Bridge safety analysis method, device, equipment and medium

Abstract

The invention relates to the technical field of bridge structure health monitoring and safety early warning, and discloses a bridge safety analysis method, device and equipment and a medium, and the method comprises the steps: collecting the building data of a bridge, and extracting the building parameters; converting the building parameters into feature vectors of a data mode, obtaining a feature space of the data mode, and mapping the feature vectors into the feature space to obtain spatial features of the building data; modeling the spatial features to obtain joint state characterization of the bridge, and identifying structural abnormal parameters of the bridge according to the joint state characterization; carrying out feature vector similarity calculation on the structure abnormal region of the bridge to obtain a feature vector matching degree quantized value; according to the feature vector matching degree quantized value, adjusting the data modal weight and carrying out weighted fusion on the spatial features to obtain weighted fusion representation; and analyzing the safety level and the damage degree of the bridge according to the weighted fusion representation. The accuracy of bridge safety analysis can be improved.

Description

Technical Field

[0001] This invention relates to the field of bridge structural health monitoring and safety early warning technology, and in particular to a bridge safety analysis method, device, equipment and medium. Background Technology

[0002] Bridge structural health monitoring and safety early warning are key areas for ensuring the stable operation of transportation infrastructure. With the expansion of bridge construction scale and the increase in service life, the requirements for comprehensiveness, real-time performance, and intelligence in monitoring are continuously increasing. Mobile platforms, such as drones and robots, are increasingly widely used in inspection scenarios. The acquisition of multi-source data on bridge structure, load, and environment by various sensors, including vision, lidar, sonar, and electromagnetic induction, is becoming a development trend. Deep integration and intelligent interpretation of multimodal information are gradually becoming core requirements. Therefore, to meet the needs of comprehensive and high-precision bridge monitoring, it is necessary to fuse and intelligently analyze multi-source, multimodal data to improve the accuracy of bridge safety analysis.

[0003] Current technologies largely rely on manual inspections or single-sensor data collection for single-modal bridges. After parameter extraction, traditional statistical or simple image processing methods are used, without converting the parameters into corresponding modal feature vectors and mapping them to the feature space. This lack of cross-modal joint modeling means that data is simply superimposed or combined with fixed weights, without quantifying the matching degree through feature vector similarity calculations or dynamically adjusting modal weights. This approach results in incomplete data coverage, loss of multimodal interaction information, inability to adapt to sensor reliability and environmental changes, biased extraction of structural anomaly feature parameters, insufficient robustness and efficiency in analysis, and ultimately, low accuracy in bridge safety analysis. Summary of the Invention

[0004] This invention provides a bridge safety analysis method, apparatus, equipment, and medium to address the problem of low accuracy in bridge safety analysis.

[0005] Firstly, a bridge safety analysis method is provided, including: The construction data of the target bridge is collected by multiple preset sensors according to different data modes, and the construction parameters are extracted from the construction data collected by the target sensors. The building parameters are converted into feature vectors corresponding to the data modes of the target sensor, and the feature space corresponding to the data modes of the target sensor is obtained. The feature vectors are mapped into the feature space to obtain the spatial features of the building data collected by the target sensor. Cross-modal joint modeling is performed on the spatial features corresponding to all sensors to obtain the joint state characterization of the target bridge, and the structural anomaly feature parameters of the target bridge are identified based on the joint state characterization. Based on the structural anomaly feature parameters, the spatial feature vector corresponding to the structural anomaly region of the target bridge is compared with the anomaly feature vector retrieved from the preset bridge anomaly feature vector library to calculate the feature vector similarity and obtain the feature vector matching metric value. The weights of the data modes corresponding to each sensor are adjusted according to the metric values ​​of the feature vector matching, and the spatial features corresponding to all sensors are weighted and fused according to the adjusted weights to obtain a weighted fusion representation. The safety level and damage level of the target bridge are analyzed based on the weighted fusion characterization.

[0006] Secondly, a bridge safety analysis device is provided, comprising: The building parameter extraction module is used to collect building data of the target bridge using multiple preset sensors according to different data modes, and extract the building parameters from the building data collected by the target sensors. The feature vector mapping module is used to convert the building parameters into feature vectors of the data mode corresponding to the target sensor, obtain the feature space corresponding to the data mode of the target sensor, and map the feature vectors into the feature space to obtain the spatial features of the building data collected by the target sensor. The structural anomaly feature parameter identification module is used to perform cross-modal joint modeling of the spatial features corresponding to all sensors to obtain the joint state characterization of the target bridge, and to identify the structural anomaly feature parameters of the target bridge based on the joint state characterization. The feature vector similarity calculation module is used to perform feature vector similarity calculation on the spatial feature vector corresponding to the structural anomaly region of the target bridge and the anomaly feature vector retrieved from the preset bridge anomaly feature vector library based on the structural anomaly feature parameters, and obtain the feature vector matching metric value. The spatial feature weighted fusion module is used to adjust the weight of the data mode corresponding to each sensor according to the feature vector matching quantification value, and to perform weighted fusion of the spatial features corresponding to all sensors according to the adjusted weights to obtain a weighted fusion representation. The bridge safety level and damage degree analysis module is used to analyze the safety level and damage degree of the target bridge based on the weighted fusion characterization.

[0007] Thirdly, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the bridge safety analysis method described above.

[0008] Fourthly, a computer-readable storage medium is provided, which stores a computer program that, when executed by a processor, implements the steps of the bridge safety analysis method described above.

[0009] The aforementioned bridge safety analysis methods, devices, equipment, and media enable the acquisition of bridge structural data through multiple sensors and modalities via a client-side approach. This data is then used to accurately capture structural anomalies through feature mapping and cross-modal joint modeling. Modal weights are adjusted based on feature vector similarity quantification, and weighted fusion comprehensively integrates effective information. This approach achieves comprehensive perception of bridge internal and external features and environmental characteristics, reducing the omission of hidden defects. Furthermore, dynamic weighting enhances the relevance and reliability of the analysis, enabling accurate assessment of safety levels and damage levels. This addresses the problem of low accuracy in bridge safety analysis. Attached Figure Description

[0010] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a schematic diagram of an application environment for the bridge safety analysis method in one embodiment of the present invention; Figure 2 This is a flowchart illustrating a bridge safety analysis method according to an embodiment of the present invention; Figure 3 This is a schematic diagram of a bridge safety analysis device according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of a computer device according to an embodiment of the present invention; Figure 5 This is another structural schematic diagram of a computer device according to one embodiment of the present invention. Detailed Implementation

[0012] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0013] The bridge safety analysis method provided in this invention can be applied to, for example... Figure 1In this application environment, the client communicates with the server via a network. The server can collect bridge construction data through multiple sensors and multiple modalities via the client. Through feature mapping and cross-modal joint modeling, it accurately captures structural anomalies. Combining feature vector similarity quantification to adjust modal weights, and after weighted fusion, it comprehensively integrates effective information. This achieves comprehensive perception of the bridge's internal and external features and environmental characteristics, reducing the omission of hidden defects. Furthermore, dynamic weighting improves the targeting and reliability of the analysis, enabling accurate assessment of safety levels and damage levels, thus solving the problem of low accuracy in bridge safety analysis. The client can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. The server can be implemented using a standalone server or a server cluster consisting of multiple servers. The invention will be described in detail below through specific embodiments.

[0014] Please see Figure 2 As shown, Figure 2 A flowchart illustrating the bridge safety analysis method provided in this embodiment of the invention includes the following steps: S1. Use multiple preset sensors to collect the building data of the target bridge according to different data modes, and extract the building parameters in the building data collected by the target sensors.

[0015] In this embodiment of the invention, the preset multiple sensors can be multi-view high-definition cameras, lidar, sonar sensors, electromagnetic induction modules, environmental and load sensors, and other sensors with multiple data modalities integrated on platforms such as drones / robots. Data modalities refer to different data types corresponding to each sensor, such as images, 3D point clouds and deformation, underwater / bridge bottom surface defect detection signals, steel bar / bolt status and concrete internal change signals, and environmental and load signals such as temperature / humidity / wind speed / vibration / vehicle load. The target bridge in this embodiment can be a cable-stayed bridge, suspension bridge, beam bridge, or other bridge types. The construction data includes multi-source data information related to bridge structure, construction dimensions, inclination, load, and environment. In this embodiment of the invention, the step of collecting construction data of the target bridge using the preset multiple sensors according to different data modalities includes: Identify the correspondence between multiple preset sensors and different data modes; The correspondence is used to control the multiple sensors to collect data from different structural parts of the target bridge, thereby obtaining the sensing data corresponding to the data mode; The sensor data is assimilated to obtain the architectural data of the target bridge.

[0016] In detail, the correspondence is as follows: multi-view high-definition cameras correspond to bridge structure surface images, lidar corresponds to 3D point clouds and deformation data, sonar sensors correspond to defect detection signals at underwater or bridge bottom surfaces, electromagnetic induction modules correspond to the state of steel bars / bolts and internal changes in concrete, and environmental and load sensors correspond to environmental and load signals such as temperature / humidity / wind speed / vibration / vehicle load. The correspondence between sensors with different data modes and different data modes can be identified by querying and retrieving different sensor types.

[0017] Specifically, the sensing data includes bridge structure surface images captured by multi-view high-definition cameras, three-dimensional point clouds and deformation data acquired by lidar, underwater or bridge deck defect signals detected by sonar sensors, steel bar / bolt status and concrete internal change signals monitored by electromagnetic induction modules, and environmental and load signals such as temperature / humidity / wind speed / vibration / vehicle load recorded by environmental and load sensors. The UAV / robot platform is equipped with multiple sensors that correspond to each data mode. Based on the correspondence, the sensors are applied to different structural parts of the target bridge, such as the bridge deck, piers, towers, and steel cables. Each sensor works synchronously to directly capture the original information under the corresponding data mode, thereby obtaining the sensing data.

[0018] Furthermore, the sensor data undergoes preprocessing such as denoising, calibration, and alignment registration on the edge computing unit. Multi-view images are stitched and enhanced, point clouds are filtered and registered, sonar and electromagnetic signals have feature parameters extracted, and load and environmental signals are time-series filtered. All processed sensor data are converted into feature vectors in a unified format, thereby obtaining the architectural data of the target bridge.

[0019] In this embodiment of the invention, the target sensor is a multi-view high-definition camera, lidar, sonar sensor, electromagnetic induction module, and environmental and load sensors integrated on an unmanned aerial vehicle / robot platform; the building parameters include the location and length of cracks in the bridge structure, the degree of cable loosening, the displacement of supports, three-dimensional point cloud and deformation-related parameters, the state parameters of reinforcing bars and bolts, the internal change parameters of concrete, and environmental and load-related parameters such as temperature, humidity, wind speed, vibration, and vehicle load.

[0020] In detail, the building data collected by the target sensor is preprocessed in the edge computing unit, including denoising, calibration, alignment and registration. Multi-view images are stitched and enhanced, point clouds are filtered and registered, feature parameters of sonar and electromagnetic signals are extracted, and load and environmental signals are time-series filtered. At the same time, the consistency of detection results from different sensors is verified. Finally, building parameters such as crack location and length, cable loosening degree, support displacement, three-dimensional point cloud and deformation-related parameters are output.

[0021] S2. The building parameters are converted into feature vectors of the data mode corresponding to the target sensor, and the feature space corresponding to the data mode of the target sensor is obtained. The feature vectors are mapped into the feature space to obtain the spatial features of the building data collected by the target sensor.

[0022] In this embodiment of the invention, the feature vector is a unified format data carrier converted from the original multimodal bridge data collected by the sensor after preprocessing such as denoising, calibration, alignment and registration, stitching enhancement, filtering, feature parameter extraction, and time series filtering. It contains key feature information of each modal data and is used to input into the multimodal deep model for fusion analysis, providing data support for bridge health status analysis and early warning.

[0023] In detail, after the target sensor collects the original architectural parameters related to the bridge structure, load and environment, it performs preprocessing such as denoising, calibration and alignment registration on the edge computing unit. Among these processes, multi-view images are stitched and enhanced, point clouds are filtered and registered, feature parameters are extracted from sonar and electromagnetic signals, and load and environmental signals are time-series filtered. After these processes, various types of data are converted into feature vectors in a unified format for subsequent fusion analysis.

[0024] In this embodiment of the invention, the feature space is a unified vector space that contains the feature vectors of the data modality corresponding to the target sensor after being processed by the encoder. This enables the features of the data modality to be in the same dimensional system as the features of other sensor data modalities, providing a foundation for the subsequent application of cross-modal fusion mechanisms, interactive reasoning of multi-source information, and the generation of joint representations of bridge health status.

[0025] In detail, after the features of the data mode corresponding to the target sensor are preprocessed and converted into feature vectors, they are mapped by an encoder dedicated to the data mode to form a unified vector space that can accommodate the feature and is in the same dimensional system as the features of other sensor data modes. This is the feature space corresponding to the data mode of the target sensor.

[0026] In this embodiment of the invention, the spatial feature is a representation formed by mapping the feature vector corresponding to the target sensor to a unified vector space through its dedicated encoder. It includes the key features of the building data collected by the sensor and can be in the same dimensional system as the features of other sensor data modes, providing a foundation for subsequent cross-modal fusion, multi-source information interaction reasoning and the generation of joint representations of bridge health status.

[0027] In this embodiment of the invention, mapping the feature vector to the feature space to obtain the spatial features of the building data collected by the target sensor includes: The feature vector is filtered and denoised to obtain a denoised feature vector; The denoised feature vectors are aligned and registered to obtain the registered feature vectors. The feature dimensions of the registered feature vectors are uniformly adjusted to preset values ​​to obtain feature vectors with uniform dimensions. The dimension-unified feature vector is mapped to the feature space using a preset feature mapping matrix to obtain a mapping space representation. Extract the inherent device parameters and environmental adaptation parameters of the target sensor respectively; Based on the inherent parameters of the device and the environmental adaptation parameters, the spatial coordinates of the mapped spatial representation are calibrated to obtain the spatial characteristics of the building data collected by the target sensor.

[0028] In detail, the denoised feature vector is a vector obtained by filtering and denoising the feature vector acquired by the sensor after preprocessing and transformation, removing noise interference and retaining effective features. Methods such as mean filtering, median filtering, or Kalman filtering can be used to filter the feature vector to achieve denoising. The registration feature vector is a vector obtained by aligning and registering the denoised feature vector to achieve data calibration and alignment, facilitating subsequent fusion analysis. In this embodiment, a registration method based on feature point matching can be used for alignment and registration. The dimension-unified feature vector is a vector obtained by uniformly adjusting the feature dimensions of the registration feature vector to a preset value to meet the dimensional requirements of multimodal fusion analysis. The denoised feature vector is obtained by filtering and denoising the feature vector acquired by the sensor after preprocessing and transformation to remove noise interference. Alignment and registration are then performed on this denoised feature vector to achieve data calibration and alignment, resulting in a registration feature vector. The feature dimensions of the registration feature vector are then uniformly adjusted to a preset value. For example, the feature dimensions of the registration feature vector can be uniformly adjusted to 512 dimensions through encoder mapping, multilayer perceptron transformation, or feature normalization methods, ultimately obtaining a dimension-unified feature vector.

[0029] Specifically, the preset feature mapping matrix is ​​a preset parameter matrix adapted to a unified feature space and used to realize the transformation of dimension-unified feature vectors; the mapping space representation is a bridge-related feature representation formed after the dimension-unified feature vectors are mapped to the feature space by this matrix, which can interact with the modal features of other sensors, providing a foundation for subsequent cross-modal fusion analysis; the dimension-unified feature vectors are transformed by the preset feature mapping matrix adapted to the unified feature space and mapped to the corresponding feature space, forming a mapping space representation that can participate in cross-modal fusion inference.

[0030] Furthermore, the inherent parameters of the device are the fixed hardware attribute parameters of the target sensor itself, while the environmental adaptation parameters are parameters related to the bridge inspection environment and used to adapt to the working state of the sensor. For the target sensor mounted on the UAV / robot platform, its inherent hardware attribute parameters are directly extracted, and the relevant parameters for adapting to the inspection environment conditions are also extracted, thus completing the separate extraction of the inherent parameters and environmental adaptation parameters of the target sensor device.

[0031] Furthermore, by combining the inherent hardware attribute parameters of the target sensor with the relevant parameters adapted to the bridge detection environment, the spatial coordinates of the mapping space representation formed by mapping the dimension-unified feature vector to the feature space through a preset feature mapping matrix are calibrated. Finally, spatial features that can reflect the key features of the building data collected by the target sensor and are adapted to subsequent fusion analysis are obtained.

[0032] S3. Perform cross-modal joint modeling of the spatial features corresponding to all sensors to obtain the joint state characterization of the target bridge, and identify the structural anomaly characteristic parameters of the target bridge based on the joint state characterization.

[0033] In this embodiment of the invention, the joint state representation refers to the multi-source data of the bridge collected by each sensor, which is preprocessed and mapped to a unified vector space by their respective encoders. Then, cross-modal joint modeling is achieved through a cross-modal fusion mechanism. The final generated unified representation can comprehensively reflect the multi-dimensional information of the bridge structure, load, environment and other aspects, as well as the interaction characteristics of the data of each modality. It can be used to generate structured diagnostic results such as bridge damage description, deformation parameters and risk assessment level.

[0034] In this embodiment of the invention, the step of performing cross-modal joint modeling of the spatial features corresponding to all sensors to obtain the joint state characterization of the target bridge includes: Based on the spatial features corresponding to all sensors, query vectors, key vectors, and value vectors of the data modality corresponding to each sensor are constructed through linear transformations. Calculate the cosine similarity between the query vector corresponding to each data modality and the key vectors corresponding to all data modalities, and use the cosine similarity as the association weight corresponding to each data modality; The cross-modal fusion features are obtained by weighting and summing the value vectors corresponding to all data modalities based on the association weights. The cross-modal fusion features are sequentially subjected to linear transformation and nonlinear activation processing to extract deep interaction features between the data modes, thereby obtaining the joint state representation of the target bridge.

[0035] In detail, the query vector is a vector obtained by linear transformation based on sensor spatial features and used for mapping between different modalities to allocate attention weights; the key vector is a vector obtained by linear transformation based on sensor spatial features and used for matching and calculating relevance with the query vector; and the value vector is a vector obtained by linear transformation based on sensor spatial features and carries the core feature information of the corresponding modality.

[0036] In this embodiment of the invention, the spatial features of the bridge structure, load, environment, etc. collected by each sensor are used as the basis. The data modes of each sensor are transformed into corresponding query vectors, key vectors and value vectors through linear transformation. With the help of these vectors, the query vectors and key vectors are interacted in the cross-modal fusion mechanism to determine the importance of each modal feature. Then, the value vector is combined to realize cross-modal joint modeling of data modes from different sensors, which provides support for the subsequent generation of joint state representation of the bridge.

[0037] Specifically, cosine similarity is a quantitative indicator that measures the similarity between the query vector of each data modality and the key vectors of all data modalities; association weight is a numerical value that reflects the importance or contribution of each data modality in cross-modal fusion, obtained based on the cosine similarity calculation results. Taking the query vectors corresponding to each data modality and the key vectors corresponding to all data modalities as the calculation objects, the cosine similarity formula is used to calculate the cosine similarity value representing the similarity between the two. These similarity values ​​are then converted into association weights corresponding to each data modality, providing a basis for adaptively allocating the weights of each modality feature in subsequent cross-modal fusion.

[0038] Furthermore, the cross-modal fusion feature is a unified feature that integrates the core interactive information of multimodal data from different sensors by weighting and summing all value vectors carrying the core feature information of the corresponding modality based on the association weights of each data modality. Based on the association weights and value vectors corresponding to each data modality, the value vector of each data modality is multiplied by its own association weight, and all value vectors after weight adjustment are summed to form a cross-modal fusion feature that can support the subsequent joint generation of bridge health status and the output of structured diagnostic results.

[0039] Furthermore, deep interaction features are the core features extracted from cross-modal fusion features after linear transformation and nonlinear activation processing, which can reflect the deep correlation and interaction between different data modes. Based on cross-modal fusion features, the features are first linearly transformed to adjust the dimensions and distribution, and then nonlinear activation processing is used to mine the complex correlation between data modes, thereby extracting deep interaction features. Finally, a joint state representation that can comprehensively reflect the target bridge structure, load, environment and damage status and other multi-dimensional information is formed.

[0040] In this embodiment of the invention, the structural anomaly characteristic parameters refer to specific quantitative indicators that reflect the abnormal state of the target bridge structure, identified based on joint state characterization. These include parameters related to bridge structural damage and deformation, such as crack location and length, cable loosening degree, and support displacement.

[0041] In this embodiment of the invention, identifying the structural anomaly feature parameters of the target bridge based on the joint state characterization includes: The joint state representation is screened for abnormal features using a preset structural anomaly feature judgment standard to obtain the set of abnormal features of the target bridge; Based on the set of abnormal features, the abnormal regions of the structural parts of the target bridge are located to obtain the structural abnormal regions. Identify the structural anomaly type corresponding to the joint state representation of the structural anomaly region; By combining the structural anomaly types and structural anomaly regions, structural anomaly characteristic parameters of the target bridge are obtained.

[0042] In detail, the preset structural anomaly feature judgment criteria are pre-set quantitative indicators and judgment rules used to measure whether there are anomalies in the bridge structure; the anomaly feature set is a summary of all features reflecting the abnormal state of the target bridge structure that meet the judgment criteria after screening; based on the preset structural anomaly feature judgment criteria and joint state representation, the joint state representation that comprehensively reflects the multi-dimensional information of bridge structure, load, environment and damage status is screened through the judgment criteria, and the features that meet the anomaly judgment conditions are extracted and integrated to form the anomaly feature set of the target bridge.

[0043] Specifically, structurally abnormal areas are the specific spatial locations on the target bridge where structural anomalies such as cracks, corrosion, and loose bolts exist. Based on the location information and descriptions of the anomalies contained in the set of anomaly features, this information is correlated with the spatial distribution of the bridge structure to accurately locate the specific areas on the target bridge where structural anomalies exist.

[0044] Furthermore, structural anomaly type refers to the specific anomaly type presented in the structural anomaly area, including specific categories reflecting bridge structural damage or deformation such as cracks, corrosion, bolt loosening, and support displacement. Taking the structural anomaly area and its corresponding joint state characterization as the core, and relying on the structured diagnostic results generated by the multimodal large model, the characteristic information such as damage and deformation related to the anomaly area is extracted from the joint state characterization, thereby accurately identifying the specific structural anomaly type corresponding to the area.

[0045] Furthermore, taking the identified structural anomaly types and the located structural anomaly areas as the core, the relevant information of the two is integrated and summarized to form target bridge structural anomaly characteristic parameters that include quantitative indicators such as the specific type of anomaly, the corresponding spatial location and crack length, and the support displacement.

[0046] S4. Based on the structural anomaly feature parameters, perform feature vector similarity calculation on the spatial feature vector corresponding to the structural anomaly region of the target bridge and the anomaly feature vector retrieved from the preset bridge anomaly feature vector library to obtain the feature vector matching metric value.

[0047] In this embodiment of the invention, the feature vector matching quantification value is a numerical value obtained by combining structural anomaly feature parameters and performing feature vector similarity calculation on the spatial features corresponding to the structural anomaly region of the target bridge, which is used to quantify the similarity between the spatial feature vector and the relevant reference feature vector.

[0048] In this embodiment of the invention, the step of calculating the similarity between the spatial feature vector corresponding to the structural anomaly region of the target bridge and the anomaly feature vector retrieved from a preset bridge anomaly feature vector database based on the structural anomaly feature parameters, to obtain a feature vector matching metric value, includes: Using the aforementioned structural anomaly feature parameters, target spatial feature vectors matching the structural anomaly type are selected from the spatial features corresponding to the structural anomaly region of the target bridge. Search the preset bridge anomaly feature vector library for anomaly feature vectors that belong to the same structural anomaly type as the target spatial feature vector and have a similarity exceeding a preset similarity threshold, and calculate the similarity value between the target spatial feature vector and the anomaly feature vector; Extract the sensor confidence weights associated with the structural anomaly feature parameters; The similarity values ​​are weighted and summed according to the sensor confidence weights to obtain the feature vector matching metric between the target space feature vector and the abnormal feature vector.

[0049] In detail, the target spatial feature vector is a uniformly formatted feature vector selected from the spatial features corresponding to the structural anomaly area of ​​the target bridge, which matches the structural anomaly type and meets the requirements of the structural anomaly feature parameters. Based on the structural anomaly feature parameters, the spatial features corresponding to the structural anomaly area, and the structural anomaly type, feature vectors that match the structural anomaly type and meet the structural anomaly feature parameter conditions are accurately selected from the spatial features to obtain the target spatial feature vector.

[0050] Specifically, the pre-set bridge anomaly feature vector library is a collection of uniformly formatted anomaly feature vectors converted from historical multimodal bridge anomaly data after preprocessing, stored in the cloud. It includes standard or historical anomaly feature vectors corresponding to different structural anomaly types. The pre-set similarity threshold is a critical similarity value set by the system to determine whether the target spatial feature vector and the anomaly feature vectors in the library belong to the same structural anomaly type. The similarity value is a numerical value that quantifies the similarity between the target spatial feature vector and the anomaly feature vectors in the library, calculated using measurement methods such as cosine similarity. The value is usually between 0 and 1, with the closer the value is to 1, the higher the similarity. The uniformly formatted target spatial feature vectors are generated from bridge data collected by multiple front-end sensors after preprocessing. The system selects anomaly feature vectors from the pre-set bridge anomaly feature vector library stored in the cloud that belong to the same structural anomaly type as the target vector and have a similarity level that reaches the pre-set similarity threshold. Then, using multi-source consistency measurement methods such as cosine similarity, the system calculates the quantitative similarity value between the target spatial feature vector and these selected anomaly feature vectors, thereby completing the determination of whether the bridge structural anomaly corresponding to the target vector matches the known anomaly type and the degree of similarity.

[0051] Furthermore, the sensor credibility weight is determined based on the sensor's own credibility and is used to dynamically adjust the weight value of each sensor in data fusion. Based on the structural anomaly feature parameter, the associated weight value reflecting the contribution of each sensor's credibility to data fusion is extracted from the parameter, providing support for the weight allocation in subsequent multimodal data fusion.

[0052] Furthermore, the target space feature vector is formed by preprocessing and multimodal fusion of bridge data collected by multiple front-end sensors. The abnormal feature vector comes from a preset bridge abnormal feature vector library. The similarity value is calculated using multi-source consistency measurement methods such as cosine similarity. The sensor confidence weight is dynamically adjusted and determined by the system based on factors such as the performance of each sensor and the reliability of the working environment. The confidence weight corresponding to each sensor is multiplied by the similarity value related to that sensor, and then all the product results are summed. The final sum is the feature vector matching metric value between the target space feature vector and the abnormal feature vector.

[0053] S5. Adjust the weight of the data mode corresponding to each sensor according to the metric value of the feature vector matching, and perform weighted fusion of the spatial features corresponding to all sensors according to the adjusted weights to obtain a weighted fusion representation.

[0054] In this embodiment of the invention, the weight of each sensor's corresponding data modality refers to the proportion of importance of the data modalities collected by various sensors such as vision, lidar, sonar, and electromagnetic induction in the process of multimodal fusion analysis when participating in joint characterization generation and realizing health status inference.

[0055] In detail, the feature vector matching quantization value is obtained by calculating cosine similarity to evaluate the consistency of the detection results of different sensors for the same bridge part. Combining sensor confidence, environmental conditions and the attention mechanism in the cross-modal Transformer architecture, the weights of the corresponding data modes of each sensor are adaptively adjusted. When the quantization value is close to 1, the corresponding weight is increased to enhance the confidence of defect judgment. When the consistency is low, the weight is adjusted to achieve mutual calibration of the results between sensors, thereby completing the weight adjustment and improving the reliability of fusion analysis.

[0056] In this embodiment of the invention, the weighted fusion representation refers to the joint representation that reflects the health status of the bridge structure, formed by integrating the spatial features corresponding to all sensors through cross-modal fusion mechanisms such as weighted summation and Transformer attention mechanism, based on the dynamically adjusted modal weights of each sensor data in multimodal fusion analysis.

[0057] In this embodiment of the invention, the step of weighted fusing of the spatial features corresponding to all sensors according to the adjusted weights to obtain a weighted fusion representation includes: The adjusted weights are multiplied with the spatial features corresponding to all sensors to obtain the weighted spatial features of each sensor. Cosine similarity is calculated for the matching degree between all weighted spatial features to obtain the feature matching strength value; Based on the feature matching strength value, feature combinations with a matching degree higher than a preset matching degree threshold are selected from all weighted spatial features, and complementary feature information in the feature combinations is extracted. Based on the complementary information of the features, all weighted spatial features are fused and spliced ​​to obtain a weighted fusion representation.

[0058] In detail, the weighted spatial feature is a feature form that reflects the importance of the sensor data in the fusion analysis after multiplying the spatial features corresponding to each sensor with the dynamically adjusted weights of its own data modes. Based on the multimodal fusion analysis process, the raw data collected by each sensor is preprocessed and converted into feature vectors, and then mapped to a unified vector space by an encoder to obtain spatial features. The weights of the corresponding data modes of each sensor are adjusted by combining sensor reliability, environmental conditions and multi-source consistency measurement results. The weighted spatial features of each sensor are obtained by multiplying the adjusted weights with the spatial features of the corresponding sensors.

[0059] Specifically, the matching degree is the degree of consistency among the weighted spatial features of all sensors for the detection results of the same bridge section; the feature matching strength value is a specific value obtained by quantifying the degree of consistency through the cosine similarity algorithm. A value close to 1 indicates that the multi-source features have similar responses to the same defect and high consistency. After the spatial features corresponding to each sensor are multiplied with the adjusted weights to obtain the weighted spatial features, the cosine similarity calculation rule is applied to the correspondence between all weighted spatial features to obtain the feature matching strength value that reflects the consistency between each weighted spatial feature.

[0060] Furthermore, the preset matching threshold is a benchmark value pre-set by the system to determine whether the matching degree between weighted spatial features meets the standard; the feature combination is a set of several weighted spatial features whose matching degree is higher than the preset threshold; the complementary feature information is information unique to the modal features of different sensors in the feature combination that can complement each other to improve the bridge state perception; after obtaining the feature matching strength value between each weighted spatial feature through cosine similarity calculation, each feature matching strength value is compared with the preset matching threshold, and weighted spatial features with matching degree exceeding the threshold are selected to form a feature combination. Then, complementary feature information unique to each modal feature that can make up for the limitations of single modal information is extracted from the combination.

[0061] Furthermore, after acquiring the weighted spatial features of each sensor and the complementary feature information extracted from them, the weighted spatial features and complementary feature information are combined and fused together by relying on the cross-modal fusion mechanism. Through unified encoding and interactive reasoning of the multimodal deep model, a weighted fusion representation that can accurately and comprehensively reflect the health status of the bridge is formed.

[0062] S6. Analyze the safety level and damage level of the target bridge based on the weighted fusion characterization.

[0063] In this embodiment of the invention, the safety level is a comprehensive risk assessment result derived from the structural health status, damage condition, and environmental and load effects of the target bridge. It is a classification reflecting the overall safety status of the bridge and its ability to bear loads normally. The degree of damage is a qualitative or quantitative description of the severity of specific defects in the bridge structure, which intuitively reflects the degree of impact of the defects on the safety of the bridge structure.

[0064] In detail, based on the analyzed bridge structural safety status data, the logic judgment module comprehensively assesses the location, size, quantity, and overall health status of specific defects, clarifies the severity of various defects such as bridge deck cracks and loose steel cables, and combines environmental and load factors and the impact of defects on structural safety to divide a comprehensive risk assessment result that reflects the overall safety status and load-bearing capacity of the bridge. At the same time, it generates a structured report and early warning information containing relevant information.

[0065] Specifically, the synergistic advantages of edge computing and cloud processing are leveraged. Drones / robots are equipped with high-performance embedded computing platforms responsible for real-time preprocessing and rapid inference, enabling local alarms and handling of emergencies. The cloud deploys large-scale computing resources to perform complex multimodal large-model inference and historical data analysis, and periodically pushes updated model parameters to edge devices. Through edge-cloud-edge collaboration, both the real-time responsiveness of detection results and the ability to utilize greater computing power in the cloud for large-scale model analysis of big data enable adaptive optimization of the system.

[0066] Furthermore, the fusion analysis results are processed by the logical judgment module to assess the safety level and damage risk of the bridge structure. The system generates structured reports and early warning information, which are pushed to bridge maintenance personnel in real time via wireless network or mobile terminal. The report may include detailed defect descriptions, location markings, and risk assessments to assist decision-makers in taking timely maintenance measures.

[0067] As can be seen, in the above scheme, bridge construction data is collected by multiple sensors and multiple modes. Structural anomalies are accurately captured through feature mapping and cross-modal joint modeling. Modal weights are adjusted by quantifying feature vector similarity. After weighted fusion, effective information is fully integrated. This not only achieves comprehensive perception of the bridge's internal and external features and environmental characteristics and reduces the omission of hidden defects, but also improves the targeting and reliability of the analysis through dynamic weight adjustment. It can accurately assess the safety level and degree of damage and solve the problem of low accuracy in bridge safety analysis.

[0068] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0069] In one embodiment, a bridge safety analysis device 100 is provided, which corresponds one-to-one with the bridge safety analysis method described in the above embodiments. For example... Figure 3 As shown, the bridge safety analysis device 100 includes a structural parameter extraction module 101, a feature vector mapping module 102, a structural anomaly feature parameter identification module 103, a feature vector similarity calculation module 104, a spatial feature weighted fusion module 105, and a bridge safety level and damage degree analysis module 106. Detailed descriptions of each functional module are as follows: The building parameter extraction module 101 is used to collect building data of the target bridge using multiple preset sensors according to different data modes, and extract building parameters from the building data collected by the target sensors. The feature vector mapping module 102 is used to convert the building parameters into feature vectors of the data mode corresponding to the target sensor, obtain the feature space corresponding to the data mode of the target sensor, and map the feature vectors into the feature space to obtain the spatial features of the building data collected by the target sensor. The structural anomaly feature parameter identification module 103 is used to perform cross-modal joint modeling of the spatial features corresponding to all sensors to obtain the joint state characterization of the target bridge, and to identify the structural anomaly feature parameters of the target bridge based on the joint state characterization. The feature vector similarity calculation module 104 is used to perform feature vector similarity calculation on the spatial feature vector corresponding to the structural anomaly region of the target bridge and the anomaly feature vector retrieved from the preset bridge anomaly feature vector library based on the structural anomaly feature parameters, and obtain the feature vector matching metric value. The spatial feature weighted fusion module 105 is used to adjust the weight of the data mode corresponding to each sensor according to the feature vector matching metric value, and to perform weighted fusion of the spatial features corresponding to all sensors according to the adjusted weights to obtain a weighted fusion representation. The bridge safety level and damage degree analysis module 106 is used to analyze the safety level and damage degree of the target bridge based on the weighted fusion characterization.

[0070] In one embodiment, the building parameter extraction module 101, when collecting building data of the target bridge using multiple preset sensors according to different data modes, is used to: Identify the correspondence between multiple preset sensors and different data modes; The correspondence is used to control the multiple sensors to collect data from different structural parts of the target bridge, thereby obtaining the sensing data corresponding to the data mode; The sensor data is assimilated to obtain the architectural data of the target bridge.

[0071] In one embodiment, the feature vector mapping module 102, when performing the mapping of the feature vectors to the feature space to obtain the spatial features of the building data collected by the target sensor, is used to: The feature vector is filtered and denoised to obtain a denoised feature vector. The denoised feature vector is then aligned and registered to obtain a registered feature vector. The feature dimensions of the registered feature vector are then uniformly adjusted to a preset value to obtain a dimension-unified feature vector. The dimension-unified feature vector is mapped to the feature space using a preset feature mapping matrix to obtain a mapping space representation. Extract the inherent device parameters and environmental adaptation parameters of the target sensor respectively; Based on the inherent parameters of the device and the environmental adaptation parameters, the spatial coordinates of the mapped spatial representation are calibrated to obtain the spatial characteristics of the building data collected by the target sensor.

[0072] In one embodiment, the structural anomaly feature parameter identification module 103, when performing cross-modal joint modeling of the spatial features corresponding to all sensors to obtain the joint state characterization of the target bridge, is used to: Based on the spatial features corresponding to all sensors, query vectors, key vectors, and value vectors of the data modality corresponding to each sensor are constructed through linear transformations. Calculate the cosine similarity between the query vector corresponding to each data modality and the key vectors corresponding to all data modalities, and use the cosine similarity as the association weight corresponding to each data modality; The cross-modal fusion features are obtained by weighting and summing the value vectors corresponding to all data modalities based on the association weights. The cross-modal fusion features are sequentially subjected to linear transformation and nonlinear activation processing to extract deep interaction features between the data modes, thereby obtaining the joint state representation of the target bridge.

[0073] In one embodiment, the structural anomaly feature parameter identification module 103, when performing the identification of structural anomaly feature parameters of the target bridge based on the joint state characterization, is further configured to: The joint state representation is screened for abnormal features using a preset structural anomaly feature judgment standard to obtain the set of abnormal features of the target bridge; Based on the set of abnormal features, the abnormal regions of the structural parts of the target bridge are located to obtain the structural abnormal regions. Identify the structural anomaly type corresponding to the joint state representation of the structural anomaly region; By combining the structural anomaly types and structural anomaly regions, structural anomaly characteristic parameters of the target bridge are obtained.

[0074] In one embodiment, the feature vector similarity calculation module 104, when performing feature vector similarity calculation on the spatial feature vector corresponding to the structural anomaly region of the target bridge and the anomaly feature vector retrieved from a preset bridge anomaly feature vector library based on the structural anomaly feature parameters, and obtaining a feature vector matching metric value, is used for: Using the aforementioned structural anomaly feature parameters, target spatial feature vectors matching the structural anomaly type are selected from the spatial features corresponding to the structural anomaly region of the target bridge. Search the preset bridge anomaly feature vector library for anomaly feature vectors that belong to the same structural anomaly type as the target spatial feature vector and have a similarity exceeding a preset similarity threshold, and calculate the similarity value between the target spatial feature vector and the anomaly feature vector; Extract the sensor confidence weights associated with the structural anomaly feature parameters; The similarity values ​​are weighted and summed according to the sensor confidence weights to obtain the feature vector matching metric between the target space feature vector and the abnormal feature vector.

[0075] In one embodiment, the spatial feature weighted fusion module 105, when performing weighted fusion of spatial features corresponding to all sensors according to the adjusted weights to obtain a weighted fusion representation, is used for: The adjusted weights are multiplied with the spatial features corresponding to all sensors to obtain the weighted spatial features of each sensor. Cosine similarity is calculated for the matching degree between all weighted spatial features to obtain the feature matching strength value; Based on the feature matching strength value, feature combinations with a matching degree higher than a preset matching degree threshold are selected from all weighted spatial features, and complementary feature information in the feature combinations is extracted. Based on the complementary information of the features, all weighted spatial features are fused and spliced ​​to obtain a weighted fusion representation.

[0076] This invention provides a bridge safety analysis device that collects bridge construction data through multiple sensors and multiple modalities. It accurately captures structural anomalies through feature mapping and cross-modal joint modeling. It combines feature vector similarity quantification to adjust modal weights and integrates effective information through weighted fusion. This not only achieves comprehensive perception of the bridge's internal and external features and environmental characteristics, reducing the omission of hidden defects, but also improves the targeting and reliability of the analysis through dynamic weight adjustment. It can accurately assess the safety level and damage degree, and solve the problem of low accuracy in bridge safety analysis.

[0077] Specific limitations regarding the bridge safety analysis device can be found in the limitations of the bridge safety analysis method described above, and will not be repeated here. Each module in the aforementioned bridge safety analysis device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0078] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 4 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with external clients via a network connection. When the computer program is executed by the processor, it implements the functions or steps of a bridge security analysis method on the server side.

[0079] In one embodiment, a computer device is provided, which may be a client, and its internal structure diagram may be as follows: Figure 5 As shown, the computer device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with an external server via a network connection. When executed by the processor, the computer program implements the functions or steps of a bridge security analysis method on the client side.

[0080] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the following steps: The construction data of the target bridge is collected by multiple preset sensors according to different data modes, and the construction parameters are extracted from the construction data collected by the target sensors. The building parameters are converted into feature vectors corresponding to the data modes of the target sensor, and the feature space corresponding to the data modes of the target sensor is obtained. The feature vectors are mapped into the feature space to obtain the spatial features of the building data collected by the target sensor. Cross-modal joint modeling is performed on the spatial features corresponding to all sensors to obtain the joint state characterization of the target bridge, and the structural anomaly feature parameters of the target bridge are identified based on the joint state characterization. Based on the structural anomaly feature parameters, the spatial feature vector corresponding to the structural anomaly region of the target bridge is compared with the anomaly feature vector retrieved from the preset bridge anomaly feature vector library to calculate the feature vector similarity and obtain the feature vector matching metric value. The weights of the data modes corresponding to each sensor are adjusted according to the metric values ​​of the feature vector matching, and the spatial features corresponding to all sensors are weighted and fused according to the adjusted weights to obtain a weighted fusion representation. The safety level and damage level of the target bridge are analyzed based on the weighted fusion characterization.

[0081] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor: The construction data of the target bridge is collected by multiple preset sensors according to different data modes, and the construction parameters are extracted from the construction data collected by the target sensors. The building parameters are converted into feature vectors corresponding to the data modes of the target sensor, and the feature space corresponding to the data modes of the target sensor is obtained. The feature vectors are mapped into the feature space to obtain the spatial features of the building data collected by the target sensor. Cross-modal joint modeling is performed on the spatial features corresponding to all sensors to obtain the joint state characterization of the target bridge, and the structural anomaly feature parameters of the target bridge are identified based on the joint state characterization. Based on the structural anomaly feature parameters, the spatial feature vector corresponding to the structural anomaly region of the target bridge is compared with the anomaly feature vector retrieved from the preset bridge anomaly feature vector library to calculate the feature vector similarity and obtain the feature vector matching metric value. The weights of the data modes corresponding to each sensor are adjusted according to the metric values ​​of the feature vector matching, and the spatial features corresponding to all sensors are weighted and fused according to the adjusted weights to obtain a weighted fusion representation. The safety level and damage level of the target bridge are analyzed based on the weighted fusion characterization.

[0082] It should be noted that the functions or steps that can be implemented by the computer-readable storage medium or computer device described above can be referred to the relevant descriptions on the server side and client side in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0083] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0084] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0085] It should be noted that if any software tools or components not belonging to our company appear in the embodiments of this application, they are merely for illustrative purposes and do not represent actual use.

[0086] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method of bridge safety analysis, characterized by, The method comprises the following steps: Collecting building data of the target bridge by a plurality of preset sensors according to different data modalities respectively, and extracting building parameters in the building data collected by the target sensor; Converting the building parameters into a feature vector of the data modality corresponding to the target sensor, obtaining a feature space corresponding to the data modality corresponding to the target sensor, mapping the feature vector into the feature space, and obtaining the spatial feature of the building data collected by the target sensor; Joint modeling of the spatial features corresponding to all sensors in a cross-modality manner to obtain a joint state representation of the target bridge, and identifying a structural abnormal feature parameter of the target bridge according to the joint state representation; According to the structural abnormal feature parameter, the spatial feature vector corresponding to the structural abnormal area of the target bridge is calculated with the abnormal feature vector retrieved from the preset bridge abnormal feature vector library, and a feature vector similarity calculation is performed to obtain a feature vector matching degree quantization value; According to the feature vector matching degree quantization value, the weight of each sensor corresponding to the data modality is adjusted, and the spatial features corresponding to all sensors are weighted and fused according to the adjusted weight to obtain a weighted fusion representation; According to the weighted fusion representation, the safety level and damage degree of the target bridge are analyzed.

2. The bridge safety analysis method of claim 1, wherein, The method comprises the following steps: Identifying the correspondence between the plurality of preset sensors and different data modalities; Using the correspondence to control the plurality of sensors to collect data from different structural parts of the target bridge to obtain sensor data corresponding to the data modalities; Performing data format homogenization on the sensor data to obtain the building data of the target bridge.

3. The bridge safety analysis method of claim 1, wherein, The method comprises the following steps: Filtering and denoising the feature vector to obtain a denoised feature vector, aligning and registering the denoised feature vector to obtain a registered feature vector, and uniformly adjusting the feature dimension of the registered feature vector to a preset value to obtain a dimension-unified feature vector; Mapping the dimension-unified feature vector into the feature space using a preset feature mapping matrix to obtain a mapping space representation; Respectively extracting the device inherent parameters and environment adaptation parameters of the target sensor; According to the device inherent parameters and the environment adaptation parameters, the mapping space representation is calibrated in space coordinates to obtain the spatial feature of the building data collected by the target sensor.

4. The bridge safety analysis method of claim 1, wherein, The method comprises the following steps: Based on the spatial features corresponding to all sensors, query vectors, key vectors and value vectors of each sensor corresponding to the data modality are respectively constructed through linear transformation; Calculate the cosine similarity between the query vector corresponding to each data modality and the key vector corresponding to all data modalities, and use the cosine similarity as the correlation weight corresponding to each data modality; According to the correlation weight, the value vectors corresponding to all data modalities are weighted and summed to obtain a cross-modality fusion feature. The cross-modal fusion features are sequentially subjected to linear transformation and nonlinear activation processing, deep interaction features between the data modalities are extracted, and a joint state representation of the target bridge is obtained.

5. The bridge safety analysis method of claim 1, wherein, The structural abnormal feature parameters of the target bridge are identified according to the joint state representation, including: Abnormal feature screening is performed on the joint state representation by using a preset structural abnormal feature determination standard, and an abnormal feature set of the target bridge is obtained; An abnormal region of a structural part of the target bridge is located according to the abnormal feature set, and a structural abnormal region is obtained; The structural abnormal type of the joint state representation corresponding to the structural abnormal region is identified; The structural abnormal type and the structural abnormal region are collected to obtain the structural abnormal feature parameters of the target bridge.

6. The bridge safety analysis method of claim 1, wherein, The feature vector similarity calculation is performed between the spatial feature vector corresponding to the structural abnormal region of the target bridge and the abnormal feature vector retrieved from the preset bridge abnormal feature vector library according to the structural abnormal feature parameters, and a feature vector matching degree quantization value is obtained, including: The target spatial feature vector matching the structural abnormal type is screened out from the spatial feature corresponding to the structural abnormal region of the target bridge by using the structural abnormal feature parameters; The abnormal feature vector belonging to the same structural abnormal type as the target spatial feature vector and having a similarity exceeding a preset similarity threshold is retrieved from the preset bridge abnormal feature vector library, and the similarity value between the target spatial feature vector and the abnormal feature vector is calculated; The sensor reliability weight associated with the structural abnormal feature parameters is extracted; The similarity value is weighted and summed according to the sensor reliability weight to obtain the feature vector matching degree quantization value between the target spatial feature vector and the abnormal feature vector.

7. The bridge safety analysis method of claim 1, wherein, The spatial features corresponding to all sensors are weighted and fused according to the adjusted weights to obtain a weighted fusion representation, including: The weighted spatial features of each sensor are obtained by performing product operation on the spatial features corresponding to all sensors respectively using the adjusted weights; The cosine similarity calculation is performed on the matching degree between all weighted spatial features respectively to obtain a feature matching strength value; The feature combination with a matching degree higher than a preset matching degree threshold is screened out from all weighted spatial features according to the feature matching strength value, and the feature complementary information in the feature combination is extracted; The weighted fusion representation is obtained by fusing and splicing all weighted spatial features according to the feature complementary information.

8. A bridge safety analysis apparatus characterized by comprising: It includes: The building parameter extraction module is used to extract the building parameters in the building data collected by the target sensor by using a plurality of preset sensors to collect the building data of the target bridge according to different data modalities; The feature vector mapping module is used to convert the building parameters into the feature vector of the data modality corresponding to the target sensor, obtain the feature space corresponding to the data modality corresponding to the target sensor, map the feature vector into the feature space, and obtain the spatial feature of the building data collected by the target sensor; The structural anomaly feature parameter identification module is configured to perform cross-modal joint modeling on the spatial features corresponding to all sensors to obtain a joint state representation of the target bridge, and identify a structural anomaly feature parameter of the target bridge according to the joint state representation. The feature vector similarity calculation module is configured to perform feature vector similarity calculation on a spatial feature vector corresponding to a structural anomaly region of the target bridge and an abnormal feature vector retrieved from a preset bridge abnormal feature vector library according to the structural anomaly feature parameter, to obtain a feature vector matching degree quantization value. The spatial feature weighted fusion module is configured to adjust the weight of each data modality corresponding to a sensor according to the feature vector matching degree quantization value, and perform weighted fusion on the spatial features corresponding to all sensors according to the adjusted weight, to obtain a weighted fusion representation. The bridge safety level and damage degree analysis module is configured to analyze the safety level and damage degree of the target bridge according to the weighted fusion representation.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The processor executes the computer program to implement the steps of the bridge safety analysis method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 9. The computer program is executed by the processor to implement the steps of the bridge safety analysis method according to any one of claims 1 to 7.

Images