is generated based on the beam sounding data through the sound ray tracing algorithm; The second derivative of the grid terrain is calculated based on the initial terrain grid, the area with curvature is marked as a suspected scour area, the abnormal area coordinates are obtained, and the MVDR scanning path is automatically planned with the abnormal area coordinates as the center; The snapshot data is collected in the suspected scour area, the diagonal loading is performed, the MVDR optimization problem is solved, and the high-resolution beam response is output; The frequency offset of the ADCP four-beam receiving signal is calculated, and the flow velocity of each depth unit is obtained; The trailing edge of the echo is detected, the bottom velocity is calculated, and the water body flow velocity is corrected; The MVDR scanning data is interpolated to the initial terrain grid, the terrain model is generated through the Kriging interpolation, the ADCP flow velocity data is interpolated to the terrain grid according to the inverse distance weight, and the flow velocity profile is obtained.

[0011] As a preferred scheme of the bridge underwater safety detection method based on the photoacoustic dual mode, wherein: the optical image and the acoustic point cloud are fused through a graph neural network to construct a three-dimensional model, surface defects are identified and scour pit features are calculated to obtain a detection result, including the following steps, Based on the PTP timestamp and the RTK positioning data, the optical image pixel coordinates are converted into a world coordinate system unified with the acoustic point cloud through a perspective transformation matrix, the time deviation and spatial offset are compensated, the ResNet-50 network is used to extract the enhanced optical image feature vector, and the spatial attention mechanism is applied to highlight the defect area; The PointNet++ network is used to process the acoustic point cloud data to obtain a geometric feature vector to calculate the local curvature to enhance the terrain features; The optical feature pixel points and the acoustic point cloud clusters are taken as two types of nodes to construct a graph structure of spatial position and feature attribute, a node connection relationship is established based on spatial proximity and feature similarity dual criteria to form a dynamic graph topology, three-layer graph attention convolution is performed, and photoacoustic feature depth fusion is performed; Based on the fusion features, the crack probability is calculated, and a crack threshold is set to determine the defect evaluation corrosion degree; The volume of the grid terrain is calculated, fuzzy logic rules are established for risk assessment, a visual model with defect labels and risk levels is generated, and a detection result is obtained.

[0012] As a preferred scheme of the bridge underwater safety detection method based on the photoacoustic dual mode, the detection result is converted into engineering visualization, a real three-dimensional model is generated, and a flow velocity vector field and a sediment concentration distribution are superimposed, including the following steps, The optical image, the acoustic point cloud, the flow velocity profile and the sediment concentration data are spatially aligned through a unified ENU coordinate system to obtain a time-space synchronous standardized data set, a water-tight triangular mesh model is generated based on the acoustic point cloud data by applying an improved Poisson reconstruction algorithm, a three-dimensional surface with a topological structure is obtained, and the enhanced optical image is projected onto the three-dimensional mesh surface to generate a model with realistic texture; The discrete ADCP measurement point data is converted into a continuous three-dimensional flow velocity field by using the inverse distance weighting method, and the Runge-Kutta fourth-order method is used for numerical solution to generate flow line trajectories representing the water flow dynamics; A three-dimensional sediment concentration distribution field is constructed based on the Kriging interpolation method, a depth buffer technology is used to realize accurate superposition of the three-dimensional model, the flow field and the sediment field, a Phong lighting model is applied to enhance the realism and detail recognition of the scene, and a three-dimensional measurement tool is embedded in the visualization interface to establish a data-visualization linkage pipeline, thereby realizing automatic updating and rendering of new detection results.

[0013] As a preferred scheme of the bridge underwater safety detection method based on the photoacoustic dual mode, multi-index fusion early warning decision is performed, a fuzzy logic decision model is established to process the detection data, a dynamic early warning threshold is set and an early warning report is generated, including the following steps, Crack propagation rate, corrosion depth, local scour depth and flow velocity gradient core indicators are extracted from the detection results to generate a standardized parameter matrix, a trapezoidal is defined for each indicator, fuzzy rules for each parameter are established, and an IF-THEN rule set is established; The base threshold is adjusted according to real-time hydrological data, the sensitivity is automatically improved during the flood period, the dynamic threshold is calculated based on the extreme value statistical theory, and the adaptive early warning line is output; The Mamdani reasoning method is used to obtain a comprehensive risk value through sup-min composite operation, the output risk value is mapped to a four-level early warning system to generate a preliminary early warning conclusion, the D-S evidence theory is applied to integrate optical and acoustic multi-dimensional evidence, the risk value, key parameter curve, disposal suggestion and other elements are combined based on a template engine to generate an early warning report.

[0014] In a second aspect, the present application provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, any step of the bridge underwater safety detection method based on the photoacoustic dual mode according to the first aspect of the present application is realized.

[0015] In a third aspect, the present application provides a computer readable storage medium having stored thereon a computer program, wherein the computer program, when executed by a processor, implements any step of the bridge underwater safety detection method based on photoacoustic dual-mode as claimed in the first aspect of the present application.

[0016] The present application has the advantages that: the optical image and the acoustic point cloud are deeply fused by using the graph neural network, the weight is dynamically optimized by the attention mechanism, the detection accuracy is improved by the spatio-temporal registration algorithm, the multi-index fusion model is constructed based on the fuzzy logic and the D-S evidence theory for the adaptive early warning decision, the upgrade from single parameter alarm to multi-dimensional decision is realized, the false alarm rate is reduced, the dynamic threshold mechanism makes the early warning lead time increase by two times, the optical parameters are optimized in real time by the multi-frequency acoustic inversion in the environment perception layer, the detection layer combines the MVDR beam forming and the mutual information minimization algorithm, and the three-dimensional visualization and intelligent early warning are realized in the decision layer. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0018] Fig. 1 It is a flow chart of the bridge underwater safety detection method based on photoacoustic dual-mode.

[0019] Fig. 2 It is a schematic diagram of the flow velocity profile.

[0020] Fig. 3 It is a schematic diagram of multi-beam coarse scanning. DETAILED DESCRIPTION

[0021] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings of the specification.

[0022] In the following description, many specific details are set forth in order to provide a thorough understanding of the present application, but the present application can also be implemented in other ways different from those described herein, and those skilled in the art can make similar generalizations without departing from the connotation of the present application, therefore the present application is not limited to the specific embodiments disclosed below.

[0023] Secondly, the embodiment or embodiments referred to herein can include specific features, structures or characteristics contained in at least one implementation of the present application. In different places in this specification, one embodiment does not refer to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive with other embodiments.

[0024] Reference Figs. 1-3 This is one embodiment of the present invention, which provides a method for underwater safety inspection of bridges based on photoacoustic dual-mode, including the following steps: S1. Use a multi-frequency acoustic suspended sediment concentration profiler to obtain particle size distribution and concentration data of water areas, and establish an environmental parameter database.

[0025] S1.1 Start the multi-frequency acoustic suspended sediment concentration profiler, perform 30 minutes of electronic component stabilization and preheating, configure the four-band working mode, set the pulse width, interval, and transmission power, collect the reference echo signals of each frequency band in sediment-free distilled water, and record the noise background.

[0026] Furthermore, after starting the multi-frequency acoustic suspended sediment concentration profiler, a 30-minute stabilization and preheating process is first performed on the electronic components to ensure that the sensors and circuits reach their optimal working state. After the preheating is completed, the four-band operating mode is configured (e.g., 600kHz, 1.2MHz, 1.95MHz, and 3.6MHz), and the pulse width (e.g., 0.2ms), pulse interval (e.g., 50ms), and transmit power (e.g., 600W) are set. Subsequently, reference echo signals of each frequency band are collected in sediment-free distilled water, and the system noise floor is recorded by a 16-bit high-precision ADC.

[0027] S1.2. Use the built-in CTD sensor to measure water temperature, salinity, and depth, and calculate the sound velocity profile.

[0028] Specifically, the expression is, ; in, For depth The speed of sound at that location, Water temperature For practical salinity.

[0029] S1.3. Based on the spiral scanning path, control commands are generated to guide the movement trajectory of the transducer array. Long-short pulse combinations are emitted in order of frequency priority to generate acoustic detection signals. The echoes of each frequency band are received through a 16-bit high-precision ADC sampling rate, and the weak signals are amplified by applying time gain compensation.

[0030] Specifically, the expression is, ; in, Time of sound wave propagation The gain compensation value, The sound wave attenuation coefficient, This is the sound wave propagation time.

[0031] S1.4 Based on the modified mixed scattering model, the scattering intensity of each frequency band at different depth cells is obtained, the intensity-depth matrix is ​​established, and the multi-frequency simultaneous equations are solved by Tikhonov regularization to output the concentration distribution of fine, medium and coarse particle sizes. Bubble interference is identified using high-frequency data, and abnormal measurement points are removed by SVM classifier and averaged by sliding window to establish an environmental parameter database.

[0032] Specifically, the expression is, ; in, The feature vector of the sample to be classified. For the first Support vectors, For the first The class labels of the support vectors. To determine the hyperplane bias.

[0033] Specifically, the expression is, ; in, For a moment The weighted average, To adjust the sliding window size, These are the original observations. It is a forgetting factor.

[0034] S2. Configure shipborne detection equipment based on the initial environmental parameter database, set up a three-axis stabilized platform equipped with a polarization imaging camera, a multispectral LED array, deploy a multi-beam detection module and ADCP equipment, and establish a unified coordinate system.

[0035] S2.1 Read turbidity, water depth and bottom sediment indicators from the initial environmental parameter database, generate a basic parameter table for equipment configuration, calculate the light attenuation coefficient based on the turbidity data, obtain a penetration capability assessment report for each band, and calculate the platform compensation angle based on the ship's IMU real-time attitude data.

[0036] Specifically, the expression is, ; in, The current attitude angle, The rate of change of attitude angle, This is the proportional gain coefficient. The differential gain coefficient, To compensate for the angle, For time derivative.

[0037] S2.2, automatically select the working mode according to the turbidity level, generate the polarization angle sequence, set the multi-spectral LED output ratio according to the optical evaluation report, and obtain the illumination scheme.

[0038] Further, based on the turbidity data in the initial environmental parameter database, the water environment is divided into three levels of low turbidity (e.g. turbidity < 50 NTU), medium turbidity (e.g. turbidity 50-200 NTU) and high turbidity (e.g. turbidity > 200 NTU); according to the turbidity level, the working mode of the polarization imaging camera is automatically selected, the low turbidity environment adopts the configuration of 1080p resolution and 30fps frame rate, and the high turbidity environment adopts the configuration of 720p resolution and 15fps frame rate; four groups of standard polarization angle sequences of 0°, 45°, 90° and 135° are generated, and three exposure images of underexposure, normal and overexposure are synchronously collected for each polarization angle; referring to the waveband penetration ability analysis result in the optical evaluation report, the output ratio of white light, blue light and ultraviolet light of the multi-spectral LED array is set (e.g. 70% white light, 20% blue light and 10% ultraviolet light in low turbidity environment, and 30% white light, 60% blue light and 10% ultraviolet light in high turbidity environment), and finally the complete illumination scheme containing polarization parameters, exposure parameters and spectral combination is output. S2.3, automatically select the working frequency according to the water depth data, obtain the frequency configuration instruction, calculate the pulse length based on the flow velocity range and the depth, and output the profile measurement scheme.

[0039] Specifically, the expression is, ; wherein, is the two-way propagation distance, is the pulse length, is the maximum flow velocity range; S2.4, achieve microsecond-level time synchronization of all devices through PTPv2 protocol, obtain unified timestamp data, and establish ENU geodetic coordinate system.

[0040] Further, microsecond-level time synchronization of the polarization optical imaging camera, the multi-beam detection module and the ADCP device is achieved through the PTPv2 precise time protocol, the master-slave clock architecture and the boundary clock mechanism are adopted to control the time deviation of each device within 1 μs; based on the absolute position and attitude data obtained by the GPS receiver and the inertial measurement unit, the ENU (East-North-Sky) geodetic coordinate system is established, the multi-source sensor data is uniformly converted to the coordinate system through the quaternion conversion matrix, and the accurate unification of the time and space reference is realized.

[0041] S3, collect multi-angle original intensity images through the polarization optical imaging module, process them through the mutual information minimization algorithm, perform multi-exposure fusion, and output enhanced images through feature pyramid network fusion and intelligent tone mapping.

[0042] S3.1, based on the control of the proportion of the multi-spectral LED array, provide stable lighting environment for image acquisition, drive the polarization plate to rotate to 0°, 45°, 90°, 135° four standard angles in turn, generate polarization angle control sequence, at each polarization angle, trigger the camera three exposure time to collect images, get the original intensity image.

[0043] Further, based on the multi-spectral LED array output proportion (for example, white light 70%, blue light 20%, ultraviolet light 10%) determined by the optical evaluation report, configure the LED drive current to provide a stable lighting environment; by stepping motor precise control polarization plate to rotate to 0°, 45°, 90°, 135° four standard angles in turn, angle positioning accuracy to ± 0.5°; at each polarization angle position, trigger the camera to collect image sequence with underexposure (for example, 1 / 100s), normal (for example, 1 / 60s), overexposure (for example, 1 / 30s) three exposure time; record the GPS time stamp, IMU attitude data and depth sensor reading of each frame of image, finally output the original intensity image containing multi-angle polarization state and multi-exposure dynamic range.

[0044] S3.2, use mutual information minimization algorithm to calculate the separation degree of target information light and background scattered light in each polarization image.

[0045] Specifically, the expression is, ; Wherein, is the background scattered light signal set, is the target information light signal set, is the joint probability distribution of background and target, is the marginal probability of background signal, is the marginal probability of target signal.

[0046] S3.3, compare the mutual information values of different polarization angle combinations to get the despeckling intermediate image, calculate the local contrast of each exposure image, and generate a quality score map.

[0047] Specifically, the expression is, ; Wherein, is the image pixel coordinate position, is the maximum pixel intensity value in the local window with as the center, is the minimum pixel intensity value in the local window with as the center S3.4, based on the quality score, use softmax function to calculate the pixel level weight of each exposure image, and output the weight distribution matrix.

[0048] Specifically, the expression is, ; wherein, is a fusion weight of the i-th exposure image at the pixel, is a quality evaluation score of the i-th exposure image at the pixel, is to amplify the quality difference between different exposure images, is an exposure image.

[0049] S3.5, a 5-layer Laplacian pyramid is constructed, and different exposure images are fused on each layer of the pyramid according to the weight map to output an HDR fusion image.

[0050] Specifically, the expression is, ; wherein, is a Laplacian pyramid coefficient of the i-th layer, is a Gaussian pyramid image of the i-th layer; S3.6, according to the image histogram, the pixels are divided into dark, intermediate and bright regions, a non-sharpening mask algorithm is used to enhance high-frequency details to obtain an enhanced image.

[0051] Specifically, the expression is, ; wherein, is an enhanced image, is an original input image, is a blurred image; S4, based on the control multi-beam detection module, a time delay summation beam forming algorithm is used for coarse scanning, and an MVDR fine scanning is started in the flushing area, and an ADCP device is used to measure the flow velocity profile based on the Doppler frequency offset.

[0052] S4.1, according to the water depth measurement data, the working frequency is automatically selected, the pulse length and beam width parameters are set, the phase consistency detection of the transducer array is performed, the time delay deviation of each array element is corrected, the array element time delay is obtained based on each beam direction, the Hanning window is used for weighting to obtain the output beam signal, and based on the beam sounding data, the initial terrain grid is generated through the sound ray tracing algorithm.

[0053] Specifically, the expression is, ; wherein, is a true distance corrected by a sound velocity profile, ​​​​​​is the original ranging value, is the reference sound speed, is the path difference of sound ray propagation.

[0054] S4.2, calculate the second derivative of the gridded terrain based on the initial terrain grid, mark the area with curvature as the suspected scour area, obtain the abnormal area coordinates, and automatically plan the MVDR scanning path with the abnormal area coordinates as the center.

[0055] Further, based on the initial terrain grid data, the central difference method is used to calculate the second derivative of each grid node, solve the local curvature value, set the curvature threshold (for example, κ>0.2m -1 ) to mark the high curvature area as the suspected scour area, and extract the abnormal area boundary coordinates; take the center point of the abnormal area as the reference to automatically generate the MVDR fine scanning path, the path spacing is set to 1 / 5 of the coarse scanning resolution (for example, 0.1m×0.1m), and the coverage range is expanded to 5m radius outside the abnormal area, while adjusting the beamforming parameter to focus the scanning energy.

[0056] S4.3, collect snapshot data in the suspected scour area, perform diagonal loading, solve the MVDR optimization problem, and output high-resolution beam response.

[0057] Specifically, the expression is, ; Wherein, is the covariance matrix of the received signal, is the array received signal of the th snapshot, is the conjugate transpose of .

[0058] Specifically, the expression is, ; Wherein, is the beamforming weight vector, is the array manifold vector in the direction .

[0059] S4.4, calculate the frequency offset of the ADCP four-beam received signal to obtain the flow velocity of each depth unit.

[0060] Specifically, the expression is, ; Wherein, is the average Doppler shift, is the upstream beam receiving frequency, is the downstream beam receiving frequency.

[0061] S4.5, detect the echo tail, calculate the bottom speed and correct the water flow speed.

[0062] Specifically, the expression is, ; wherein, is the corrected true water flow speed, is the seabed movement speed, is the water flow speed directly measured by the ADCP.

[0063] S4.6, interpolate the MVDR scanning data to the initial terrain grid, generate the terrain model by Kriging interpolation, interpolate the ADCP flow speed data to the terrain grid according to the inverse distance weighting, and obtain the flow speed profile.

[0064] Further, based on the high-resolution point cloud data obtained by the MVDR fine scanning, the Kriging interpolation algorithm is used to interpolate the discrete measurement point data to the initial terrain grid, and a spatial correlation model is established through semi-variogram analysis to generate a continuous terrain surface; at the same time, the discrete flow speed data measured by the ADCP device is spatially interpolated according to the inverse distance weighting method, and the weight coefficient is calculated by the inverse square of the distance, and a three-dimensional flow speed field is established on the same terrain grid; finally, the comprehensive profile data set containing terrain elevation and flow speed vector is output, realizing the unified representation of geometric deformation and water flow dynamic characteristics.

[0065] S5, fuse the optical image and acoustic point cloud through the graph neural network to construct a three-dimensional model, identify surface defects and calculate the scour pit features, and obtain the detection result.

[0066] S5.1, based on the PTP timestamp and RTK positioning data, convert the optical image pixel coordinates to the world coordinate system unified with the acoustic point cloud through the perspective transformation matrix, compensate for the time deviation and spatial offset, extract the enhanced optical image feature vector using the ResNet-50 network, and apply the spatial attention mechanism to highlight the defect area.

[0067] Further, based on the timestamp data synchronized by the PTPv2 protocol and the centimeter-level position information obtained by the RTK positioning system, first, the optical image is processed for space-time alignment: the camera intrinsic matrix obtained by Zhang Zhengyou calibration and the extrinsic matrix measured by IMU are used to convert the image pixel coordinates to three-dimensional points in the world coordinate system, compensate for the space-time deviation caused by device motion, and the spatial compensation accuracy is then used. The pre-trained ResNet-50 network is used to extract the multi-scale features of the enhanced optical image, and a spatial attention module is introduced after the fourth convolutional layer of the network, which highlights the response strength of the crack, corrosion and other defect areas through spatial weighting of the feature map, and finally outputs the optical feature vector with spatial attention weight.

[0068] S5.2, the acoustic point cloud data is processed using a PointNet++ network to obtain a geometric feature vector to calculate a local curvature enhanced terrain feature.

[0069] Specifically, the expression is, ; wherein, is a local curvature, is a local surface normal, is a query point of curvature, is a centroid of a neighborhood point set.

[0070] S5.3, the optical feature pixel points and the acoustic point cloud clusters are taken as two types of nodes, a graph structure of spatial position and feature attribute is constructed, a node connection relationship is established based on spatial proximity and feature similarity dual criteria, a dynamic graph topology structure is formed, three-layer graph attention convolution is performed, and optical and acoustic feature deep fusion is performed.

[0071] Further, based on the optical feature pixel points extracted by the ResNet-50 network and the acoustic point cloud clusters processed by the PointNet++ network, a heterogeneous graph structure is constructed in the ENU coordinate system: the optical feature pixel points are taken as the first type of nodes, carrying 512-dimensional texture feature vectors; the acoustic point cloud clusters (5cm radius aggregation) are taken as the second type of nodes, carrying 256-dimensional geometric feature vectors. The node relationship is established through double connection criteria: spatial proximity criterion (Euclidean distance <10cm) and feature similarity criterion (cosine similarity >0.8), three-layer graph attention convolution (GAT) is performed on the dynamically generated graph topology structure, the node attention coefficients are calculated at each layer, and finally an 128-dimensional joint feature vector of fused optical texture and acoustic geometry is output, realizing deep coupling of cross-modal features.

[0072] S5.4, based on the fused feature, a crack probability is calculated, and a crack threshold is set to determine the defect evaluation corrosion degree.

[0073] Specifically, the expression is, ; wherein, is a crack existence probability, is a weight matrix, is a bias term.

[0074] S5.5, the volume of the gridded terrain is calculated, fuzzy logic rules are established for risk assessment, a visual model with defect labels and risk levels is generated, and a detection result is obtained.

[0075] Further, based on the grid terrain data generated by photoacoustic fusion, the volume of each grid unit is calculated using tetrahedral unit segmentation method, and the total volume of the scour pit is obtained by accumulation; a fuzzy logic rule base is established, including crack length (e.g. 0-5 mm), corrosion area (e.g. 0-10 cm 2 ), scour depth (e.g. 0-50 cm) and flow velocity gradient (e.g. 0-2 m / s / m), and a trapezoidal membership function is defined to fuzz each parameter (e.g. shallow scour depth is 0-10 cm, medium is 8-30 cm, and deep is 25-50 cm); 25 IF-THEN rules are executed by Mamdani reasoning method (e.g. IF crack length is large AND scour depth is deep THEN risk is high), and a risk index of 0-1 is output; finally, a three-dimensional visualization model with red-yellow-green three-color risk labeling is generated, where high-risk areas (>0.7) are marked in red and additional structural reinforcement suggestions are added, medium-risk areas (0.3-0.7) are marked in yellow and regular monitoring is suggested, and low-risk areas (<0.3) are marked in green as normal areas.

[0076] S6, convert the detection results into engineering visualization, generate a realistic three-dimensional model and superimpose the flow velocity vector field and sediment concentration distribution.

[0077] S6.1, align the optical image, acoustic point cloud, flow velocity profile and sediment concentration data in space through a unified ENU coordinate system to obtain a spatiotemporally synchronized standardized dataset, apply an improved Poisson reconstruction algorithm to generate a watertight triangular mesh model based on acoustic point cloud data, obtain a three-dimensional surface with topological structure, project the enhanced optical image onto the three-dimensional mesh surface, and generate a model with realistic texture.

[0078] Further, based on PTPv2 time synchronization and RTK positioning data, the multi-angle optical images collected by the polarized optical imaging camera, the acoustic point cloud obtained by the multi-beam detection module, the flow velocity profile measured by the ADCP device, and the sediment concentration data inverted by the multi-frequency acoustic suspended sediment concentration profiler are uniformly converted to the ENU-based geodetic coordinate system; the acoustic point cloud data is processed by the improved Poisson reconstruction algorithm, the octree depth is set to 10 levels (e.g. corresponding to 0.1 m resolution), the Poisson equation is solved, and a watertight triangular mesh model is generated; the enhanced optical image (resolution e.g. 2048x2048) is projected onto the three-dimensional mesh surface using perspective texture mapping technology, and bilinear interpolation is used to eliminate the seams, and finally a three-dimensional visualization model with realistic texture is output.

[0079] S6.2, use the inverse distance weighting method to convert discrete ADCP measurement point data into continuous three-dimensional flow velocity field, use Runge-Kutta fourth-order method to solve numerically, and generate flow line trajectories representing water flow dynamics.

[0080] Further, based on the discrete flow velocity measurement point data (including east, north, and sky components) collected by the ADCP device, inverse distance weighting (IDW) is used for spatial interpolation: the distance between each grid node and the surrounding measurement points is calculated, the weight coefficient (the search radius is set to 5 times the average measurement point spacing), and the discrete flow velocity v_i is weighted and averaged to obtain a continuous three-dimensional flow velocity field; in the generated flow velocity field, the Runge-Kutta fourth-order method is used to numerically solve the streamline differential equation, and the time step Δt is adaptively adjusted according to the CFL condition (for example, 0.1-1 second), the initial seed points are uniformly arranged at intervals of 0.2m in the area of interest, and the integral termination condition is set to reach the boundary or exceed 1000 steps of iteration, and finally a family of streamline trajectories reflecting the dynamic characteristics of the water flow is generated.

[0081] S6.3, based on the Kriging interpolation method, a three-dimensional sediment concentration distribution field is constructed, a depth buffer technology is used to realize accurate superposition of the three-dimensional model, the flow velocity field and the sediment field, a Phong lighting model is applied to enhance the realism and detail recognition of the scene, and a three-dimensional measurement tool is embedded in the visualization interface to establish a data-visualization linkage pipeline to realize automatic updating and rendering of new detection results.

[0082] Further, based on the discrete concentration data collected by the multi-frequency acoustic suspended sediment concentration profiler, a three-dimensional sediment concentration distribution field is constructed using the Kriging interpolation method: an exponential half-variogram function model is established, the search radius is set to 3 times the average sampling interval, the Kriging equation set is solved to obtain the spatial optimal unbiased estimate; the accurate superposition of the three-dimensional model (from Poisson reconstruction), the flow velocity field (from ADCP interpolation) and the sediment field is realized through the depth buffer technology, and the z-buffer algorithm is used to process the occlusion relationship; the Phong lighting model is applied to render the scene, the ambient light intensity, diffuse reflectance coefficient, specular reflectance coefficient, and reflectivity are set, the three-dimensional measurement tool is integrated into the visualization interface to support real-time calculation of distance, area, and volume; a data-visualization linkage pipeline based on a message queue is established, and when a new detection result is received, the scene update is automatically triggered, and the rendering frame rate is maintained above 30fps.

[0083] S7, multi-index fusion early warning decision is executed, a fuzzy logic decision model is established to process the detection data, a dynamic early warning threshold is set, and an early warning report is generated.

[0084] S7.1, crack propagation rate, corrosion depth, local scour depth, and flow velocity gradient core indicators are extracted from the detection results, a standardized parameter matrix is generated, a trapezoidal is defined for each indicator, fuzzy rules for each parameter are established, and an IF-THEN rule set is established.

[0085] Further, four key parameters are extracted from the photoacoustic fusion detection results: crack propagation rate (unit: mm / day), corrosion depth (unit: mm), local scour depth (unit: cm), and flow velocity gradient (unit: (m / s) / m). Each parameter is subjected to min-max normalization processing to map the original value to the [0, 1] interval, a standardized parameter matrix is generated, a trapezoidal is defined for each index, a fuzzy rule for each parameter is established, and an IF-THEN rule set is established.

[0086] S7.2, adjust the basic threshold according to the real-time hydrological data, automatically improve the sensitivity during the flood period, calculate the dynamic threshold based on the extreme value statistical theory, and output the self-adaptive warning line.

[0087] Specifically, the expression is, ; Among them, the dynamic threshold, the arithmetic mean of the historical data, the standard deviation of the historical data.

[0088] S7.3, using the Mamdani reasoning method, the comprehensive risk value is obtained by sup-min composite operation, the output risk value is mapped to the four-level warning system, a preliminary warning conclusion is generated, the D-S evidence theory is used to integrate the optical and acoustic multi-dimensional evidence, based on the template engine, the risk value, the key parameter curve, the disposal suggestion and other elements are combined to generate a warning report.

[0089] Further, the Mamdani reasoning method is used to process the 25 IF-THEN rules in the fuzzy logic rule base, and the comprehensive risk value is calculated by sup-min composite operation: first, the minimum input membership degree of each rule premise part is taken as the rule triggering strength, then the rule conclusion is truncated, and finally the membership union of all activated rule conclusions is taken to obtain the comprehensive risk value R in the interval [0, 1]. The risk value R is mapped to the four-level warning system: normal (0-0.3), attention (0.3-0.6), warning (0.6-0.8), and emergency (0.8-1.0), to generate a preliminary warning conclusion containing the risk level.

[0090] The embodiment also provides a computer device suitable for the bridge underwater safety detection method based on the photoacoustic dual mode, which comprises a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions to realize the bridge underwater safety detection method based on the photoacoustic dual mode proposed in the above embodiment.

[0091] The computer device can be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected by a system bus. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for running the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is configured to perform wired or wireless communication with an external terminal. The wireless communication can be achieved by WIFI, an operator network, NFC (Near Field Communication) or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device, or an external keyboard, touchpad or mouse, etc.

[0092] The embodiment also provides a storage medium having a computer program stored thereon, the program being executed by a processor to implement the method for detecting underwater safety of a bridge based on a photoacoustic dual mode as described above. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as a static random access memory (SRAM), an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic memory, a flash memory, a magnetic disk or an optical disk.

[0093] To sum up, the application adopts a graph neural network to fuse an optical image and an acoustic point cloud in depth, dynamically optimizes weights through an attention mechanism, improves detection accuracy through a spatio-temporal registration algorithm, and constructs a multi-index fusion model based on fuzzy logic and D-S evidence theory for adaptive early warning decision-making, so as to realize upgrading from single-parameter alarm to multi-dimensional decision-making, reduce false alarm rate, and double early warning lead time through a dynamic threshold mechanism. The environment perception layer optimizes optical parameters in real time through multi-frequency acoustic inversion, the detection layer combines MVDR beam forming and mutual information minimization algorithm, and the decision layer realizes three-dimensional visualization and intelligent early warning.

[0094] It should be noted that the above examples are only used to illustrate the technical solutions of the present application but not limit the present application. Although the present application is described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or equivalently replaced, without departing from the spirit and scope of the technical solutions of the present application, which should be covered in the scope of the claims of the present application.

Claims

1. A bridge underwater safety detection method based on photoacoustic dual-mode, characterized in that: comprising, The particle size distribution and concentration data of the water area are obtained by using a multi-frequency acoustic suspended sediment concentration profiler, an environmental parameter database is established, a shipborne detection device is configured according to the initial environmental parameter database, a three-axis stabilized platform carrying a polarized imaging camera, a multi-spectral LED array, a multi-beam detection module and an ADCP device are set up, and a unified coordinate system is established; The multi-angle original intensity images are collected by the polarized optical imaging module, and after processing by the mutual information minimization algorithm, multi-exposure fusion is performed, and enhanced images are output through feature pyramid network fusion and intelligent tone mapping; Based on the time delay sum beam forming algorithm, the multi-beam detection module is controlled to perform coarse scanning, the MVDR fine scanning is started in the scouring area, and the flow velocity profile is measured based on the Doppler frequency offset by using the ADCP device; The optical image and the acoustic point cloud are fused by the graph neural network to construct a three-dimensional model, surface defects are identified and scour pit characteristics are calculated, and the detection result is obtained; The detection result is converted into engineering visualization, a realistic three-dimensional model is generated and superimposed with the flow velocity vector field and the sediment concentration distribution, multi-index fusion early warning decision is executed, a fuzzy logic decision model is established to process the detection data, a dynamic early warning threshold is set and an early warning report is generated.

2. The bridge underwater safety detection method based on photoacoustic dual-mode according to claim 1, characterized in that: The particle size distribution and concentration data of the water area are obtained by using a multi-frequency acoustic suspended sediment concentration profiler, an environmental parameter database is established, including the following steps, The multi-frequency acoustic suspended sediment concentration profiler is started, the electronic element is preheated for thirty minutes for stabilization, the four-frequency working mode is configured, the pulse width, interval and transmission power are set, the reference echo signals of each frequency band are collected in the distilled water without sediment, and the noise floor is recorded; The water temperature, salinity and depth are measured by using the built-in CTD sensor, and the sound velocity profile is calculated; The control command is generated according to the spiral scanning path to guide the motion trajectory of the transducer array, the long-short pulse combination is transmitted in the order of frequency priority, the acoustic detection signal is generated, the echo of each frequency band is received by the 16-bit high-precision ADC sampling rate, and the weak signal is amplified by applying time gain compensation; Based on the modified hybrid scattering model, the scattering intensity of each frequency band at different depth units is obtained, the intensity-depth matrix is established, the multi-frequency simultaneous equations are solved by Tikhonov regularization, the concentration distribution of fine, medium and coarse particle sizes is output, the bubble interference is identified by using high-frequency data, the abnormal measurement points are removed by using the SVM classifier, the sliding window average is performed, and the environmental parameter database is established.

3. The bridge underwater safety detection method based on photoacoustic dual-mode according to claim 2, characterized in that: The shipborne detection device is configured according to the initial environmental parameter database, the three-axis stabilized platform carrying the polarized imaging camera, the multi-spectral LED array, the multi-beam detection module and the ADCP device are set up, and the unified coordinate system is established, including the following steps, The turbidity, water depth and bottom material indexes are read from the initial environmental parameter database, the device configuration basic parameter table is generated, the light attenuation coefficient is calculated according to the turbidity data, the penetration ability evaluation report of each wave band is obtained, and the platform compensation angle is calculated based on the real-time attitude data of the ship IMU; The working mode is automatically selected according to the turbidity level, the polarization angle sequence is generated, the multi-spectral LED output ratio is set according to the optical evaluation report, and the illumination scheme is obtained; According to the water depth data, the working frequency is automatically selected, the frequency configuration instruction is obtained, the pulse length is calculated based on the flow velocity range and the depth, and the profile measurement scheme is output; Through the PTPv2 protocol, all devices are synchronized to the microsecond level, unified timestamp data is obtained, and the ENU geodetic coordinate system is established.

4. The bridge underwater safety detection method based on photoacoustic dual-mode according to claim 3, characterized in that: Through the polarization optical imaging module, multi-angle original intensity images are collected, mutual information minimization algorithm is used for processing, multi-exposure fusion is performed, feature pyramid network fusion and intelligent tone mapping are used to output enhanced images, including the following steps, Based on the proportion of the control multispectral LED array, a stable illumination environment is provided for image acquisition, the polarization plate is rotated to 0°, 45°, 90° and 135° in turn, a polarization angle control sequence is generated, and at each polarization angle, the camera is triggered to collect images at three exposure times, obtaining original intensity images; The mutual information minimization algorithm is used to calculate the separation degree of target information light and background scattered light in each polarization image; The mutual information values of different polarization angle combinations are compared to obtain the despeckling intermediate image, the local contrast of each exposure image is calculated, and a quality score map is generated; Based on the quality score, the softmax function is used to calculate the pixel-level weight of each exposure image, and a weight distribution matrix is output; A 5-layer Laplacian pyramid is constructed, and different exposure images are fused on each pyramid layer according to the weight map, and an HDR fusion image is output; According to the image histogram, the pixels are divided into dark, intermediate and bright regions, the non-sharpening mask algorithm is used to enhance high-frequency details, and an enhanced image is obtained.

5. The bridge underwater safety detection method based on photoacoustic dual-mode according to claim 4, characterized in that: Based on the control of the multi-beam detection module, the time delay sum beam forming algorithm is used for coarse scanning, the MVDR fine scanning is started in the scouring area, and the ADCP device measures the flow velocity profile based on the Doppler frequency offset, including the following steps, According to the water depth measurement data, the working frequency is automatically selected, the pulse length and beam width parameters are set, the phase consistency of the transducer array is detected, the time delay deviation of each array element is corrected, the array element time delay is obtained based on each beam direction, the Hann window is used for weighting, and the output beam signal is obtained, based on the beam sounding data, the initial terrain grid is generated through the sound ray tracing algorithm; Based on the initial terrain grid, the second derivative of the grid terrain is calculated, the area with curvature is marked as a suspected scouring area, and the abnormal area coordinates are obtained, and the MVDR scanning path is automatically planned with the abnormal area coordinates as the center. In the suspected scouring area, snapshot data is collected, diagonal loading is performed, the MVDR optimization problem is solved, and a high-resolution beam response is output; The frequency offset is calculated for the ADCP four-beam receiving signal, and the flow velocity of each depth unit is obtained. The echo trailing edge is detected, the bottom velocity is calculated, and the water flow velocity is corrected; The MVDR scanning data is interpolated to the initial terrain grid, the terrain model is generated through the Kriging interpolation, the ADCP flow velocity data is interpolated to the terrain grid according to the inverse distance weighting, and the flow velocity profile is obtained.

6. The bridge underwater safety detection method based on photoacoustic dual-mode according to claim 5, characterized in that: The optical image and the acoustic point cloud are fused through the graph neural network to construct a three-dimensional model, surface defects are recognized and scour pit features are calculated, and detection results are obtained, including the following steps, Based on the PTP timestamp and RTK positioning data, the optical image pixel coordinates are converted into the world coordinate system unified with the acoustic point cloud through the perspective transformation matrix, compensating for the time deviation and spatial offset, and the ResNet-50 network is used to extract the enhanced optical image feature vector, and the spatial attention mechanism is applied to highlight the defect area; The PointNet++ network is used to process the acoustic point cloud data to obtain the geometric feature vector to calculate the local curvature to enhance the terrain features; The optical feature pixel points and acoustic point cloud clusters are taken as two types of nodes to construct a graph structure of spatial position and feature attributes, and the node connection relationship is established based on the dual criteria of spatial proximity and feature similarity to form a dynamic graph topology, and a three-layer graph attention convolution is performed for deep fusion of optical and acoustic features; Based on the fusion features, the crack probability is calculated, and a crack threshold is set to determine the defect evaluation of the corrosion degree; The volume of the grid terrain is calculated, fuzzy logic rules are established for risk assessment, and a visual model with defect labels and risk levels is generated to obtain the detection results.

7. The bridge underwater safety detection method based on photoacoustic dual-mode according to claim 6, characterized in that: The detection results are converted into engineering visualization to generate a realistic three-dimensional model and superimpose the flow velocity vector field and sediment concentration distribution, including the following steps, The optical image, acoustic point cloud, flow velocity profile and sediment concentration data are spatially aligned through a unified ENU coordinate system to obtain a standardized dataset that is synchronized in time and space, based on the acoustic point cloud data, an improved Poisson reconstruction algorithm is applied to generate a watertight triangular mesh model, and a three-dimensional surface with a topology is obtained, the enhanced optical image is projected onto the three-dimensional mesh surface to generate a model with realistic texture; The inverse distance weighting method is used to convert discrete ADCP measurement point data into a continuous three-dimensional flow field, and the Runge-Kutta fourth-order method is used for numerical solution to generate flow line trajectories representing water flow dynamics; Based on the Kriging interpolation method, a three-dimensional sediment concentration distribution field is constructed, and a depth buffer technique is used to accurately superimpose the three-dimensional model, flow field and sediment field, and a Phong lighting model is applied to enhance the realism and detail recognition of the scene, and a three-dimensional measurement tool is embedded in the visualization interface to establish a data-visualization linkage pipeline, enabling automatic updating and rendering of new detection results.

8. The bridge underwater safety detection method based on photoacoustic dual-mode according to claim 7, characterized in that: Multi-index fusion early warning decision is executed, a fuzzy logic decision model is established to process detection data, dynamic warning thresholds are set and a warning report is generated, including the following steps, Core indicators such as crack propagation rate, corrosion depth, local scour depth and flow velocity gradient are extracted from the detection results to generate a standardized parameter matrix, a trapezoidal is defined for each indicator, fuzzy rules for each parameter are established, and an IF-THEN rule set is established, The basic threshold is adjusted according to real-time hydrological data, the sensitivity is automatically improved during the flood period, the dynamic threshold is calculated based on the extreme value statistical theory, and the adaptive warning line is output; The Mamdani reasoning method is used to obtain the comprehensive risk value through sup-min composite operation, the output risk value is mapped to the four-level warning system to generate the preliminary warning conclusion, the D-S evidence theory is used to integrate multi-dimensional evidence from optical and acoustic data, and the risk value, key parameter curve, disposal suggestions and other elements are combined based on the template engine to generate the warning report. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is characterized in that: The computer program is executed by the processor to realize the steps of the bridge underwater safety detection method based on the photoacoustic dual mode in any one of claims 1-8.

10. A computer readable storage medium having stored thereon a computer program, characterized in that: The computer program is executed by the processor to realize the steps of the bridge underwater safety detection method based on the photoacoustic dual mode in any one of claims 1-8.

Images