A bridge structure detection method and system based on 3D Gaussian scattering

Abstract

The present invention relates to the field of bridge detection, and specifically to a bridge structure detection method and system based on 3D Gaussian scattering. The method includes: simultaneously collecting 3D Gaussian scattering data and environmental sensor data of the bridge structure; dynamically fusing the collected 3D Gaussian scattering data with the environmental sensor data based on a pre-trained deep learning model; adjusting the weight of the 3D Gaussian scattering data in real time according to the environmental sensor data, compensating for the influence of environmental factors on the scattering data, and obtaining a fused comprehensive data set; modeling the fused comprehensive data set using a 3D Gaussian function to generate a Gaussian radiation field model; performing intelligent detection on the health status of the bridge structure, obtaining detection results, identifying potential damage and providing maintenance suggestions. The present invention improves the accuracy of bridge detection, reduces the interference of environmental factors, and realizes dynamic monitoring and intelligent evaluation.

Description

Technical Field

[0001] The present invention relates to the field of bridge detection, and in particular to a bridge structure detection method and system based on 3D Gaussian scattering. Background Art

[0002] As an important part of modern transportation infrastructure, the structural safety and reliability of bridges are directly related to the safety and efficiency of transportation. With the acceleration of urbanization and the continuous increase in transportation volume, bridge structures are facing increasingly complex use environments and higher safety requirements. Traditional bridge detection methods mainly rely on manual inspections, simple sensor deployment, and limited image processing technology. Although these methods can provide health status information of bridge structures to a certain extent, they have many limitations, such as low detection accuracy, poor efficiency, strong interference from environmental factors, and difficulty in achieving dynamic monitoring.

[0003] In recent years, with the development of 3D Gaussian scattering technology, deep learning, dynamic data fusion and augmented reality (AR) technology, new technical means have been provided for bridge inspection. 3D Gaussian scattering technology can model the scattering characteristics of bridge structures with high precision, deep learning algorithms can effectively process complex scattering data, dynamic data fusion technology can compensate for the interference of environmental factors on the inspection results, and augmented reality technology can intuitively display the health status of the bridge structure.

[0004] However, the existing technology has not yet deeply integrated these advanced technologies to form an efficient, intelligent and dynamically optimized bridge detection system. Most of the existing detection methods are the application of a single technology, lacking systematicity and synergy, and it is difficult to meet the needs of modern bridge structure complexity and high-precision detection. For example, the existing technology with patent number CN118332655A proposes a road bridge detection method, which obtains road bridge stress distribution images through finite element simulation and non-destructive testing technology, and combines a multimodal convolutional neural network (CNN-LSTM) fusion model for feature extraction and stress analysis. Although this method improves the detection accuracy to a certain extent, it relies on complex finite element simulation and a variety of non-destructive testing technologies, has a large amount of data processing, and is difficult to adapt to dynamic environmental changes.

[0005] In addition, the existing technologies for evaluating the health status of bridge structures are mostly qualitative analyses, lacking quantitative severity assessments and targeted maintenance recommendations. In practical applications, the health status of bridge structures is affected by a variety of environmental factors, such as temperature, humidity, wind speed, and vibration. These environmental factors not only affect the accuracy of the test data, but may also lead to misjudgments or missed judgments. Therefore, how to achieve high-precision, dynamic monitoring and intelligent evaluation in complex environments is a technical problem that needs to be solved in the current field of bridge inspection. Summary of the invention

[0006] The purpose of the present invention is to provide a bridge structure detection method and system based on 3D Gaussian scattering, so as to solve the problems in the prior art of low bridge detection accuracy, great interference from environmental factors, and difficulty in realizing dynamic monitoring and intelligent evaluation.

[0007] To achieve the above purpose, the following technical solution is adopted.

[0008] A bridge structure detection method based on 3D Gaussian scattering, comprising:

[0009] S1, simultaneously collect 3D Gaussian scattering data of the bridge structure and environmental sensor data;

[0010] S2. Based on the pre-trained deep learning model, the collected 3D Gaussian scattering data is dynamically fused with the environmental sensor data; the weight of the 3D Gaussian scattering data is adjusted in real time according to the environmental sensor data to compensate for the impact of environmental factors on the scattering data, and a fused comprehensive data set is obtained;

[0011] S3, using 3D Gaussian function to model the fused comprehensive data set and generate a Gaussian radiation field model;

[0012] S4. Intelligently detect the health status of the bridge structure, obtain test results, identify potential damage and provide maintenance recommendations.

[0013] Optionally, it also includes, S5, using augmented reality technology to visualize the Gaussian radiation field model in real time, generating a three-dimensional reconstruction model of the bridge structure, and superimposing the three-dimensional reconstruction model with the actual bridge structure for display; in the visualization interface, intuitively displaying the health status of the bridge structure through color coding or dynamic labeling.

[0014] Optionally, in the process of modeling the fused comprehensive data set using the 3D Gaussian function, the following steps are also included:

[0015] S31. Adaptively optimize the parameters of the Gaussian radiation field model through a deep learning algorithm. During the optimization process, a closed-loop feedback mechanism is introduced to feed back the received detection results to the modeling process to dynamically adjust the parameters of the 3D Gaussian function.

[0016] Optionally, the 3D Gaussian scattering data includes surface scattering data, internal structure scattering data and environmental scattering data; and the environmental sensor data includes temperature, humidity, wind speed and vibration data.

[0017] Optionally, step S2 specifically includes the following steps:

[0018] Use the pre-trained deep learning model to extract features from the collected 3D Gaussian scattering data and environmental sensor data, and identify the impact pattern of environmental factors on the scattering data; generate a feature vector set containing Gaussian scattering features and environmental impact patterns;

[0019] According to the identified environmental impact mode, the weight of each data point is dynamically calculated, and the weight calculation formula is: Among them, w i is the weight of the i-th data point, f i is the i-th component of the environmental feature vector, α and β are the parameters of the deep learning model obtained through training; in the weight calculation process, an adaptive adjustment mechanism is introduced to dynamically adjust the parameters α and β according to the changes in environmental factors;

[0020] Obtain the dynamic weight value corresponding to each data point to compensate for the impact of environmental factors on scattering data in real time;

[0021] The extracted 3D Gaussian scattering data and the environmental sensor data are weighted fused according to the calculated dynamic weights to generate a fused comprehensive data set.

[0022] Optionally, the step of adaptively optimizing the parameters of the Gaussian radiation field model by using a deep learning algorithm includes:

[0023] S311, based on the preset initial conditions, setting initial parameters for the 3D Gaussian function, including position, covariance matrix, color and transparency, to obtain an initial parameter set of the Gaussian radiation field model;

[0024] S312. Optimizing the parameters of the Gaussian radiation field model using a deep learning algorithm. During the optimization process, the parameters of the Gaussian function are adjusted according to the fused comprehensive data set to minimize the error between the modeling and the actual detection data, thereby generating an optimized Gaussian radiation field model parameter set.

[0025] S313, receiving the detection results, and feeding back the detection results to the Gaussian radiation field modeling process; dynamically adjusting the optimized Gaussian radiation field model parameter set according to the detection results to improve the model's sensitivity to potential damage areas; generating a Gaussian radiation field model parameter set after feedback adjustment.

[0026] Optionally, the specific steps of step S4 include:

[0027] Preprocessing the generated Gaussian radiation field model data to extract feature vectors related to the health status of the bridge structure, wherein the feature vectors include scattering intensity, distribution pattern, location information, and environmental influencing factors; further extracting and encoding the extracted feature vectors to generate high-dimensional feature vectors;

[0028] The extracted high-dimensional feature vector is input into the pre-trained machine classification model to classify the health status and output the health status classification results of the bridge structure, including four levels: normal, slightly damaged, moderately damaged, and severely damaged;

[0029] Using the confidence scoring mechanism, the confidence of each classification result is evaluated, and the areas with confidence below the preset threshold are screened out and marked as areas for further analysis;

[0030] For the marked area to be further analyzed, the damaged area is identified and the features are enhanced, the precise position and shape of the damaged area are output, and the enhanced damage feature vector is generated;

[0031] Using the enhanced damage feature vector and combining it with environmental sensor data, we conduct multi-dimensional feature analysis on the damaged area, extract the expansion trend, depth and area characteristics of the damaged area, generate a quantitative assessment report on the severity of the damaged area, and provide uncertainty analysis results.

[0032] The health status classification results, the quantitative assessment report of the severity of the damaged area and the uncertainty analysis results are integrated as the test results;

[0033] Generate targeted maintenance suggestions based on the test results; the maintenance suggestions include repair priority, repair method and estimated repair time;

[0034] Utilize the preset rule engine, combined with the damage type and severity, to call the corresponding maintenance strategy library; at the same time, introduce a dynamic optimization mechanism to dynamically adjust the maintenance strategy according to the real-time detection results and optimize the maintenance recommendations.

[0035] Optionally, the specific steps of dynamically adjusting the optimized Gaussian radiation field model parameter set according to the detection result in step S313 include:

[0036] Analyze the received detection results and extract characteristic information of the damaged area, including damage type, severity, location information and size information;

[0037] Generate a parameter adjustment strategy based on the extracted damaged area feature information, the parameter adjustment strategy includes:

[0038] According to the position information of the damaged area, the center position of the Gaussian function is adjusted to make it closer to the center of the damaged area;

[0039] According to the size information of the damaged area, the covariance matrix of the Gaussian function is adjusted so that its elliptical shape better matches the contour of the damaged area;

[0040] According to the severity of the damaged area, the color and transparency of the Gaussian function are adjusted to enhance the visibility of the damaged area in the visualization;

[0041] Applying the generated parameter adjustment strategy to the Gaussian radiation field model to dynamically adjust the optimized Gaussian radiation field model parameter set;

[0042] The effect of the adjusted Gaussian radiation field model is verified, and the detection accuracy and sensitivity of the Gaussian radiation field model to the damaged area are evaluated. Based on the verification results, the parameter adjustment strategy is further optimized to ensure the optimal performance of the Gaussian radiation field model.

[0043] A bridge structure detection system based on 3D Gaussian scattering, comprising:

[0044] A data acquisition module for simultaneously acquiring 3D Gaussian scattering data of the bridge structure and environmental sensor data;

[0045] The data processing module is used to dynamically fuse the collected 3D Gaussian scattering data with the environmental sensor data based on the pre-trained deep learning model, adjust the weight of the 3D Gaussian scattering data in real time according to the environmental sensor data, compensate for the influence of environmental factors on the scattering data, and obtain a fused comprehensive data set;

[0046] The modeling module is used to model the fused comprehensive data set using a 3D Gaussian function to generate a Gaussian radiation field model. During the modeling process, the parameters of the Gaussian radiation field model are adaptively optimized through a deep learning algorithm, and a closed-loop feedback mechanism is introduced to feed back the received detection results to the modeling process to dynamically adjust the parameters of the 3D Gaussian function.

[0047] An intelligent detection module is used to perform intelligent detection on the health status of the bridge structure, obtain detection results, identify potential damage and provide maintenance suggestions, send the detection results to the visualization module for visualization processing, and send the detection results to the modeling module;

[0048] The visualization module is used to use augmented reality technology to visualize the Gaussian radiation field model in real time, generate a three-dimensional reconstruction model of the bridge structure, and overlay the three-dimensional reconstruction model with the actual bridge structure for display.

[0049] Optionally, the intelligent detection module includes:

[0050] A feature extraction unit is used to pre-process the generated Gaussian radiation field model data, extract feature vectors related to the health status of the bridge structure, further extract and encode the extracted feature vectors, and generate high-dimensional feature vectors;

[0051] A classification unit, used for inputting the extracted high-dimensional feature vector into a pre-trained machine classification model to perform health status classification and output the health status classification result of the bridge structure;

[0052] A confidence evaluation unit is used to use a confidence scoring mechanism to perform confidence evaluation on each classification result, filter out areas with confidence values ​​lower than a preset threshold, and mark them as areas for further analysis;

[0053] The damage identification unit is used to identify and enhance the features of the damaged area for the marked area to be further analyzed, output the precise position and shape of the damaged area, and generate an enhanced damage feature vector;

[0054] The severity assessment unit is used to use the damage feature vector after feature enhancement and combine it with the environmental sensor data to perform multi-dimensional feature analysis on the damage area, extract the expansion trend, depth and area characteristics of the damage area, generate a quantitative assessment report on the severity of the damage area, and provide uncertainty analysis results;

[0055] An integration unit, used to integrate the health status classification result, the quantitative assessment report of the severity of the damaged area and the uncertainty analysis result as the detection result;

[0056] The maintenance suggestion generation unit is used to generate targeted maintenance suggestions based on the detection results, including repair priority, repair method and estimated repair time, and use the preset rule engine to call the corresponding maintenance strategy library in combination with the damage type and severity. At the same time, by introducing a dynamic optimization mechanism, the maintenance strategy is dynamically adjusted according to the real-time detection results to optimize the maintenance suggestions.

[0057] Compared with the prior art, the present invention has the following beneficial effects:

[0058] The present invention achieves high-precision modeling and dynamic monitoring of bridge structures by integrating 3D Gaussian scattering technology and deep learning algorithms. Through dynamic data fusion technology, the weight of 3D Gaussian scattering data is adjusted in real time to compensate for the interference of environmental factors on the detection results, significantly improving the detection accuracy and reliability; at the same time, the intelligent detection module is used to quantitatively evaluate the health status of the bridge structure, identify potential damage and provide targeted maintenance suggestions, solving the problems of low detection accuracy, large interference from environmental factors and lack of dynamic monitoring capabilities in the existing technology.

[0059] The present invention uses a deep learning model to extract features and dynamically fuse 3D Gaussian scattering data and environmental sensor data to generate a fused comprehensive data set. This multi-dimensional data fusion technology can effectively compensate for the interference of environmental factors on the detection results, improving the quality and reliability of the data; at the same time, through the dynamic adjustment of the weight mechanism, the data fusion process is optimized in real time according to the changes in environmental factors, further improving the detection accuracy.

[0060] The present invention uses a deep learning algorithm to adaptively optimize the parameters of the Gaussian radiation field model and dynamically adjusts the model parameters through a closed-loop feedback mechanism. This intelligent detection technology can quantitatively evaluate the health status of the bridge structure, identify potential damage and generate a detailed detection report; at the same time, through multi-dimensional feature analysis, it can extract the expansion trend, depth and area characteristics of the damaged area, providing a scientific basis for maintenance decisions.

[0061] The present invention uses augmented reality technology to visualize the Gaussian radiation field model in real time, generate a three-dimensional reconstruction model of the bridge structure, and overlay the three-dimensional reconstruction model with the actual bridge structure. This visualization technology can intuitively display the health status of the bridge structure, making it easier for technicians to conduct on-site assessment and maintenance. Through color coding or dynamic labeling, technicians can quickly identify potential damage areas and improve maintenance efficiency.

[0062] The present invention dynamically adjusts the maintenance strategy according to the real-time detection results and optimizes the maintenance suggestions. Through the preset rule engine and dynamic optimization mechanism, combined with the damage type and severity, the corresponding maintenance strategy library is called to provide scientific support for bridge maintenance. This dynamic maintenance strategy optimization technology can significantly improve maintenance efficiency, reduce maintenance costs, and ensure the safety and reliability of bridge structures.

[0063] In summary, the present invention forms an efficient, intelligent and dynamically optimized bridge detection system by deeply integrating 3D Gaussian scattering technology, deep learning, dynamic data fusion and augmented reality technology. The system can significantly improve the detection accuracy and reliability, provide a scientific basis for bridge maintenance, and solve many limitations in the existing technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 It is a schematic diagram of the steps of an embodiment of a bridge structure detection method based on 3D Gaussian scattering of the present invention.

[0065] Figure 2 It is a module schematic diagram of an embodiment of a bridge structure detection system based on 3D Gaussian scattering of the present invention. DETAILED DESCRIPTION

[0066] The present invention will be described in detail below with reference to the accompanying drawings and in combination with embodiments. It should be noted that the embodiments and features in the embodiments of the present application can be combined with each other without conflict.

[0067] The following detailed description is an exemplary description, which is intended to provide further detailed description of the present invention. Unless otherwise specified, all technical terms used in the present invention have the same meaning as those generally understood by those skilled in the art to which the present application belongs. The terms used in the present invention are only for describing specific embodiments, and are not intended to limit exemplary embodiments according to the present invention.

[0068] Example 1

[0069] like Figure 1 As shown, the present invention provides a bridge structure detection method based on 3D Gaussian scattering. The method fuses 3D Gaussian scattering data and environmental sensor data, uses a deep learning model for dynamic fusion and intelligent detection, and can identify potential damage to the bridge structure with high precision and provide maintenance recommendations.

[0070] The following are the specific implementation steps of this method:

[0071] S1, simultaneously collect 3D Gaussian scattering data of the bridge structure and environmental sensor data;

[0072] S2. Based on the pre-trained deep learning model, the collected 3D Gaussian scattering data is dynamically fused with the environmental sensor data; the weight of the 3D Gaussian scattering data is adjusted in real time according to the environmental sensor data to compensate for the impact of environmental factors on the scattering data, and a fused comprehensive data set is obtained;

[0073] S3, using 3D Gaussian function to model the fused comprehensive data set and generate a Gaussian radiation field model;

[0074] S4. Intelligently detect the health status of the bridge structure, obtain test results, identify potential damage and provide maintenance recommendations.

[0075] Specifically, first, 3D Gaussian scattering data and environmental sensor data of the bridge structure are collected simultaneously. 3D Gaussian scattering data includes surface scattering data, internal structure scattering data and environmental scattering data, which are obtained by high-precision scattering sensors. Environmental sensor data includes temperature, humidity, wind speed and vibration data, which are obtained by environmental sensors deployed at key parts of the bridge. The data acquisition module ensures that all sensors work synchronously to obtain a temporally consistent data set.

[0076] Specifically, the collection of 3D Gaussian scattering data is completed by high-precision scattering sensors. These sensors can capture the scattering signals of the surface, internal structure and surrounding environment of the bridge structure. The specific steps are as follows: Select high-precision scattering sensors suitable for bridge detection. These sensors should have high sensitivity and high resolution to capture tiny scattering changes. Arrange multiple scattering sensors at key parts of the bridge (such as piers, bridge decks, beams, etc.) to ensure that the bridge structure can be fully covered. The sensors collect scattering data in real time, including surface scattering data, internal structure scattering data and environmental scattering data. These data reflect the scattering characteristics of the bridge structure under different conditions.

[0077] Environmental sensor data is collected to compensate for the impact of environmental factors on scattered data. Environmental sensors include temperature, humidity, wind speed and vibration sensors. The specific steps are as follows: Select a suitable environmental sensor, such as the BME280 sensor, which can measure temperature, humidity and air pressure at the same time. In addition, you can also choose a dedicated wind speed sensor and vibration sensor. Place environmental sensors at key locations on the bridge to ensure that the environmental conditions of the bridge can be monitored in real time. The sensors collect environmental data in real time, including temperature, humidity, wind speed and vibration data. These data are used in the subsequent data fusion process to compensate for the impact of environmental factors on scattered data.

[0078] To ensure the temporal consistency of the data, all sensors must work in sync. A synchronous trigger mechanism is used to ensure that the scatter sensors and environmental sensors collect data at the same time point. A timestamp is added to each data point so that the data can be accurately aligned in subsequent processing. The collected data is calibrated to ensure data consistency between different sensors.

[0079] The collected raw data needs to be preprocessed to improve data quality and usability. The specific steps are as follows: Remove noise and outliers to ensure data accuracy. Data normalization: Normalize the data to a uniform range for subsequent processing. Data fusion preparation: Align the preprocessed scattering data and environmental data to prepare for subsequent data fusion.

[0080] Through the above steps, the present invention can simultaneously collect 3D Gaussian scattering data and environmental sensor data of the bridge structure, and ensure the temporal consistency and high quality of the data. These data will provide basic support for subsequent dynamic fusion and intelligent detection.

[0081] Next, based on the pre-trained deep learning model, the collected 3D Gaussian scattering data is dynamically fused with the environmental sensor data. In the present invention, data fusion is the process of dynamically fusion of the collected 3D Gaussian scattering data with the environmental sensor data. This process is implemented by a deep learning model (such as improved YOLOv8), and the specific steps are as follows:

[0082] First, a pre-trained deep learning model (such as improved YOLOv8) is used to extract features from the collected 3D Gaussian scattering data and environmental sensor data. The improved YOLOv8 model extracts multi-scale feature maps through its backbone network. These feature maps contain semantic information at different levels and help detect targets of different sizes. At the same time, the model identifies the impact pattern of environmental factors on scattering data and generates a set of feature vectors containing Gaussian scattering features and environmental impact patterns.

[0083] The weight of each data point is dynamically calculated based on the identified environmental impact pattern. The weight calculation formula is as follows:

[0084]

[0085] Among them, w i is the weight of the i-th data point, f i is the i-th component of the environmental feature vector, and α and β are deep learning model parameters obtained through training. An adaptive adjustment mechanism is introduced in the weight calculation process to dynamically adjust parameters α and β according to changes in environmental factors. This adaptive adjustment mechanism is similar to the dynamic weight adjustment in multi-task learning, and can optimize weights in real time according to changes in environmental factors.

[0086] Finally, the extracted 3D Gaussian scattering data and the environmental sensor data are weighted fused according to the calculated dynamic weights to generate a fused comprehensive data set. The weighted fusion process ensures that the impact of environmental factors on the scattering data is compensated, improving the quality and reliability of the data. The details are as follows:

[0087] The 3D Gaussian scattering data and environmental sensor data of each data point are multiplied by their corresponding weights.

[0088] The weighted data are added point by point to generate a fused comprehensive data set.

[0089] Through the above steps, the present invention can effectively fuse 3D Gaussian scattering data and environmental sensor data, compensate for the influence of environmental factors on scattering data, and provide a high-quality data foundation for subsequent modeling and detection.

[0090] The fused comprehensive data set is modeled using a 3D Gaussian function to generate a Gaussian radiation field model. The modeling process of the present invention is to model the fused comprehensive data set using a 3D Gaussian function to generate a Gaussian radiation field model. Each 3D Gaussian function is described by parameters such as position, covariance matrix, color, and transparency. During the modeling process, the parameters of the Gaussian radiation field model are adaptively optimized using a deep learning algorithm. A closed-loop feedback mechanism is introduced during the optimization process to feed back the detection results of the machine learning algorithm to the modeling process, and dynamically adjust the parameters of the 3D Gaussian function. The details are as follows:

[0091] First, based on the preset initial conditions, the initial parameters of the 3D Gaussian function are set, including position, covariance matrix, color, and transparency, to obtain the initial parameter set of the Gaussian radiation field model. The selection of these initial parameters is crucial for the subsequent optimization process. For example, the position parameter can be initialized to the point position in the sparse point cloud, the covariance matrix can be initialized to the unit matrix, and the color and transparency can be preset according to actual needs.

[0092] Next, the deep learning algorithm is used to optimize the parameters of the Gaussian radiation field model. During the optimization process, the parameters of the Gaussian function are adjusted according to the fused comprehensive data set to minimize the error between the modeling and the actual detection data, and generate an optimized Gaussian radiation field model parameter set. This process can be achieved through the following steps:

[0093] Initialize the Gaussian radiation field model using preset initial parameters.

[0094] The fused comprehensive dataset is input into the model.

[0095] Calculate the error between the model output and the actual detection data.

[0096] Based on the error, the model parameters are adjusted using deep learning algorithms such as gradient descent.

[0097] Repeat the above steps until the error reaches the preset threshold or the maximum number of iterations is reached.

[0098] Finally, the test results are received and fed back to the Gaussian radiation field modeling process. The optimized Gaussian radiation field model parameter set is dynamically adjusted according to the test results to improve the model's sensitivity to potential damage areas and generate a Gaussian radiation field model parameter set that has been adjusted after feedback. This process can be achieved through the following steps:

[0099] The detection results are analyzed to extract the characteristic information of the potential damage area.

[0100] The detection results are fed back into the modeling process to adjust the model parameters according to the characteristic information of the damaged area.

[0101] According to the feedback information, the parameters of the Gaussian function, such as position, covariance matrix, color and transparency, are dynamically adjusted.

[0102] The Gaussian radiation field model is updated to generate a set of parameters adjusted by feedback.

[0103] In order to optimize the parameters of the Gaussian to represent the scene, the Gaussian needs to be rendered into the image in a differentiable way. The specific method is to integrally project the influence function of the 3D Gaussian along the depth dimension and approximate it as a 2D Gaussian influence function in pixel coordinates. In this way, the dynamic adjustment of the Gaussian parameters can be achieved, ensuring that each Gaussian represents the same physical area while moving dynamically in time, thereby achieving 6-DOF tracking of the entire scene.

[0104] Dynamic adjustment: During the optimization process, the parameters α and β are dynamically adjusted according to changes in environmental factors to ensure that the model can adapt to different environmental conditions.

[0105] During the first time step, all parameters of the Gaussians are optimized. After the first time step, the size, color, opacity, and background logic of the Gaussians are fixed, and only the position and rotation are updated.

[0106] Through the above steps, the present invention can use 3D Gaussian functions to perform high-precision modeling of bridge structures, and dynamically optimize model parameters through deep learning algorithms and closed-loop feedback mechanisms, thereby improving the model's sensitivity to potential damage areas and detection accuracy.

[0107] The intelligent detection process of the present invention is to conduct a high-precision assessment of the health status of the bridge structure, identify potential damage and provide targeted maintenance suggestions. The following are the specific implementation steps:

[0108] 1. Data preprocessing and feature extraction

[0109] First, the generated Gaussian radiation field model data is preprocessed to extract the feature vectors related to the health status of the bridge structure. The feature vectors include scattering intensity, distribution pattern, location information and environmental influencing factors. The details are as follows:

[0110] Normalize the Gaussian radiation field model data to a uniform range for easy subsequent processing.

[0111] A deep learning model (such as improved YOLOv8) is used to extract features from the normalized data to generate a feature vector containing scattering intensity, distribution pattern, location information, and environmental influencing factors.

[0112] The extracted feature vectors are further encoded to generate high-dimensional feature vectors to better represent the complexity of the data.

[0113] 2. Health status classification

[0114] The extracted high-dimensional feature vector is input into the pre-trained machine classification model to classify the health status. The specific steps are as follows:

[0115] Choose an appropriate machine learning classification model, such as a support vector machine (SVM) or a deep neural network.

[0116] Based on the input high-dimensional feature vector, the model outputs the classification results of the health status of the bridge structure, including four levels: normal, slightly damaged, moderately damaged, and severely damaged.

[0117] 3. Confidence Assessment and Screening

[0118] Using the confidence scoring mechanism, each classification result is evaluated for confidence, and areas with confidence below the preset threshold are screened out and marked as areas for further analysis. The specific steps are as follows:

[0119] Calculate the confidence score for each classification result to evaluate the model's confidence in the classification result.

[0120] Set a confidence threshold, filter out areas with confidence levels below the threshold, and mark them as areas that require further analysis.

[0121] 4. Damage area identification and feature enhancement

[0122] For the marked areas to be further analyzed, the damaged areas are identified and their features are enhanced. The specific steps are as follows:

[0123] The marked areas are further analyzed using deep learning models (such as improved YOLOv8) to identify the precise location and shape of the damaged area.

[0124] The identified damaged area is feature enhanced to generate an enhanced damage feature vector to better highlight the damaged area.

[0125] 5. Multi-dimensional feature analysis and quantitative evaluation

[0126] Using the enhanced damage feature vector and combining it with environmental sensor data, we can perform multi-dimensional feature analysis on the damaged area. The specific steps are as follows:

[0127] Extract the expansion trend, depth and area characteristics of the damaged area and generate a quantitative assessment report on the severity of the damaged area.

[0128] Provide uncertainty analysis results to evaluate the reliability of test results.

[0129] 6. Test results integration

[0130] The health status classification results, the quantitative assessment report of the severity of the damaged area and the uncertainty analysis results are integrated as the final detection results.

[0131] 7.Maintenance suggestion generation

[0132] Generate targeted maintenance recommendations based on the test results. The specific steps are as follows:

[0133] Utilize the preset rule engine, combined with the damage type and severity, to call the corresponding maintenance strategy library.

[0134] Introduce a dynamic optimization mechanism to dynamically adjust maintenance strategies based on real-time detection results and optimize maintenance recommendations.

[0135] Generate maintenance recommendations that include repair priority, repair method, and estimated repair time.

[0136] Through the above steps, the present invention can perform high-precision assessment of the health status of the bridge structure, identify potential damage, and provide scientific maintenance recommendations to ensure the safety and reliability of the bridge.

[0137] As a preferred example, during the modeling process, the parameters of the Gaussian radiation field model are dynamically adjusted according to the detection results. The specific steps are as follows:

[0138] 1. Analysis of test results

[0139] First, the received detection results are analyzed to extract the characteristic information of the damaged area, including damage type, severity, location information and size information. The specific steps are as follows:

[0140] The damaged area marking information in the detection results is extracted, including damage type (such as cracks, corrosion, etc.), severity (such as slight, moderate, severe), location information (such as coordinate points) and size information (such as length, width).

[0141] For each damaged area, its feature vector is extracted, including scattering intensity, distribution pattern, location information and environmental influencing factors.

[0142] 2. Parameter adjustment strategy generation

[0143] Generate parameter adjustment strategy based on the extracted damage area feature information. The specific steps are as follows:

[0144] According to the position information of the damaged area, the center position of the Gaussian function is adjusted to make it closer to the center of the damaged area.

[0145] The centroid coordinates of the damaged area are calculated, and the center position of the Gaussian function is adjusted to the centroid position.

[0146] According to the size information of the damaged area, the covariance matrix of the Gaussian function is adjusted so that its elliptical shape better matches the contour of the damaged area.

[0147] According to the length and width of the damaged area, the lengths of the major and minor axes of the covariance matrix were adjusted to ensure that the Gaussian ellipse could cover the entire damaged area.

[0148] According to the severity of the damaged area, the color and transparency of the Gaussian function are adjusted to enhance the visibility of the damaged area in the visualization.

[0149] For severely damaged areas, increase the color saturation and transparency to make them more prominent in visualization; for slightly damaged areas, reduce the color saturation and transparency appropriately.

[0150] 3. Parameter adjustment application

[0151] The generated parameter adjustment strategy is applied to the Gaussian radiation field model to dynamically adjust the optimized Gaussian radiation field model parameter set. The specific steps are as follows:

[0152] According to the generated adjustment strategy, the parameters of the Gaussian function are updated, including position, covariance matrix, color, and transparency.

[0153] The updated parameters are applied to the Gaussian radiation field model to generate a new model representation.

[0154] 4. Effect verification and optimization

[0155] Verify the effect of the adjusted Gaussian radiation field model, evaluate the detection accuracy and sensitivity of the Gaussian radiation field model to the damaged area, and further optimize the parameter adjustment strategy based on the verification results to ensure the optimal performance of the Gaussian radiation field model. The specific steps are as follows:

[0156] Prepare a set of validation data with known damage areas to evaluate the detection accuracy of the model.

[0157] Calculate the detection accuracy of the adjusted model on the validation data, including recall, precision, and F1 score.

[0158] Evaluate the model's sensitivity to injury areas of varying severity to ensure that the model can accurately identify minor injuries.

[0159] According to the evaluation results, the parameter adjustment strategy is further optimized, such as adjusting the parameters α and β in the weight calculation formula.

[0160] Through the above steps, the present invention can detect the health status of the bridge structure with high precision, identify potential damage, and provide scientific maintenance suggestions, which significantly improves the efficiency and reliability of bridge detection.

[0161] Example 2

[0162] like Figure 2 As shown, the detection system of the present invention includes the following modules:

[0163] The data acquisition module is used to simultaneously collect 3D Gaussian scattering data and environmental sensor data of the bridge structure. The 3D Gaussian scattering data is collected by high-precision sensors, including surface scattering, internal structure scattering, and environmental scattering data. Environmental sensor data includes temperature, humidity, wind speed, and vibration data, which are obtained by sensors deployed at key locations on the bridge. The data acquisition module ensures that all sensors work synchronously to obtain a temporally consistent data set.

[0164] The data processing module dynamically fuses the collected 3D Gaussian scattering data with the environmental sensor data based on a pre-trained deep learning model (such as the improved YOLOv8). The weight of the 3D Gaussian scattering data is adjusted in real time according to the environmental sensor data to compensate for the impact of environmental factors on the scattering data, and a fused comprehensive data set is obtained. The weight calculation formula is as follows:

[0165]

[0166] Among them, w i is the weight of the i-th data point, f i is the i-th component of the environmental feature vector, α and β are the parameters of the deep learning model obtained through training. An adaptive adjustment mechanism is introduced in the weight calculation process to dynamically adjust the parameters α and β according to the changes in environmental factors.

[0167] The modeling module uses the 3D Gaussian function to model the fused comprehensive data set and generate a Gaussian radiation field model. During the modeling process, the parameters of the Gaussian radiation field model are adaptively optimized through a deep learning algorithm, and a closed-loop feedback mechanism is introduced to feed back the received detection results to the modeling process to dynamically adjust the parameters of the 3D Gaussian function. The specific steps include:

[0168] Set the initial parameters for the 3D Gaussian function, including position, covariance matrix, color, and transparency.

[0169] The deep learning algorithm is used to optimize the parameters of the Gaussian radiation field model to minimize the error between the modeling and the actual detection data.

[0170] The optimized Gaussian radiation field model parameter set is dynamically adjusted according to the detection results to improve the model's sensitivity to potential damage areas.

[0171] The intelligent detection module is used to perform intelligent detection of the health status of the bridge structure, obtain detection results, identify potential damage and provide maintenance suggestions. The detection results are sent to the visualization module for visualization processing and fed back to the modeling module. The intelligent detection module includes the following units:

[0172] Feature extraction unit: preprocesses the generated Gaussian radiation field model data, extracts feature vectors related to the health status of the bridge structure, and generates high-dimensional feature vectors.

[0173] Classification unit: The extracted high-dimensional feature vector is input into the pre-trained machine classification model to output the classification result of the health status of the bridge structure.

[0174] Confidence Assessment Unit: Utilizes the confidence scoring mechanism to filter out areas with confidence levels below a preset threshold and mark them as areas for further analysis.

[0175] Damage identification unit: performs damage identification and feature enhancement on these areas, and outputs the precise location and shape of the damaged area.

[0176] Severity assessment unit: Combined with environmental sensor data, it conducts multi-dimensional feature analysis on the damaged area and generates a quantitative severity assessment report.

[0177] Integration unit: Integrates the classification results, damage assessment report and uncertainty analysis results as the final detection results.

[0178] Maintenance suggestion generation unit: Generates maintenance suggestions based on the test results, including repair priority, repair method and estimated repair time. Uses the preset rule engine, combined with the damage type and severity, to call the corresponding maintenance strategy library.

[0179] The visualization module uses augmented reality technology to visualize the Gaussian radiation field model in real time, generate a 3D reconstruction model of the bridge structure, and overlay the 3D reconstruction model with the actual bridge structure. In the visualization interface, the health status of the bridge structure is intuitively displayed through color coding or dynamic annotation.

[0180] It is known from common technical knowledge that the present invention can be implemented by other embodiments that do not deviate from its spirit or essential features. Therefore, the above disclosed embodiments are only illustrative in all respects and are not exclusive. All changes within the scope of the present invention or within the scope equivalent to the present invention are included in the present invention.

Claims

1. A bridge structure detection method based on 3D Gaussian scattering, characterized in that: include, S1, simultaneously collect 3D Gaussian scattering data of the bridge structure and environmental sensor data; S2. Based on the pre-trained deep learning model, the collected 3D Gaussian scattering data is dynamically fused with the environmental sensor data; the weight of the 3D Gaussian scattering data is adjusted in real time according to the environmental sensor data to compensate for the impact of environmental factors on the scattering data, and a fused comprehensive data set is obtained; S3, using 3D Gaussian function to model the fused comprehensive data set and generate a Gaussian radiation field model; S4. Intelligently detect the health status of the bridge structure, obtain test results, identify potential damage and provide maintenance suggestions; In the process of modeling the fused comprehensive data set using the 3D Gaussian function, The following steps are included: S31. Adaptively optimize the parameters of the Gaussian radiation field model through a deep learning algorithm; during the optimization process, introduce a closed-loop feedback mechanism to feed back the received detection results to the modeling process and dynamically adjust the parameters of the 3D Gaussian function; The specific steps of step S4 include: Preprocessing the generated Gaussian radiation field model data to extract feature vectors related to the health status of the bridge structure, wherein the feature vectors include scattering intensity, distribution pattern, location information, and environmental influencing factors; further extracting and encoding the extracted feature vectors to generate high-dimensional feature vectors; The extracted high-dimensional feature vector is input into the pre-trained machine classification model to classify the health status and output the health status classification results of the bridge structure, including four levels: normal, slightly damaged, moderately damaged, and severely damaged; Using the confidence scoring mechanism, the confidence of each classification result is evaluated, and the areas with confidence below the preset threshold are screened out and marked as areas for further analysis; For the marked area to be further analyzed, the damaged area is identified and the features are enhanced, the precise position and shape of the damaged area are output, and the enhanced damage feature vector is generated; Using the enhanced damage feature vector and combining it with environmental sensor data, we conduct multi-dimensional feature analysis on the damaged area, extract the expansion trend, depth and area characteristics of the damaged area, generate a quantitative assessment report on the severity of the damaged area, and provide uncertainty analysis results. The health status classification results, the quantitative assessment report of the severity of the damaged area and the uncertainty analysis results are integrated as the test results; Generate targeted maintenance recommendations based on the test results; The maintenance proposal includes repair priority, repair method and estimated repair time; Utilize the preset rule engine, combined with the damage type and severity, to call the corresponding maintenance strategy library; at the same time, introduce a dynamic optimization mechanism to dynamically adjust the maintenance strategy according to the real-time detection results and optimize the maintenance recommendations.

2. The bridge structure detection method based on 3D Gaussian scattering according to claim 1 is characterized in that: It also includes, S5, using augmented reality technology to visualize the Gaussian radiation field model in real time, generating a three-dimensional reconstruction model of the bridge structure, and superimposing the three-dimensional reconstruction model with the actual bridge structure; in the visualization interface, intuitively displaying the health status of the bridge structure through color coding or dynamic labeling.

3. The bridge structure detection method based on 3D Gaussian scattering according to claim 1 is characterized in that: The 3D Gaussian scattering data includes surface scattering data, internal structure scattering data and environmental scattering data; the environmental sensor data includes temperature, humidity, wind speed and vibration data.

4. The bridge structure detection method based on 3D Gaussian scattering according to claim 1 is characterized in that: Step S2 specifically The following steps are included: Use the pre-trained deep learning model to extract features from the collected 3D Gaussian scattering data and environmental sensor data, and identify the impact pattern of environmental factors on the scattering data; generate a feature vector set containing Gaussian scattering features and environmental impact patterns; According to the identified environmental impact mode, the weight of each data point is dynamically calculated, and the weight calculation formula is: Among them, w i is the weight of the ith data point, f i is the i-th component of the environmental feature vector, α and β are the parameters of the deep learning model obtained through training; in the weight calculation process, an adaptive adjustment mechanism is introduced to dynamically adjust the parameters α and β according to the changes in environmental factors; Obtain the dynamic weight value corresponding to each data point to compensate for the impact of environmental factors on scattering data in real time; The extracted 3D Gaussian scattering data and the environmental sensor data are weighted fused according to the calculated dynamic weights to generate a fused comprehensive data set.

5. The bridge structure detection method based on 3D Gaussian scattering according to claim 1 is characterized in that: The step of adaptively optimizing the parameters of the Gaussian radiation field model by using a deep learning algorithm comprises: S311, based on the preset initial conditions, setting initial parameters for the 3D Gaussian function, including position, covariance matrix, color and transparency, to obtain an initial parameter set of the Gaussian radiation field model; S312. Optimizing the parameters of the Gaussian radiation field model using a deep learning algorithm. During the optimization process, the parameters of the Gaussian function are adjusted according to the fused comprehensive data set to minimize the error between the modeling and the actual detection data, thereby generating an optimized Gaussian radiation field model parameter set. S313, receiving the detection results, and feeding back the detection results to the Gaussian radiation field modeling process; dynamically adjusting the optimized Gaussian radiation field model parameter set according to the detection results to improve the model's sensitivity to potential damage areas; generating a Gaussian radiation field model parameter set after feedback adjustment.

6. The bridge structure detection method based on 3D Gaussian scattering according to claim 5 is characterized in that: The specific steps of dynamically adjusting the optimized Gaussian radiation field model parameter set according to the detection result in step S313 include: Analyze the received detection results and extract characteristic information of the damaged area, including damage type, severity, location information and size information; Generate a parameter adjustment strategy based on the extracted damaged area feature information, the parameter adjustment strategy includes: According to the position information of the damaged area, the center position of the Gaussian function is adjusted to make it closer to the center of the damaged area; According to the size information of the damaged area, the covariance matrix of the Gaussian function is adjusted so that its elliptical shape better matches the contour of the damaged area; According to the severity of the damaged area, the color and transparency of the Gaussian function are adjusted to enhance the visibility of the damaged area in the visualization; Applying the generated parameter adjustment strategy to the Gaussian radiation field model to dynamically adjust the optimized Gaussian radiation field model parameter set; The effect of the adjusted Gaussian radiation field model is verified, and the detection accuracy and sensitivity of the Gaussian radiation field model to the damaged area are evaluated. Based on the verification results, the parameter adjustment strategy is further optimized to ensure the optimal performance of the Gaussian radiation field model.

7. A bridge structure detection system based on 3D Gaussian scattering, based on a bridge structure detection method based on 3D Gaussian scattering according to any one of claims 1 to 6, characterized in that: include, A data acquisition module for simultaneously acquiring 3D Gaussian scattering data of the bridge structure and environmental sensor data; The data processing module is used to dynamically fuse the collected 3D Gaussian scattering data with the environmental sensor data based on the pre-trained deep learning model, adjust the weight of the 3D Gaussian scattering data in real time according to the environmental sensor data, compensate for the influence of environmental factors on the scattering data, and obtain a fused comprehensive data set; The modeling module is used to model the fused comprehensive data set using a 3D Gaussian function to generate a Gaussian radiation field model. During the modeling process, the parameters of the Gaussian radiation field model are adaptively optimized through a deep learning algorithm, and a closed-loop feedback mechanism is introduced to feed back the received detection results to the modeling process to dynamically adjust the parameters of the 3D Gaussian function. An intelligent detection module is used to perform intelligent detection on the health status of the bridge structure, obtain detection results, identify potential damage and provide maintenance suggestions, send the detection results to the visualization module for visualization processing, and send the detection results to the modeling module; A visualization module is used to visualize the Gaussian radiation field model in real time using augmented reality technology, generate a three-dimensional reconstruction model of the bridge structure, and overlay the three-dimensional reconstruction model with the actual bridge structure for display; The intelligent detection module comprises: A feature extraction unit is used to pre-process the generated Gaussian radiation field model data, extract feature vectors related to the health status of the bridge structure, further extract and encode the extracted feature vectors, and generate high-dimensional feature vectors; A classification unit, used for inputting the extracted high-dimensional feature vector into a pre-trained machine classification model to perform health status classification and output the health status classification result of the bridge structure; A confidence evaluation unit is used to use a confidence scoring mechanism to perform confidence evaluation on each classification result, filter out areas with confidence values ​​lower than a preset threshold, and mark them as areas for further analysis; The damage identification unit is used to identify and enhance the features of the damaged area for the marked area to be further analyzed, output the precise position and shape of the damaged area, and generate an enhanced damage feature vector; The severity assessment unit is used to use the damage feature vector after feature enhancement and combine it with the environmental sensor data to perform multi-dimensional feature analysis on the damage area, extract the expansion trend, depth and area characteristics of the damage area, generate a quantitative assessment report on the severity of the damage area, and provide uncertainty analysis results; The integration unit is used to integrate the health status classification results, the quantitative assessment report of the severity of the damaged area and the uncertainty analysis results as the detection results; the maintenance suggestion generation unit is used to generate targeted maintenance suggestions based on the detection results, including repair priority, repair method and estimated repair time, and use the preset rule engine to call the corresponding maintenance strategy library in combination with the damage type and severity. At the same time, by introducing a dynamic optimization mechanism, the maintenance strategy is dynamically adjusted according to the real-time detection results to optimize the maintenance suggestions.

Images