Civil engineering structure defect detection method and system based on image processing technology

Abstract

The invention relates to the field of constructional engineering, in particular to a civil engineering structure defect detection method and system based on an image processing technology, and the method comprises the steps: data collection, preprocessing, generative adversarial network processing of crack multi-modal data, adaptive dynamic weight three-dimensional reconstruction and crack complexity evaluation. Generating a three-dimensional crack model and estimating the depth and the width of the crack by generating a crack visual form, a temperature characteristic and an acoustic wave characteristic, dynamically evaluating a crack risk level in combination with a complexity index, and preferentially processing a high-risk area; according to the invention, through fusion of multi-modal data, higher-precision and automatic civil engineering structure crack detection is realized, and the defects of multi-scale fusion and accurate crack positioning in the prior art are effectively overcome.

Description

Technical Field

[0001] The present invention relates to the field of construction engineering, and in particular to a method and system for detecting defects in civil engineering structures based on image processing technology. Background Art

[0002] In the field of modern construction engineering, the safety and stability of civil engineering structures are of vital importance. In particular, crack detection in concrete buildings is particularly important because the presence of cracks may affect the structural integrity and increase construction risks. Therefore, early detection and evaluation of cracks in civil engineering structures are particularly important. Civil engineering structure defect detection based on image processing technology has become one of the main means in the prior art. However, the prior art (Chinese invention patent, publication number: CN118379270A) such as the CA saliency detection algorithm has the limitation of insufficient accuracy. It relies on pixel grayscale differences and distances for judgment and fails to effectively handle errors caused by complex textures on the building surface. In addition, when calculating the saliency of multi-scale layers, the weight distribution is unreasonable, and information is lost during image downsampling, which further reduces the detection accuracy. These technical defects make the existing methods perform poorly in dealing with complex building environments and are difficult to meet the needs of high-precision detection. Summary of the invention

[0003] In view of the many problems existing in the above-mentioned prior art, the present invention provides a method and system for detecting defects in civil engineering structures based on image processing technology. The present invention processes multimodal data of building cracks, including visible light images, thermal imaging, and acoustic reflection data, by generating adversarial networks and adaptive dynamic weight 3D reconstruction technology, generates 3D crack model data, and evaluates the complexity and risk level of cracks, providing more accurate detection results. The comprehensive analysis of crack depth, width, and complexity index effectively improves the accuracy and efficiency of detection.

[0004] A method for detecting defects in civil engineering structures based on image processing technology comprises the following steps:

[0005] Collect visible light image data, thermal imaging data and acoustic wave reflection data of civil engineering structures, and perform denoising and normalization preprocessing on the collected data to generate crack edge data, temperature anomaly area data and internal defect location data;

[0006] The crack edge data, temperature anomaly area data and internal defect location data are processed by generative adversarial networks to generate visual morphological data, temperature characteristic data and acoustic wave characteristic data of simulated cracks, and crack matching degree data is generated through morphological matching algorithm and reverse reasoning;

[0007] Based on the crack matching data, the adaptive dynamic weight 3D reconstruction technology is used to generate 3D crack model data by combining visual morphology data, temperature feature data and acoustic wave feature data. The 3D crack model data is used to estimate the depth and width of the cracks, and the crack depth data and crack width data are generated.

[0008] Based on the crack depth data, crack width data, three-dimensional crack model data, temperature characteristic data and acoustic wave characteristic data, the crack complexity index is calculated, the risk level of the crack is evaluated, the crack risk level data is generated, and the crack areas for priority treatment are identified.

[0009] Preferably, the preprocessing of data includes:

[0010] De-noising the visible light image data, using a bilateral filtering algorithm to perform image de-noising to retain the sharp features of the crack edge;

[0011] Performing temperature normalization on the thermal imaging data, and adjusting the distribution of temperature values ​​through linear mapping;

[0012] The acoustic wave reflection data is subjected to denoising, and the interference of environmental noise is eliminated by using time delay difference calculation to generate acoustic wave characteristic signals for identifying internal defects of civil engineering structures.

[0013] Preferably, the training process of the generative adversarial network includes:

[0014] The generator generates visual morphological data of simulated cracks, temperature characteristic data of simulated cracks and acoustic wave characteristic data of simulated cracks through crack edge data, temperature abnormality area data and internal defect position data;

[0015] The discriminator is used to determine the matching degree between the generated simulated crack data and the actual crack data. The generator and the discriminator are interactively trained by optimizing the loss function, so that the difference between the generated simulated crack data and the actual crack data is gradually reduced.

[0016] Preferably, the crack matching data is calculated by the following formula:

[0017] M=α·V m +β·T m +γ·S m

[0018] Where M represents the crack matching degree; V m Indicates the visual morphology matching degree; T m Indicates the temperature characteristic matching degree; S m Indicates the matching degree of acoustic wave features; α, β, and γ are the weight coefficients of visual morphology, temperature characteristics, and acoustic wave characteristics, respectively. The weight coefficients are dynamically adjusted according to the confidence of the actual crack characteristics.

[0019] Preferably, the adaptive dynamic weighted 3D reconstruction comprises:

[0020] Initialize the weight parameters of each modal data;

[0021] The SIFT feature point matching algorithm is used to extract the key points of the crack at different viewing angles and generate a preliminary three-dimensional point cloud of the crack.

[0022] The depth and width parameters of the crack are dynamically adjusted according to the temperature characteristic data and the acoustic wave characteristic data to generate optimized three-dimensional crack model data.

[0023] Preferably, the crack depth estimation is performed by a convolutional neural network, wherein the optimized three-dimensional crack model data is input, segmented according to different depth layers, and the depth changes of the cracks are calculated layer by layer to generate crack depth data.

[0024] Preferably, the crack width is estimated by analyzing the cross-sectional morphology of the three-dimensional crack model to generate crack width data, thereby ensuring that the widths on different cross sections are accurately measured.

[0025] Preferably, the fracture complexity index is calculated by the following formula:

[0026]

[0027] Among them, C represents the crack complexity index; L represents the crack length; D represents the crack depth; B represents the number of crack branches; S represents the morphological complexity of the crack; T represents the temperature anomaly index; V represents the acoustic wave characteristic index, and w1, w2, w3, w4, and w5 are the weight coefficients of crack length, crack depth, crack branching, temperature anomaly, and acoustic wave characteristics, respectively.

[0028] Preferably, the fracture risk level converts the fracture complexity index into a risk level through nonlinear mapping, and the risk level is divided into low risk, medium risk and high risk, and the priority treatment area is identified to generate priority treatment area data.

[0029] A system for implementing the method for detecting defects in civil engineering structures based on image processing technology, the system comprising:

[0030] An image acquisition module for acquiring visible light image data, thermal imaging data, and acoustic wave reflection data of civil engineering structures;

[0031] A data preprocessing module, communicating with the image acquisition module, is used to perform denoising and normalization processing on the visible light image data, thermal imaging data and acoustic wave reflection data to generate crack edge data, temperature abnormality area data and internal defect location data;

[0032] A generative adversarial network module is used to generate visual morphological data of simulated cracks, temperature characteristic data of simulated cracks and acoustic wave characteristic data of simulated cracks according to the crack edge data, temperature abnormality area data and internal defect position data, and to generate crack matching degree data through morphological matching algorithm and reverse reasoning;

[0033] A three-dimensional reconstruction module is used to generate three-dimensional crack model data based on the crack matching data, in combination with the visual morphology data of the simulated crack, the temperature characteristic data of the simulated crack and the acoustic wave characteristic data of the simulated crack by using an adaptive dynamic weight three-dimensional reconstruction technology, and to estimate the depth and width of the crack by using the three-dimensional crack model data to generate crack depth data and crack width data;

[0034] A complexity index calculation module, used to calculate the crack complexity index according to the crack depth data, crack width data, three-dimensional crack model data, temperature characteristic data and acoustic wave characteristic data;

[0035] The risk assessment module is used to assess the risk level of the cracks according to the crack complexity index, generate crack risk level data, and identify crack areas for priority treatment.

[0036] Compared with the prior art, the advantages and beneficial effects of the present invention are:

[0037] The present invention realizes efficient and accurate crack detection by adopting generative adversarial network (GAN), adaptive dynamic weight 3D reconstruction technology, and crack complexity index calculation method;

[0038] The present invention processes the visual morphology, temperature and acoustic wave characteristics of cracks by generating adversarial networks to generate crack matching data, and then uses adaptive weight technology to combine different modal data to generate three-dimensional crack model data to ensure the accuracy of crack detection and depth assessment;

[0039] The introduction of the crack complexity index of the present invention enables the detection system to dynamically adjust the risk assessment criteria and give priority to high-risk crack areas, solving the important problem of the failure to accurately process multi-dimensional information and misjudgment of defects in the prior art. Through this technical means, the problems of insufficient accuracy of existing detection algorithms and limited multimodal data processing capabilities are solved, and the automation and reliability of civil engineering structure defect detection are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a schematic diagram of the process of the present invention;

[0041] Figure 2 A structural diagram of the reconstruction of the three-dimensional fracture model in the present invention;

[0042] Figure 3This is a schematic diagram of the process flow of generating an adversarial network (GAN) in the present invention;

[0043] Figure 4 The fracture complexity index and weight distribution diagram in the present invention

[0044] Figure 5 This is a system structure block diagram of the present invention. DETAILED DESCRIPTION

[0045] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present disclosure. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure.

[0046] A method for detecting defects in civil engineering structures based on image processing technology comprises the following steps:

[0047] Collect visible light image data, thermal imaging data and acoustic wave reflection data of civil engineering structures, and perform denoising and normalization preprocessing on the collected data to generate crack edge data, temperature anomaly area data and internal defect location data;

[0048] Preferably, the preprocessing of data includes:

[0049] De-noising the visible light image data, using a bilateral filtering algorithm to perform image de-noising to retain the sharp features of the crack edge;

[0050] Performing temperature normalization operation on the thermal imaging data, and adjusting the distribution of temperature values ​​through linear mapping;

[0051] The acoustic wave reflection data is subjected to denoising, and the interference of environmental noise is eliminated by using time delay difference calculation to generate acoustic wave characteristic signals for identifying internal defects of civil engineering structures.

[0052] In the present invention, cracks and potential defects in civil engineering structures are identified through the collection and preprocessing of multimodal data. Specifically, the system first synchronously collects visible light images, thermal imaging data, and acoustic wave reflection data through multiple sensors. These three types of data each provide observation information on different dimensions of cracks. Among them, visible light image data is used to obtain the geometric shape of the building surface, which can directly capture the appearance and edge characteristics of the cracks; thermal imaging data uses infrared detection technology to obtain the surface and internal temperature distribution. Because the location of the cracks is usually accompanied by thermal anomalies, this type of data can effectively assist in discovering those hidden cracks that are difficult to detect with visible light alone; acoustic wave reflection data detects the integrity of the internal structure of the building through the propagation and reflection characteristics of ultrasonic waves. Cracks will cause changes in the reflection and delay characteristics of acoustic waves during propagation, which provides more in-depth information for the identification of internal defects. In order to ensure that these data have high accuracy and availability in subsequent analysis, the preprocessing step is a crucial link.

[0053] The primary task of data preprocessing is to denoise and normalize the collected visible light images, thermal imaging data, and acoustic reflection data. In the preprocessing of visible light images, the bilateral filtering algorithm is used to denoise the data. This algorithm can not only effectively remove noise, but also retain the sharp features of the crack edge. Compared with other common filtering methods, bilateral filtering can maintain the clarity of local edges while eliminating image noise, which is particularly suitable for scenes with complex crack morphology and large background noise. The crack edge data generated by this step retains clear boundary information, making subsequent crack detection and morphological recognition more accurate. In the preprocessing of thermal imaging data, the normalization operation converts the collected temperature data into a unified scale through linear mapping to make the temperature difference characteristics of the crack area more obvious. The normalized data can significantly enhance the temperature difference between the crack area and the normal area, making the thermal anomaly characteristics of the crack more prominent in the subsequent temperature analysis. The denoising of acoustic reflection data mainly eliminates environmental noise unrelated to the crack by calculating the time delay difference. Since cracks will cause a specific time delay effect on the propagation of sound waves, by analyzing the time delay differences of the sound wave reflection signals, external interference signals can be effectively filtered out and sound wave characteristic signals can be generated for identifying internal defects in civil engineering structures.

[0054] After data preprocessing, the system generates crack edge data, temperature anomaly area data, and internal defect location data. Crack edge data is based on the processing of visible light images, providing accurate geometric boundary information; temperature anomaly area data provides thermal anomaly information on the building surface through thermal imaging analysis, which can usually highlight the thermal characteristics of the crack area; and internal defect location data is obtained through time delay analysis of acoustic wave reflection signals, which can identify the damage caused by cracks to the internal structure of the building. These three types of preprocessed data provide an important basis for subsequent analysis and risk assessment, ensuring the accuracy and reliability of crack detection. In practical applications, the combination of these data can effectively improve the multi-dimensional coverage and depth perception capabilities of crack detection, especially in complex building environments. Through the collaborative analysis of multimodal data, potential structural problems can be captured more accurately, providing a scientific basis for the safe maintenance of buildings.

[0055] The crack edge data, temperature anomaly area data and internal defect location data are processed by generative adversarial networks to generate visual morphological data, temperature characteristic data and acoustic wave characteristic data of simulated cracks, and crack matching degree data is generated through morphological matching algorithm and reverse reasoning;

[0056] In the present invention, the crack edge data, temperature anomaly area data and internal defect location data are processed by a generative adversarial network (GAN) in order to generate multi-dimensional simulation data of the crack, specifically including visual morphology data, temperature feature data and acoustic wave feature data. The core principle of the generative adversarial network is to achieve the generation of crack simulation data through the alternating optimization of two mutually antagonistic neural networks, the generator and the discriminator. The generator is responsible for generating realistic simulated cracks based on the input data, while the discriminator is used to judge the matching degree between the generated data and the actual data, and feed back to the generator to optimize the generation effect. In this process, the crack edge data, temperature anomaly area data and internal defect location data respectively provide the geometric morphology, thermal anomaly characteristics and acoustic wave reflection characteristics of the crack. The generator uses these multimodal data as input to generate realistic visual morphology, temperature characteristics and acoustic wave characteristics of simulated cracks.

[0057] First, the crack edge data is obtained through visible light image processing, which is the basis for describing the geometric appearance of the crack. The generator generates visual morphological data of simulated cracks based on this data to ensure that the generated crack edge features are consistent with the shape and size of the actual crack. In this step, the generator analyzes and abstracts the edge data through a multi-layer convolutional neural network to generate visual morphological data similar to the real crack edge. When processing the temperature anomaly area data, the generator uses the input thermal imaging data to generate simulated temperature feature data that matches the temperature difference in the actual crack area through the temperature distribution pattern. The focus of this process is to ensure that the simulated data can reflect the thermal anomaly in the crack area, especially in areas where the heat conduction changes caused by the cracks are more obvious. The generated temperature feature data must be highly consistent with the thermal imaging results of the actual building. In the processing of sound wave reflection data, the generator generates simulated sound wave feature data based on the input sound wave reflection features to simulate the impact of cracks on sound wave propagation. The presence of cracks will change the path and delay of sound wave propagation. The generator generates sound wave feature data by simulating these changes. The data reflects the specific impact of the internal structure of the crack on the sound wave propagation.

[0058] like Figure 3 As shown in the figure, after the generative adversarial network generates the above three types of simulated crack feature data, it enters the discriminant stage. The task of the discriminator is to compare the simulated data generated by the generator with the actual data to determine whether the generated data is realistic enough. The discriminator compares the actual crack data with the generated data by input, and further processes them through the morphological matching algorithm to analyze the differences between the two in geometric morphology, temperature distribution and acoustic wave propagation characteristics. The morphological matching algorithm calculates the similarity between the generated data and the actual data in each feature dimension to obtain the morphological matching degree data. These matching degrees reflect the similarity between the simulated cracks and the actual cracks in each dimension, and then provide reverse feedback to the generator, so that the generator can further optimize the generated data.

[0059] In this process, the reverse reasoning mechanism also plays an important role. Reverse reasoning is based on the interaction between the generator and the discriminator. By backpropagating the error information output by the discriminator, the error information is fed back to the generator, prompting the generator to adjust and optimize the generated data in the next round of training. Through continuous iterative optimization, the generator can gradually generate more realistic simulated crack feature data. Ultimately, the simulated crack data generated by the system can highly approximate the morphology, temperature characteristics, and acoustic wave propagation characteristics of real cracks, ensuring a high degree of similarity between the multi-dimensional characteristics of the simulated cracks and the actual cracks.

[0060] The core effect of this process is reflected in the generation of visual, temperature and acoustic wave characteristic data of cracks. These simulated data provide a reliable data basis for subsequent 3D reconstruction and complexity assessment. Through morphological matching algorithms and reverse reasoning, the system can not only ensure the accuracy of simulated data, but also continuously optimize the generation effect through iterative training, so that each round of generated data can more accurately reflect the crack characteristics in actual civil engineering structures. This method can effectively improve the detection accuracy in the crack detection process, especially in complex crack environments. Through the collaborative processing of multimodal data, the generative adversarial network can generate complete crack characteristic data, laying a solid foundation for subsequent analysis and evaluation.

[0061] Preferably, the training process of the generative adversarial network includes:

[0062] The generator generates visual morphological data of simulated cracks, temperature characteristic data of simulated cracks and acoustic wave characteristic data of simulated cracks through crack edge data, temperature abnormality area data and internal defect position data;

[0063] The discriminator is used to determine the matching degree between the generated simulated crack data and the actual crack data. The generator and the discriminator are interactively trained by optimizing the loss function, so that the difference between the generated simulated crack data and the actual crack data is gradually reduced.

[0064] In the present invention, the method realizes the generation and optimization of simulated cracks through the training process of generative adversarial network (GAN). The principle of generative adversarial network is to gradually generate simulated crack data similar to actual crack characteristics through the interaction between two adversarial neural network models, generator and discriminator. The task of the generator is to generate the visual morphology, temperature characteristics and acoustic wave characteristics of the simulated crack based on the input data, while the discriminator is used to judge the similarity between these generated data and the actual crack data. Through the optimization of the loss function, the difference between the generated simulated crack data and the real crack data is gradually reduced.

[0065] During the training process of the generator, the input data includes crack edge data, temperature anomaly area data, and internal defect location data. These data are multimodal information obtained through previous data collection and preprocessing, and each type of data represents different characteristic dimensions of the crack. The crack edge data is obtained through visible light image processing, reflecting the geometric edge information of the crack; the temperature anomaly area data is generated through thermal imaging data, describing the thermal anomaly characteristics at the crack; and the internal defect location data is extracted based on the acoustic wave reflection data, which can reflect the impact of the crack on the acoustic wave propagation characteristics. After receiving these multimodal data, the generator uses a deep neural network model to learn the features of the input data and generates visual morphological data of simulated cracks, temperature feature data of simulated cracks, and acoustic feature data of simulated cracks. This process captures the spatial, temperature, and acoustic wave features in the input data through the operation of multiple convolutional layers and deconvolutional layers, and forms the final simulated crack through the deep learning ability of the generator.

[0066] After the generator generates preliminary simulated crack data, the discriminator is responsible for comparing these simulated data with the actual crack data. The input of the discriminator includes the simulated crack data generated by the generator and the actual crack data collected through real building cracks. The task of the discriminator is to discriminate between the two and determine whether the generated data is close enough to the real data. The working principle of the discriminator is to extract high-dimensional features of the input data through the convolutional neural network layer, classify these high-dimensional features, and output the discrimination result, that is, the matching degree between the generated data and the real data. In order to improve the quality of the simulated cracks generated by the generator, the generator and the discriminator are interactively trained through the loss function. The goal of the generator is to generate simulated crack data that can "deceive" the discriminator, so that the discriminator cannot distinguish between the generated data and the real data, while the goal of the discriminator is to distinguish between the generated data and the real data as much as possible. In this adversarial training process, the capabilities of the generator and the discriminator are continuously improved, and the quality of the generated data is gradually improved.

[0067] The loss function is the core tool for interactive training of the generator and the discriminator. In the generative adversarial network, the loss function of the generator is used to measure the difference between the generated data and the real data, and the loss function of the discriminator is used to measure the accuracy of the discriminator. The generator adjusts the parameters of the generation model through the feedback of the loss function, so that the difference between the generated data and the real data is gradually reduced; similarly, the discriminator also optimizes its judgment ability through the loss function, so that it can more accurately distinguish between the generated data and the real data. In the present invention, the loss function of the generator is optimized based on the multimodal characteristics of the cracks, ensuring that the generated crack data can be close to the multidimensional characteristics of the real cracks in terms of visual, temperature and acoustic wave characteristics.

[0068] The training process of the generative adversarial network has achieved remarkable results. Through continuous iterative training, the generator can generate more and more realistic simulated crack data, which includes the geometric shape of the crack, temperature distribution characteristics, and acoustic wave propagation characteristics, and can provide high-quality input for subsequent 3D reconstruction and complexity assessment. In practical applications, this method is particularly suitable for complex crack detection scenarios. Through the generative adversarial network, a complete crack feature model can be generated. Even in the case of missing data or incomplete partial modal information, realistic crack features can be generated, providing strong technical support for the safety assessment of civil engineering structures and crack maintenance decisions.

[0069] Preferably, the crack matching data is calculated by the following formula:

[0070] M=α·V m +β·T m +γ·S m

[0071] Where M represents the crack matching degree; V m Indicates the visual morphology matching degree; T m Indicates the temperature characteristic matching degree; S m Indicates the matching degree of acoustic wave features; α, β, and γ are the weight coefficients of visual morphology, temperature characteristics, and acoustic wave characteristics, respectively. The weight coefficients are dynamically adjusted according to the confidence of the actual crack characteristics.

[0072] In the process of detecting defects in civil engineering structures of the present invention, the calculation of crack matching data is performed by combining three modal data: visual morphology, temperature characteristics and acoustic wave characteristics. The core principle of this calculation is to quantify the matching results of each modality and perform weighted average according to its importance to the crack detection results, and finally obtain a comprehensive crack matching degree.

[0073] In this formula, the visual morphology matching degree V m It is obtained by comparing the generated simulated cracks with the actual cracks in terms of geometry. Specifically, the system uses the crack edge data to extract the edge features of the cracks through image processing algorithms (such as edge detection, morphology matching, etc.), and compares these features with the crack edge features generated by the generative adversarial network to calculate the similarity between the two. This similarity reflects the degree of closeness between the generated visual morphology data and the actual cracks in terms of geometric appearance. A high degree of match means that the generated simulated cracks are consistent with the actual cracks in terms of shape, position, and edge details, thus providing a reliable basis for subsequent 3D reconstruction.

[0074] Temperature characteristic matching degree T mIt is obtained by analyzing the temperature distribution in the crack area. The data of the temperature anomaly area comes from the thermal imaging data, which reflects the temperature distribution characteristics of the crack area. Since cracks are often accompanied by local heat conduction anomalies, the temperature characteristic data of simulated cracks needs to be able to accurately reproduce this temperature anomaly. By comparing the simulated temperature characteristics with the temperature distribution of the actual crack area, the matching degree of the two in temperature characteristics is calculated. The higher the matching degree, the more accurate the generated simulation data is in reproducing the temperature difference in the crack area, and it can reflect the real physical conditions of the crack area.

[0075] Acoustic wave characteristic matching degree S m It is obtained by analyzing the acoustic reflection characteristics of the internal structure of the crack. The propagation and reflection of the crack will produce unique time delays and reflection intensity changes. By analyzing these changing characteristics in the acoustic reflection data, the system can capture the impact of the internal structure of the crack on the propagation of the acoustic wave. In the process of generating simulated crack data, the system generates simulated acoustic wave characteristic data. After comparing these data with the acoustic reflection characteristics caused by the actual crack, the matching degree of the acoustic wave characteristics can be obtained. This matching degree can reflect the authenticity of the simulated cracks in the deep structure. Especially when detecting hidden cracks inside buildings, the accuracy of the acoustic wave characteristic matching degree is crucial.

[0076] The weight coefficients α, β, and γ in this formula are dynamically adjusted and are used to weight according to the importance of different crack features in detection. For example, in some cases, the visual morphological features may be the main basis for judging cracks, and the weight of α will be relatively high; in other cases, such as when the crack is located inside the structure, the temperature or acoustic wave features may be more significant, and the weight of β or γ will increase accordingly. The adjustment of the weight coefficient is based on the confidence of the actual crack features, which means that during the detection process, the system can adaptively and dynamically adjust the contribution of each mode according to the characteristics of the specific building cracks, so that the final calculated crack matching degree can more truly reflect the comprehensive characteristics of the cracks.

[0077] Through this matching formula, the system can accurately evaluate the similarity between the generated simulated cracks and the actual cracks in various feature dimensions, providing accurate data support for subsequent analysis. The effect of this process is that the crack detection capability under the coordination of multimodal data has been greatly enhanced, ensuring that the geometric morphology, temperature characteristics and internal structural characteristics of the cracks can be fully identified and matched in complex building environments. By dynamically adjusting the weights, the system can also adaptively optimize the matching algorithm according to different detection scenarios, improve the flexibility and accuracy of detection, and thus provide a more reliable basis for the comprehensive assessment of building safety.

[0078] Based on the crack matching data, the adaptive dynamic weight 3D reconstruction technology is used to generate 3D crack model data by combining visual morphology data, temperature feature data and acoustic wave feature data. The 3D crack model data is used to estimate the depth and width of the cracks, and the crack depth data and crack width data are generated.

[0079] In the present invention, based on the crack matching data, the adaptive dynamic weight 3D reconstruction technology is used to generate an accurate 3D crack model by combining multimodal data (visual morphology data, temperature characteristic data and acoustic wave characteristic data), and on this basis, accurately estimate the depth and width of the crack. The core of this technology is to dynamically adjust the importance of different modal data and fuse these data to reconstruct the 3D structure of the crack and achieve a comprehensive analysis of the crack geometric characteristics.

[0080] The crack matching data is the result of multimodal data fusion calculated in the previous step, reflecting the matching degree of the visual morphology, temperature characteristics and acoustic characteristics of the crack in each dimension. Based on these matching degrees, the first thing that 3D reconstruction technology needs to solve is how to effectively fuse data of different modes. Adaptive dynamic weight technology is designed to solve this problem. This technology dynamically adjusts the weight coefficient of each modal data according to the contribution of different data modes to the geometric structure during the crack reconstruction process. For example, visual morphology data contributes more in describing the surface morphology of the crack, acoustic wave characteristic data can better reflect the depth and internal structure of the crack, and temperature characteristic data reflects the thermal conduction anomaly at the crack. The dynamic adjustment of weights is based on the different crack characteristics and is achieved through iterative optimization, so that the generated 3D crack model is more accurate based on multidimensional data.

[0081] During the reconstruction process, visual morphological data is mainly used to generate the surface geometry of the crack. By analyzing the crack edge data and other geometric features, the system first uses 3D point cloud technology to build a preliminary model of the crack. This preliminary model uses image processing technology to extract the key points of the crack at different perspectives (such as through the SIFT feature point matching algorithm), and uses these key points as the basis for 3D point cloud data. These point cloud data are the starting point for building a 3D model. They can describe the position and shape of the crack in space and provide support for subsequent fine modeling.

[0082] The temperature feature data is combined with the visual morphology data to further optimize the 3D crack model. The temperature feature data is obtained through thermal imaging technology and reflects the thermal conduction anomaly in the crack area. These thermal anomaly features can effectively help the system determine the change in crack depth. By integrating the temperature feature with the 3D point cloud data, the system can better capture the change in the depth direction of the crack. Especially in the case of deep cracks, the temperature feature data provides a key reference for crack depth estimation.

[0083] At the same time, the acoustic wave characteristic data is used as auxiliary information to supplement and optimize the width and depth estimation of cracks in the 3D model. The acoustic wave characteristic data is obtained through acoustic wave reflection analysis. Cracks will cause specific reflection and diffraction phenomena during the propagation of acoustic waves. Especially when the cracks penetrate deeper layers of the building, the acoustic wave data can effectively reflect the internal structure of the cracks. By combining these acoustic wave characteristic data with visual and temperature data, the system can more accurately reconstruct the 3D morphology of the cracks and effectively estimate the changes in the width and depth of the cracks.

[0084] A key step in the 3D reconstruction process is the adjustment of adaptive dynamic weights. Through multiple rounds of iterations, the system adjusts the weights according to the contribution of each modal data to the 3D reconstruction accuracy, so that the model can be optimized according to the data characteristics at each stage. For example, when the surface morphology of the crack is relatively simple, the weight of the visual morphological data will be higher, while when dealing with complex cracks (such as cracks that penetrate deep structures), the weight of the acoustic data will increase. Ultimately, the system generates a highly accurate 3D crack model by fusing all modal data, and estimates the depth and width of the crack through this model.

[0085] After generating a three-dimensional crack model, the system uses the model for geometric analysis to generate crack depth data and crack width data. The estimation of crack depth is mainly based on the vertical dimension data in the three-dimensional model. By analyzing the relative height difference between different points in the model, combined with temperature characteristics and acoustic wave characteristics, the system can infer the depth change of the crack. The estimation of crack width is based on the cross-sectional analysis of the crack. By extracting the morphological features on different sections in the three-dimensional model, the system can accurately measure the width change of the crack, especially when the crack width is uneven, the system can identify the widest and narrowest areas of the crack.

[0086] Through this 3D reconstruction technology, the system can generate a complete 3D model of the crack, and generate crack depth data and crack width data based on the model. These data provide important support for subsequent crack complexity assessment and risk assessment. Through the fusion of multimodal data and dynamic adjustment of weights, the system ensures that the geometric and physical characteristics of cracks can be accurately captured in different environments, which significantly improves the accuracy and reliability of crack detection. In civil engineering structural health monitoring, this technology can help inspectors more accurately assess the potential impact of cracks on building safety, and put forward scientific maintenance and treatment suggestions based on the geometric characteristics of cracks.

[0087] Preferably, the adaptive dynamic weighted 3D reconstruction comprises:

[0088] Initialize the weight parameters of each modal data;

[0089] The SIFT feature point matching algorithm is used to extract the key points of the crack at different viewing angles and generate a preliminary three-dimensional point cloud of the crack.

[0090] The depth and width parameters of the crack are dynamically adjusted according to the temperature characteristic data and the acoustic wave characteristic data to generate optimized three-dimensional crack model data.

[0091] like Figure 2 As shown, in the process of civil engineering structure defect detection of the present invention, the reconstruction of the three-dimensional crack model is completed by adaptive dynamic weight technology. The core of this technology is to combine multimodal data (including visual morphology, temperature characteristics and acoustic wave characteristic data) to accurately reconstruct the crack in three dimensions, and dynamically adjust the weight of each mode at different stages to optimize the geometric model of the crack to ensure that the model can fully reflect the depth, width and overall morphology of the crack. The process includes the following key steps:

[0092] 1. Initialize the weight parameters of each modal data:

[0093] First, before 3D reconstruction, the weight parameters of each modal data need to be initialized. These weight parameters are used to control the relative influence of visual morphological data, temperature characteristic data, and acoustic wave characteristic data in the 3D reconstruction process. The initial value of the weight can be set according to different types of crack characteristics. Usually, the initial weight of visual morphological data is larger because it has a more intuitive effect in describing the geometric characteristics of the crack surface. However, for deeper cracks, the weights of temperature characteristic data and acoustic wave characteristic data will gradually increase. This weight initialization provides a basis for subsequent adaptive adjustment.

[0094] 2. Generate a preliminary 3D point cloud of the crack using the SIFT feature point matching algorithm:

[0095] After the weight parameters are initialized, the system uses the SIFT (Scale Invariant Feature Transform) feature point matching algorithm to extract the key points of the crack at different perspectives and generate a preliminary 3D point cloud model. The role of the SIFT algorithm is to extract stable and consistent key points from multi-angle images of the crack, which describe the edges and geometric features of the crack. Since the SIFT algorithm has strong scale and rotation invariance, it can accurately match the key points of the crack under different perspectives and lighting conditions. By matching these key points, the system generates a preliminary 3D point cloud model, which provides a basic description of the position and morphology of the crack in 3D space. The density and distribution of the point cloud directly affect the subsequent 3D reconstruction accuracy, so this process is crucial.

[0096] Specifically, the SIFT algorithm extracts feature points from crack images taken from different perspectives, and matches the corresponding feature pairs through the algorithm to form a sparse point cloud. These point cloud data are the basis for building a 3D model of the crack, and they represent the real geometric position of the crack in space. Through further processing, the system will expand these sparse point clouds into denser 3D data to generate a more refined 3D model.

[0097] 3. Dynamically adjust the parameters of crack depth and width:

[0098] After generating the preliminary 3D point cloud, the system further dynamically adjusts the depth and width parameters of the crack based on the temperature characteristic data and the acoustic wave characteristic data. The temperature characteristic data comes from the thermal imaging results and can reflect the temperature anomalies in the crack area. These temperature anomalies are usually closely related to the depth of the crack, because the depth of the crack affects the heat conduction characteristics. By combining the temperature characteristics with the preliminary generated point cloud data, the system can accurately infer the depth changes of the crack. As the weight of the temperature characteristic data increases, the system will dynamically adjust the parameters of the point cloud model in the crack depth direction, thereby optimizing the 3D structure of the crack.

[0099] At the same time, the acoustic wave characteristic data reflects the internal structural characteristics of the crack by analyzing the change in acoustic wave reflection and time delay, which is particularly important for adjusting the crack width. Cracks will cause changes in the propagation path of the acoustic wave, and the difference in the reflection intensity and propagation time of the acoustic wave can provide accurate information on the crack width. The system combines these acoustic wave characteristic data with the three-dimensional point cloud and optimizes the crack width estimation in the model by adaptively adjusting the crack width parameters. Especially in the case of uneven crack width, the acoustic wave characteristic data can provide more detailed width distribution information, allowing the reconstructed model to capture small changes in crack width.

[0100] Through the above steps, the system can generate a highly accurate 3D crack model, which can not only reflect the geometric shape of the cracks on the building surface, but also accurately estimate the depth and width of the cracks by combining temperature and acoustic wave data. The advantage of this adaptive dynamic weight 3D reconstruction technology is that it can adaptively adjust the weight of the data according to the different characteristics of the cracks and the detection environment, so that high-quality 3D models can be obtained under different conditions. Especially for those complex crack structures, through dynamic weight adjustment, the system can effectively integrate multimodal data to ensure the comprehensiveness and accuracy of the model.

[0101] For example, in the detection process of a deep crack, the visual morphological data may not be sufficient to fully reflect the depth characteristics of the crack. At this time, the system will automatically increase the weight of the temperature characteristic data and the acoustic wave characteristic data to more accurately capture the depth changes of the crack. Similarly, in the case of cracks with uneven widths, the increase in the weight of the acoustic wave data can help optimize the width estimation in the model, making the final crack model closer to the actual situation.

[0102] Preferably, the crack depth estimation is performed by a convolutional neural network, wherein the optimized three-dimensional crack model data is input, segmented according to different depth layers, and the depth changes of the cracks are calculated layer by layer to generate crack depth data.

[0103] In the crack detection of the present invention, the estimation of the crack depth is performed by a convolutional neural network (CNN), which can accurately estimate the change of the crack in the depth direction by analyzing the input three-dimensional crack model data layer by layer, thereby generating crack depth data. The core of this process is to use the powerful feature extraction and spatial analysis capabilities of CNN to perform deep reasoning on the optimized three-dimensional crack model data, and accurately extract the crack depth change information from the three-dimensional data.

[0104] The 3D crack model data is generated by the previous adaptive dynamic weight 3D reconstruction technology. These data include the geometry, temperature characteristics and acoustic characteristics of the crack in space. These multimodal data have been integrated into a complete 3D crack model through optimization steps. In order to estimate the depth of the crack, the system provides the 3D model as input data to the convolutional neural network. The convolutional neural network has the ability to extract spatial features at multiple levels. It gradually abstracts and captures the characteristic changes of the crack at different depths through the convolution layer.

[0105] Specifically, the convolutional neural network processes the 3D crack model data in layers, dividing the 3D space into several depth layers, each of which represents the geometric features of the crack at that layer. The data of each layer is calculated by the convolution kernel to extract the spatial features of the layer, especially the geometric changes of the crack, such as the shape of the crack, the change trend of the crack edge, etc. In each convolution layer, CNN further processes these features to capture the subtle changes of the crack at different depths.

[0106] For each depth layer, the convolutional neural network not only extracts the surface features of the cracks, but also analyzes the internal structure of the cracks. In actual operation, the system will segment the three-dimensional crack model in the depth direction. For example, a crack in a building may appear as a shallow opening on the surface, but as the depth increases, the shape of the crack may change significantly, and this change is difficult to observe on the surface. Therefore, CNN can accurately calculate the changes in geometric features such as width, angle, and branching of cracks at different depths through layer-by-layer analysis, and retain these changes as depth feature data.

[0107] In this process, the convolution layer extracts the features of each depth layer by performing kernel operations on the three-dimensional data, and the pooling layer reduces redundant information and retains the most representative depth features. In this way, the system can generate a feature map containing the changes in crack depth through multiple layers of convolution and pooling operations. These feature maps describe the geometric changes of cracks at different depth layers, such as the tendency of cracks to widen or narrow at a certain layer, and the branching of cracks. These geometric changes reflect the true depth of the cracks in terms of spatial structure, and these features are ultimately integrated into complete depth data.

[0108] Through multi-level processing of convolutional neural networks, the system can generate accurate crack depth data. These depth data can reflect the true depth of cracks inside the building, including not only the cracks visible on the surface, but also the cracks hidden in the civil engineering structure. For example, some cracks may seem insignificant on the surface of the building, but may have expanded to a dangerous depth deep in the structure. The convolutional neural network can reveal these potential hidden dangers through in-depth analysis of the three-dimensional crack model.

[0109] This crack depth estimation method based on convolutional neural networks also has high accuracy and automation. Through automated hierarchical processing and deep reasoning, the system can complete the depth detection of complex cracks at a high speed, which is particularly suitable for the monitoring and maintenance of large-scale civil engineering structures. Since crack depth is an important parameter for building safety assessment, generating accurate depth data can not only help inspectors determine the severity of cracks, but also provide strong support for subsequent maintenance decisions.

[0110] For example, in the monitoring process of a high-rise building, the system can detect the depth of some surface cracks extending into the load-bearing structure of the building through this method, and this depth information is difficult to capture by traditional detection methods. Through layer-by-layer analysis of the convolutional neural network, the system can reveal the actual depth of the cracks and provide a clear reference for building maintenance.

[0111] Preferably, the crack width is estimated by analyzing the cross-sectional morphology of the three-dimensional crack model to generate crack width data, thereby ensuring that the widths on different cross sections are accurately measured.

[0112] In the present invention, the estimation of crack width is carried out by analyzing the cross-sectional morphology of the three-dimensional crack model. Specifically, the system extracts the width information of the crack at various positions from different cross-sections through the constructed three-dimensional crack model to generate accurate crack width data. This method can not only capture the changes in crack width at different depths and positions, but also ensure that the width measurement of the entire crack is comprehensive and consistent.

[0113] The 3D model of the crack is generated through the previous data collection, processing and reconstruction, which includes the depth, width and morphological information of the crack. In order to accurately estimate the crack width, the system will perform layered processing on the generated 3D model and extract multiple cross sections along the longitudinal direction of the crack. Each cross section represents the morphology and geometric characteristics of the crack at that specific location. The system analyzes the morphological data of each cross section, especially the distance between the crack edges, to derive the crack width on that cross section.

[0114] During the cross-sectional analysis, the crack edges are first accurately extracted using image processing algorithms. Edge detection algorithms are usually used, such as the Canny edge detection algorithm or the Sobel operator, which can accurately identify the edge contours of the crack. The precise location of these edge points is the basis for crack width estimation. On each cross section, the system determines the crack width by calculating the shortest distance between the edges on both sides of the crack. This width is based on high-precision data from the 3D model and can therefore accurately reflect the true physical properties of the crack.

[0115] It is worth mentioning that the crack width estimation is not limited to a certain cross section, but multiple measurements are taken along the entire length of the crack. The system extracts multiple cross sections at different depth layers and analyzes these sections one by one to ensure that the changes in crack width can be fully reflected. For example, in some complex crack situations, the width of the crack may vary significantly at different depths, and even local expansion or contraction may occur. Through multi-level cross-sectional analysis, the system can capture these subtle changes and generate accurate crack width data.

[0116] In practical applications, the estimation of crack width not only requires high-precision three-dimensional model support, but also relies on accurate edge detection and distance calculation methods. For each extracted cross section, the system first identifies the edge of the crack through image processing technology, and measures the width of the crack through mathematical geometry calculations. In some special cases, such as when the edge of the crack is more complex or has irregular shapes, the system will use more geometric algorithms to ensure the accuracy of the crack width. Common methods include calculations based on the minimum bounding box, and edge smoothing and optimization based on morphological operations. The application of these methods can effectively improve the accuracy of width measurement, especially when the crack edge is irregular or the noise is large.

[0117] In order to ensure the comprehensiveness of the crack width estimation, the system automatically extracts cross sections along the longitudinal direction of the crack multiple times, measures the width of each cross section, and generates multiple width data points. These data points not only reflect the width of the crack at different depths, but also reveal the morphological changes of the crack in the entire structure. For example, for a crack of uneven width, the system can identify the expansion of the crack at different levels, thereby helping the inspection personnel understand the specific morphological distribution of the crack.

[0118] Through this crack width estimation method based on cross-sectional morphological analysis, the system is able to generate accurate crack width data. These data can be widely used in crack assessment and maintenance of buildings, especially when assessing the severity of cracks, crack width is often a key reference indicator. The system can automatically complete the width measurement, significantly improving the detection efficiency, while ensuring that the width of different crack locations can be accurately measured.

[0119] The advantages of this method are also reflected in its high flexibility and accuracy. By analyzing the cross-section layer by layer, the system can capture the width changes of the crack at any position. Even if the crack shape is more complex or the width is irregular, it can accurately present the overall picture of the crack through multi-level analysis. In addition, the accurate estimation of crack width also provides an important basis for subsequent civil engineering structural safety assessment and crack repair decisions. For example, on the exterior wall of a building, if the system detects that the crack width has significantly expanded at certain depth levels, it may indicate that the crack at that location poses a greater threat to the structural stability of the building, and relevant departments can carry out targeted maintenance and repairs based on these data.

[0120] Based on the crack depth data, crack width data, three-dimensional crack model data, temperature characteristic data and acoustic wave characteristic data, the crack complexity index is calculated, the risk level of the crack is evaluated, the crack risk level data is generated, and the crack areas for priority treatment are identified.

[0121] In the present invention, the complexity index of the crack is calculated by comprehensively analyzing the crack depth data, crack width data, three-dimensional crack model data, temperature characteristic data and acoustic wave characteristic data, and finally generates the risk level data of the crack, and identifies the crack areas that need to be treated first. The core of this process lies in the fusion and quantitative evaluation of multimodal data. By analyzing the geometric morphology, physical characteristics and structural influence of the crack in multiple dimensions, an index model that can reflect the danger and complexity of the crack is constructed, thereby realizing the intelligent evaluation of the crack risk.

[0122] The complexity of cracks depends not only on the geometric morphology of the cracks (such as depth and width), but also on the internal structure and thermal anomalies of the cracks. In order to comprehensively evaluate the risk level of cracks, the system needs to analyze these different dimensions of data simultaneously. First, the crack depth data and crack width data can intuitively reflect the geometric size of the cracks. Generally, the deeper and wider the cracks, the greater their potential threat to civil engineering structures. Therefore, depth and width are two key parameters for evaluating crack complexity and risk level.

[0123] Secondly, the 3D crack model data can provide the spatial morphological characteristics of the cracks. By analyzing the shape, distribution, and presence of multiple branches of the cracks in 3D space, the system can further determine the complexity of the cracks. For example, some cracks may appear simple on the surface, but may have a complex multi-level structure in 3D space, especially when the cracks have obvious branches in the depth direction. This complexity greatly increases the destructiveness of the cracks to civil engineering structures.

[0124] At the same time, the temperature characteristic data provides information on thermal conductivity anomalies in the crack area. Cracks usually affect the thermal conductivity characteristics of buildings, especially in the crack area, where abnormal temperature changes may indicate that the cracks have affected the internal structure of the building materials. For example, temperature anomalies deep in the cracks often mean that the cracks have penetrated multiple layers of materials and may even affect the core structure of the building. The degree of abnormality of temperature characteristics can be further used to assess the potential risk of cracks.

[0125] The acoustic wave characteristic data determines the internal structural integrity of the crack by analyzing the impact of the crack on the propagation of the acoustic wave. The time delay and reflection changes of the acoustic wave when passing through the crack can reflect the degree of damage to the internal structure of the building caused by the crack. If there is a significant change in the propagation of the acoustic wave, it indicates that the crack has caused obvious damage to the internal structure of the building, which will significantly increase the complexity index of the crack. Especially in certain parts of the crack, the acoustic wave characteristics may indicate internal damage that cannot be observed by the naked eye. This potential structural damage is an important basis for assessing the risk level of the crack.

[0126] In practical applications, the system will calculate the complexity index of the cracks based on these multimodal data through a specific algorithm. Usually, this calculation process adopts a weighted summation method, and the weights of different data are determined by the importance of the cracks in different dimensions. For example, in some cracks, depth and width are important factors affecting their complexity, so the system will give these two dimensions a higher weight; in other cases, temperature anomalies or acoustic reflection characteristics may play a key role, and the weight of these data will increase accordingly.

[0127] In the calculation model of the complexity index, data of all dimensions will be standardized to ensure that data from different sources can be measured uniformly. For example, depth data and width data can be standardized through simple geometric proportions, while temperature characteristics and acoustic wave characteristics data can be converted into easily comparable numerical ranges through normalization methods. Ultimately, the system will generate a crack complexity index. The higher the index, the greater the structural complexity of the crack and the greater the safety threat to the building.

[0128] Based on the complexity index, the system can automatically assess the risk level of cracks. The risk level is usually divided according to the range of the complexity index. For example, the complexity index can be divided into three risk levels: low, medium, and high. The higher the risk level of the crack, the greater the need for priority treatment. The system will generate crack risk level data based on these assessment results and mark the priority crack areas on the civil engineering structure diagram. This process greatly simplifies the complexity of manual assessment, allowing inspectors to quickly identify and prioritize the repair of the most dangerous cracks.

[0129] This crack complexity assessment method based on multimodal data fusion can provide a scientific basis for comprehensive crack analysis. The system generates a complexity index by quantitatively analyzing the depth, width, three-dimensional morphology, temperature anomalies, and acoustic reflection of cracks, which significantly improves the accuracy and efficiency of crack risk assessment. Especially in complex civil engineering structures, this method can effectively identify potential high-risk cracks and give priority treatment recommendations to help building managers make reasonable maintenance decisions.

[0130] For example, in the daily monitoring of high-rise buildings, the system can generate a complete crack risk report by quantitatively analyzing the complexity of each crack. For those cracks with a higher complexity index, the system will give priority to repairing them and mark the specific locations of these cracks on the civil engineering structure diagram. This efficient and accurate assessment method can not only save detection time, but also prevent missed detection or misjudgment, providing technical support for the long-term health management of buildings.

[0131] Preferably, Figure 4As shown, the fracture complexity index is calculated by the following formula:

[0132]

[0133] Among them, C represents the crack complexity index; L represents the crack length; D represents the crack depth; B represents the number of crack branches; S represents the morphological complexity of the crack; T represents the temperature anomaly index; V represents the acoustic wave characteristic index, and w1, w2, w3, w4, and w5 are the weight coefficients of crack length, crack depth, crack branching, temperature anomaly, and acoustic wave characteristics, respectively.

[0134] 1. Crack length (L): Crack length is a key parameter in the crack geometry, which directly affects the potential impact of the crack on the structure. Generally speaking, the longer the crack, the greater the impact on the building surface or internal structure, which may cause more areas to be damaged. Therefore, in the crack complexity index, the crack length is given a weight coefficient w1 to measure the impact of the crack's spatial expansion on the building's safety.

[0135] 2. Crack depth (D): Crack depth is an important factor affecting crack complexity and risk. Deeper cracks usually have more serious impacts on civil engineering structures, especially when the cracks penetrate deep into the load-bearing structure of the building, which may lead to a decrease in structural strength. By analyzing the depth data generated by the three-dimensional crack model through a convolutional neural network (CNN), the system can accurately determine the depth of the crack. This depth information is given a higher weight w2 in the complexity index, especially when the depth is uneven or the crack runs through multiple layers of material, the importance of the depth parameter is particularly prominent.

[0136] 3. Number of crack branches and morphological complexity The morphological complexity and branch number of the crack directly affect its complexity. The number of crack branches B indicates whether the crack has multiple branches or fissures in the depth direction, while the morphological complexity S reflects the geometric complexity of the crack. Generally speaking, the more crack branches there are and the more complex the morphology, the more destructive the structure of the crack is. In order to quantify this dimension, the system analyzes the branching and geometric complexity of the crack through the three-dimensional model data. The weight w3 is used to control the influence of this parameter in the complexity index. The evaluation of morphological complexity can be measured by the fractal dimension of the three-dimensional crack model, and the number of branches is determined by the number of crack bifurcation points identified in the three-dimensional model.

[0137] 4. Temperature anomaly index (T): Temperature anomaly is an important indicator of physical changes inside the crack. Through thermal imaging data, the system can detect temperature anomalies in the crack area. Changes in thermal conductivity at the crack are often related to the fracture of structural materials, air flow penetration, etc. The more abnormal the temperature, the greater the impact of the crack on the internal structure, especially when the crack is deep or spans multiple layers of material. The temperature anomaly index T indicates the intensity of temperature fluctuations in the crack area, and the weight coefficient w4 is dynamically adjusted according to the importance of temperature anomalies to crack risk assessment.

[0138] 5. Acoustic wave characteristic index (V): The acoustic wave characteristic index V reflects the degree of damage to the internal structure of the crack by analyzing the impact of the crack on the propagation of sound waves. Cracks usually cause changes in sound wave reflection, diffraction or delay, especially when the internal structure of the building is severely damaged, the abnormality of the acoustic wave characteristics is more obvious. Through the acoustic wave reflection data, the system can capture the impact of cracks on the propagation of sound waves and convert it into the acoustic wave characteristic index V. This index is quantified by the weight w5 in the complexity calculation. Especially when the cracks are difficult to detect by visual or temperature means, the acoustic wave characteristics become an important basis for risk assessment.

[0139] In practical applications, the system will first standardize the data of all dimensions so that they are within the same numerical range for easy calculation. When calculating the crack complexity index, the system dynamically adjusts the weight coefficient of each dimension according to the importance of different crack characteristics. Generally, for surface cracks, length and depth may be the main influencing factors, so w1 and w2 will be larger; when detecting complex multi-branch cracks, the number of branches and morphological complexity are more important. The weight of the crack will be increased accordingly. The system dynamically adjusts the weight coefficient so that the crack complexity index can flexibly adapt to the characteristics of different cracks.

[0140] By calculating the crack complexity index, the system can quantitatively evaluate the structural complexity of different cracks. The higher the complexity index, the more complex the geometry of the crack and the greater the potential for destructiveness. Based on the complexity index, the system can generate risk level data for cracks, classify cracks into three levels: low, medium, and high, and identify crack areas that need to be treated first on the civil engineering structure diagram. This not only provides a scientific basis for crack repair, but also greatly improves detection efficiency, helping inspectors quickly identify high-risk cracks and prioritize repairs.

[0141] For example, in the inspection of a large bridge, the system calculated the complexity index of the crack and found that a deep crack that penetrated multiple load-bearing columns had a high complexity index. The system marked the crack as a high-risk area and recommended immediate repair. This method greatly improved the accuracy and reliability of the inspection and effectively ensured the structural safety of the bridge.

[0142] Preferably, the fracture risk level converts the fracture complexity index into a risk level through nonlinear mapping, and the risk level is divided into low risk, medium risk and high risk, and the priority treatment area is identified to generate priority treatment area data.

[0143] In the present invention, the crack risk level converts the crack complexity index into a risk level by nonlinear mapping. The core of this method is to use a nonlinear mapping function to convert the crack characteristics represented by the complexity index into an easy-to-interpret risk level, and further identify the priority areas. This process not only realizes the automated assessment of crack risk, but also provides an effective decision-making basis for the priority sorting of building crack repair.

[0144] The crack complexity index C is a comprehensive indicator calculated based on the geometric shape (length, depth, branching, etc.) and physical properties (temperature characteristics, acoustic wave reflection, etc.) of the crack, reflecting the structural complexity and potential destructiveness of the crack. In order to facilitate decision makers to understand and respond, the system needs to convert this complex value into the risk level of the crack through nonlinear mapping. The principle of nonlinear mapping is to emphasize the priority of certain cracks with higher complexity during assessment through a curve function. For example, the complexity index may grow exponentially or logarithmically in certain intervals, that is, the risk does not increase linearly with the increase in the complexity of the crack, but changes at a faster or slower rate.

[0145] The selection and parameter adjustment of the nonlinear mapping function are dynamically set according to the different characteristics of civil engineering structures and the actual situation of the cracks. For example, for key structures such as high-rise buildings or bridges, the depth and width of the cracks may have a greater impact on the safety of the overall structure, so these parameters may be amplified in the nonlinear mapping, making cracks with higher complexity index more likely to be classified as high risk. For surface cracks or decorative cracks, the system may adjust the mapping parameters to correspond to a lower risk level.

[0146] The system will first classify cracks into three levels: low risk, medium risk, and high risk, based on the numerical range of the crack complexity index C. In order to ensure the rationality of the classification, the system uses a nonlinear mapping function to process the complexity index. This nonlinear mapping can be accomplished through an exponential function, a logarithmic function, or other suitable curves. For example, the exponential function can be used to emphasize the changes in the complexity index in the high value range, that is, when the complexity index of the crack reaches a certain threshold, the risk level of the crack will rise rapidly. This nonlinear processing can capture the significant impact of cracks with higher complexity on building safety and identify potential high-risk cracks in a timely manner.

[0147] In actual operation, the system will calculate the risk level of each crack based on the complexity index:

[0148] Low risk: usually corresponds to a small range of crack complexity index, indicating that the length, depth and other morphological characteristics of the cracks will not have a significant impact on the overall safety of the building. For example, surface cracks or shallow cracks may be classified as this.

[0149] Medium risk: The complexity index is relatively medium, indicating that the cracks may have potential impacts on some structural components of the building, especially when there are many branches and great depth. The system will classify it as medium risk.

[0150] High risk: When the complexity index exceeds a certain critical threshold, the crack will be marked as high risk, which usually means that the crack has significant depth, width, and may cause damage to the building's load-bearing structure. Cracks are also more likely to be marked as high risk when the temperature characteristics are abnormally significant or the acoustic characteristics indicate internal structural damage.

[0151] After generating the risk level, the system further identifies the crack areas that need to be treated first. To achieve this goal, the system marks the specific locations of high-risk cracks in the civil engineering structure diagram and generates priority area data. This data can be used to assist decision-making, allowing maintenance personnel to focus resources on repairing high-risk cracks first and ensure the safety of the building. For example, for the load-bearing columns of bridges or the core structures of high-rise buildings, the system will particularly emphasize high-risk cracks and prompt managers to make timely repairs.

[0152] By converting the complexity index into risk level through nonlinear mapping, the system can effectively simplify the process of crack risk assessment and ensure that the priority treatment of cracks can be reasonably sorted according to their actual threat level. The advantages of this method are:

[0153] Through nonlinear mapping, the system can highlight cracks with a higher complexity index and classify them as high-risk cracks, thus avoiding omissions or misjudgments in risk assessment.

[0154] The flexibility of nonlinear mapping enables the system to dynamically adjust the risk classification according to the specific structural characteristics of the building or the type of cracks. For example, for certain important structural components, the system can ensure that cracks in these parts receive priority attention by adjusting the parameters of the mapping function.

[0155] By generating priority area data, the system can provide powerful decision support for building management and maintenance. After high-risk cracks are identified, managers can quickly find areas that need key maintenance through the regional data provided by the system, optimize the allocation of repair resources, and maximize building safety.

[0156] For example, in a crack detection project for a large dam, the system identified certain high-risk cracks in the dam through nonlinear mapping. These cracks were classified as high risk because of their large depth and obvious temperature anomalies. Based on this information, engineers can prioritize these cracks and avoid potential major structural damage.

[0157] like Figure 5 As shown, a system for implementing the method for detecting defects in civil engineering structures based on image processing technology, the system comprising:

[0158] Image acquisition module, used to collect visible light image data, thermal imaging data and acoustic reflection data of civil engineering structures; this module is used to collect multimodal data from civil engineering structures, including visible light images, thermal imaging and acoustic reflection data. These data provide basic information for subsequent crack detection. Visible light images are used to obtain surface information of cracks, thermal imaging data reveals temperature anomalies in the crack area, and acoustic reflection data is used to determine defects in the internal structure of the building.

[0159] The data preprocessing module communicates with the image acquisition module and is used to denoise and normalize the visible light image data, thermal imaging data and acoustic reflection data to generate crack edge data, temperature anomaly area data and internal defect location data; the data preprocessing module receives the data from the image acquisition module and first denoises and normalizes it. Denoising ensures the purity of the data and avoids interference from external noise, while normalization standardizes the data so that it can be processed uniformly in subsequent analysis. The preprocessed data includes crack edge data, temperature anomaly area data and internal defect location data.

[0160] The Generative Adversarial Network module is used to generate the visual morphology data, temperature feature data and acoustic feature data of simulated cracks based on the crack edge data, temperature anomaly area data and internal defect location data, and generate crack matching data through morphological matching algorithm and reverse reasoning; the GAN module uses the data generated by preprocessing to generate the visual morphology, temperature feature and acoustic feature of simulated cracks through the generator, and performs morphological matching and reasoning of cracks through the discriminator to generate crack matching data. This step enhances the accuracy of crack detection through deep learning technology, especially in the integration of complex morphology and multimodal data.

[0161] The three-dimensional reconstruction module is used to generate three-dimensional crack model data based on the crack matching data, combined with the visual morphological data of the simulated crack, the temperature characteristic data of the simulated crack and the acoustic wave characteristic data of the simulated crack, and estimate the depth and width of the crack through the three-dimensional crack model data to generate crack depth data and crack width data; based on the crack matching data generated by the generative adversarial network, combined with the characteristics of each modality (visual, temperature, acoustic wave), the adaptive dynamic weight three-dimensional reconstruction technology is used to generate a three-dimensional model of the crack. Through this model, the system can accurately estimate the depth and width of the crack and generate corresponding crack depth data and width data.

[0162] The complexity index calculation module is used to calculate the crack complexity index based on the crack depth data, crack width data, three-dimensional crack model data, temperature characteristic data and acoustic wave characteristic data; the module calculates the complexity index of the crack based on the geometric and physical characteristics of the crack (depth, width, temperature and acoustic wave characteristics). The complexity index is a comprehensive index obtained by weighted calculation, which reflects the potential threat of the crack to the building.

[0163] The risk assessment module is used to assess the risk level of cracks according to the crack complexity index, generate crack risk level data, and identify crack areas for priority treatment. The risk assessment module converts the crack complexity index into the risk level of cracks through nonlinear mapping, and generates priority area data. This step ensures that high-risk cracks are identified and treated first to prevent greater damage to the civil engineering structure.

[0164] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

1. A method for detecting defects in civil engineering structures based on image processing technology, characterized in that: The following steps are involved: Collect visible light image data, thermal imaging data and acoustic wave reflection data of civil engineering structures, and perform denoising and normalization preprocessing on the collected data to generate crack edge data, temperature anomaly area data and internal defect location data; The crack edge data, temperature anomaly area data and internal defect location data are processed by generative adversarial networks to generate visual morphological data, temperature characteristic data and acoustic wave characteristic data of simulated cracks, and crack matching degree data is generated through morphological matching algorithm and reverse reasoning; Based on the crack matching data, the adaptive dynamic weight 3D reconstruction technology is used to generate 3D crack model data by combining visual morphology data, temperature feature data and acoustic wave feature data. The 3D crack model data is used to estimate the depth and width of the cracks, and the crack depth data and crack width data are generated. Based on the crack depth data, crack width data, three-dimensional crack model data, temperature characteristic data and acoustic wave characteristic data, the crack complexity index is calculated, the risk level of the crack is evaluated, the crack risk level data is generated, and the crack areas for priority treatment are identified.

2. The method for detecting defects in civil engineering structures based on image processing technology according to claim 1, characterized in that: The pre-processing of data includes: De-noising the visible light image data, using a bilateral filtering algorithm to perform image de-noising to retain the sharp features of the crack edge; Performing temperature normalization operation on the thermal imaging data, and adjusting the distribution of temperature values ​​through linear mapping; The acoustic wave reflection data is subjected to denoising, and the interference of environmental noise is eliminated by using time delay difference calculation to generate acoustic wave characteristic signals for identifying internal defects of civil engineering structures.

3. The method for detecting defects in civil engineering structures based on image processing technology according to claim 2, characterized in that: The training process of the generative adversarial network includes: The generator generates visual morphological data of simulated cracks, temperature characteristic data of simulated cracks and acoustic wave characteristic data of simulated cracks through crack edge data, temperature abnormality area data and internal defect position data; The discriminator is used to determine the matching degree between the generated simulated crack data and the actual crack data. The generator and the discriminator are interactively trained by optimizing the loss function, so that the difference between the generated simulated crack data and the actual crack data is gradually reduced.

4. The method for detecting defects in civil engineering structures based on image processing technology according to claim 1, characterized in that: The crack matching data is calculated by the following formula: M=α·V m +β·T m +γ·S m Where M represents the crack matching degree; V m Indicates the visual morphology matching degree; T m Indicates the temperature characteristic matching degree; S m Indicates the matching degree of acoustic wave features; α, β, and γ are the weight coefficients of visual morphology, temperature characteristics, and acoustic wave characteristics, respectively. The weight coefficients are dynamically adjusted according to the confidence of the actual crack characteristics.

5. The method for detecting defects in civil engineering structures based on image processing technology according to claim 1, characterized in that: The adaptive dynamic weight 3D reconstruction comprises: Initialize the weight parameters of each modal data; The SIFT feature point matching algorithm is used to extract the key points of the crack at different viewing angles and generate a preliminary three-dimensional point cloud of the crack. The depth and width parameters of the crack are dynamically adjusted according to the temperature characteristic data and the acoustic wave characteristic data to generate optimized three-dimensional crack model data.

6. The method for detecting defects in civil engineering structures based on image processing technology according to claim 5, characterized in that: The crack depth estimation is performed through a convolutional neural network, in which the optimized three-dimensional crack model data is input, segmented according to different depth layers, and the depth changes of the cracks are calculated layer by layer to generate crack depth data.

7. The method for detecting defects in civil engineering structures based on image processing technology according to claim 6, characterized in that: The crack width is estimated by analyzing the cross-sectional morphology of the three-dimensional crack model to generate crack width data, ensuring that the width on different cross sections is accurately measured.

8. The method for detecting defects in civil engineering structures based on image processing technology according to claim 1, characterized in that: The fracture complexity index is calculated by the following formula: Among them, C represents the crack complexity index; L represents the crack length; D represents the crack depth; B represents the number of crack branches; S represents the morphological complexity of the crack; T represents the temperature anomaly index; V represents the acoustic wave characteristic index, and w1, w2, w3, w4, and w5 are the weight coefficients of crack length, crack depth, crack branching, temperature anomaly, and acoustic wave characteristics, respectively.

9. The method for detecting defects in civil engineering structures based on image processing technology according to claim 8, characterized in that: The fracture risk level converts the fracture complexity index into a risk level through nonlinear mapping. The risk level is divided into low risk, medium risk and high risk. The priority treatment areas are identified and the priority treatment area data is generated.

10. A system for implementing the civil engineering structure defect detection method based on image processing technology according to any one of claims 1 to 9, characterized in that: The system comprises: An image acquisition module for acquiring visible light image data, thermal imaging data, and acoustic wave reflection data of civil engineering structures; A data preprocessing module, communicating with the image acquisition module, is used to perform denoising and normalization processing on the visible light image data, thermal imaging data and acoustic wave reflection data to generate crack edge data, temperature abnormality area data and internal defect location data; A generative adversarial network module is used to generate visual morphological data of simulated cracks, temperature characteristic data of simulated cracks and acoustic wave characteristic data of simulated cracks according to the crack edge data, temperature abnormality area data and internal defect position data, and to generate crack matching degree data through morphological matching algorithm and reverse reasoning; A three-dimensional reconstruction module is used to generate three-dimensional crack model data based on the crack matching data, in combination with the visual morphology data of the simulated crack, the temperature characteristic data of the simulated crack and the acoustic wave characteristic data of the simulated crack by using an adaptive dynamic weight three-dimensional reconstruction technology, and to estimate the depth and width of the crack by using the three-dimensional crack model data to generate crack depth data and crack width data; A complexity index calculation module, used to calculate the crack complexity index according to the crack depth data, crack width data, three-dimensional crack model data, temperature characteristic data and acoustic wave characteristic data; The risk assessment module is used to assess the risk level of the cracks according to the crack complexity index, generate crack risk level data, and identify crack areas for priority treatment.

Images