Multimode-based airport concrete road hidden disease identification and device

Abstract

The application relates to an airport concrete road hidden disease identification method and device based on multiple modes. The method obtains data in two different modes of radar echo and laser point cloud, and extracts respective features. Then, the concept of disease feature template is introduced, and a feature fusion weight is dynamically adjusted based on a correlation value. The construction of the disease feature template enables the disease identification process to focus on the essential properties of the disease itself, effectively reduces the interference of environmental factors on feature extraction and identification accuracy, and dynamically adjusts the fusion weight value according to the first correlation value and the second correlation value. According to the matching degree of the current data and the disease template, the contribution of different modal features can be intelligently allocated. Compared with the simple linear superposition or fixed weight fusion mode, deeper feature cooperation is realized, the fused features are more representative and robust, and finally the precision and stability of the hidden disease identification are improved.

Description

Technical Field

[0001] This application relates to the field of road defect identification technology, and in particular to a multimodal-based device for identifying latent defects in airport concrete roads. Background Technology

[0002] Airport pavements, as a core infrastructure of air transport, are directly related to flight safety due to their structural integrity. Hidden defects (such as underslab delamination, base layer voids, and internal hidden cracks) are key risk points that can lead to sudden pavement damage and disrupt normal airport operations because they are not obviously damaged on the surface and their hidden dangers are concealed. Currently, there are still many technical bottlenecks in the field of airport pavement hidden defect detection, making it difficult to meet the requirements of high-precision and high-adaptability detection. The shortcomings of existing technologies are mainly reflected in the following aspects: Current pavement pavement defect detection technologies largely rely on single-modal data for defect prediction. Common detection methods fall into two categories: one is underground defect detection based on ground-penetrating radar data, which can only capture internal physical features of the pavement and cannot correlate with surface deformation information, making it prone to misjudgments and omissions, and difficult to distinguish between similar defects such as hidden cracks and loose subgrade; the other is surface deformation detection based on three-dimensional laser point cloud data, which can only reflect micro-settlement characteristics of the pavement surface and cannot identify hidden hazards within the pavement, thus failing to achieve accurate defect location and type identification. The limitations of single-modal data result in low accuracy of defect prediction, making it difficult to meet the high-standard inspection requirements of airport pavements.

[0003] Some improved solutions attempt to combine two modalities of data for detection, but they fail to achieve deep synergy of multimodal features, still do not incorporate environmental condition data, and ignore the interference of environmental factors on detection accuracy. Airport pavement inspection scenarios are complex. Environmental factors such as temperature and humidity changes, pavement aging, and material differences can cause systematic deviations in core features such as radar wave velocity and point cloud deformation, reducing the accuracy of feature extraction and consequently affecting the accuracy and stability of defect prediction. Summary of the Invention

[0004] The main objective of this disclosure is to propose a method and apparatus for identifying latent defects in airport concrete pavements based on multimodal methods, which can improve the accuracy and stability of latent defect identification.

[0005] A first aspect of this application provides a method for identifying latent defects in airport concrete pavements based on multimodal methods, the method comprising: The radar echo, laser point cloud, and corresponding environmental conditions of the target area in the airport road are obtained, and the features of the radar echo, laser point cloud, and environmental conditions are extracted respectively to obtain radar echo features, laser point cloud features, and environmental condition features. Calculate a first correlation value between the radar echo features and the disease feature template, and calculate a second correlation value between the laser point cloud features and the disease feature template; wherein, the calculation process of the disease feature template includes: ; ; in, As a template for disease characteristics, This represents the total number of disease categories. For the first The total number of inherent characteristics contained in a disease class. For the first The inherent characteristics of this type of disease, For the first The first type of disease corresponding to An inherent characteristic; the inherent characteristic is a characteristic that is not affected by the environmental conditions. The fusion weight values ​​corresponding to the radar echo features, laser point cloud features and environmental condition features are dynamically adjusted based on the first correlation value and the second correlation value, and the radar echo features, laser point cloud features and environmental condition features are fused based on the fusion weight values ​​to obtain fused features; The latent diseases in the target area are identified based on the fusion features.

[0006] The method provided in this application has at least the following beneficial effects: This method acquires data from two different modalities—radar echo and laser point cloud—and extracts their respective features, achieving comprehensive capture of information on the internal structure and surface deformation of airport roads. This avoids the limitations of single-modal data in encompassing both internal and external information, significantly improving the accuracy and completeness of defect identification. Furthermore, addressing the shortcomings of existing technologies in achieving deep multimodal feature synergy and ignoring environmental interference, this method introduces the concept of defect feature templates and dynamically adjusts feature fusion weights based on correlation values. The construction of defect feature templates, particularly their inherent features unaffected by environmental conditions, allows the defect identification process to focus on the defects themselves. By leveraging the inherent properties of the disease itself, the interference of environmental factors on feature extraction and recognition accuracy is effectively reduced. The fusion weight value is dynamically adjusted based on the first and second correlation values. It can intelligently allocate the contribution of different modal features according to the degree of matching between the current data and the disease template. For example, when radar data has a higher correlation with a certain internal disease, its weight will be increased, thus occupying a more dominant position in the fusion features. Compared with the simple linear superposition or fixed weight fusion method in the existing technology, this achieves a deeper level of feature synergy, making the fusion features more representative and robust, and ultimately improving the accuracy and stability of latent disease identification.

[0007] A second aspect of this application provides a multimodal device for identifying latent defects in airport concrete pavements, the device comprising: The data acquisition unit is used to acquire radar echoes, laser point clouds, and corresponding environmental conditions of the target area in the airport road, and extract the features of the radar echoes, laser point clouds, and environmental conditions respectively to obtain radar echo features, laser point cloud features, and environmental condition features. A relation calculation unit is used to calculate a first correlation value between the radar echo features and the disease feature template, and a second correlation value between the laser point cloud features and the disease feature template; wherein, the calculation process of the disease feature template includes: ; ; in, As a template for disease characteristics, This represents the total number of disease categories. For the first The total number of inherent characteristics contained in a disease class. For the first The inherent characteristics of this type of disease For the first The first type of disease corresponding to An inherent characteristic; the inherent characteristic is a characteristic that is not affected by the environmental conditions. The feature fusion unit is used to dynamically adjust the fusion weight values ​​corresponding to the radar echo feature, the laser point cloud feature and the environmental condition feature according to the first association value and the second association value, and fuse the radar echo feature, the laser point cloud feature and the environmental condition feature based on the fusion weight values ​​to obtain the fused feature; A disease prediction unit is used to identify latent diseases in the target area based on the fused features.

[0008] A third aspect of this application provides an electronic device including at least one controller and a memory for communicatively connecting to the controller; the memory stores instructions executable by the at least one controller to cause the at least one controller to perform a multimodal method for identifying latent defects in airport concrete pavements as described above.

[0009] A fourth aspect of this application provides a computer-readable storage medium storing computer-executable instructions for causing a computer to perform the above-described method for identifying latent defects in airport concrete pavements based on multimodality.

[0010] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a schematic diagram of a method for identifying latent defects in airport concrete pavements based on multimodal analysis, as provided in this application. Figure 2 This is a schematic diagram of a multimodal airport concrete pavement latent defect identification device provided in this application; Figure 3 This is a schematic diagram of the electronic device provided in this application. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0014] In the description of this application, the use of terms such as "first," "second," etc., is for the purpose of distinguishing technical features only and should not be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated or the order of the technical features indicated.

[0015] In the description of this application, it should be understood that the orientation descriptions, such as up, down, etc., are based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed or function in a specific orientation, and therefore should not be construed as a limitation of this application.

[0016] like Figure 1 As shown in one embodiment of this application, a method for identifying latent defects in airport concrete pavements based on multimodal methods is provided. The method includes: Step S110: Obtain radar echo, laser point cloud and corresponding environmental conditions of the target area in the airport road, and extract the features of radar echo, laser point cloud and environmental conditions respectively to obtain radar echo features, laser point cloud features and environmental condition features. Step S120: Calculate the first correlation value between radar echo features and the disease feature template, and calculate the second correlation value between laser point cloud features and the disease feature template; wherein, the calculation process of the disease feature template includes: ; ; in, As a template for disease characteristics, This represents the total number of disease categories. For the first The total number of inherent characteristics contained in a disease class. For the first The inherent characteristics of this type of disease, For the first The first type of disease corresponding to An inherent characteristic; an inherent characteristic is a characteristic that is not affected by environmental operating conditions; Step S130: Dynamically adjust the fusion weight values ​​of the corresponding radar echo features, laser point cloud features and environmental condition features according to the first correlation value and the second correlation value, and fuse the radar echo features, laser point cloud features and environmental condition features based on the fusion weight values ​​to obtain the fused features; Step S140: Identify latent diseases in the target area based on the fusion features.

[0017] For ease of understanding, the following explains some key terms in this embodiment: Multimodal analysis refers to acquiring information from different types of data sources and combining them for analysis. In this method, multimodal data mainly includes radar echo data and laser point cloud data, which reflect the internal structure and surface deformation information of airport roads from different dimensions.

[0018] Airport pavement hidden defects refer to defects or damages that exist in the airport pavement structure but have no obvious visible signs on the surface, such as voids in the bottom of the slab, voids in the base layer, and hidden internal cracks.

[0019] Radar echo refers to the signal reflected back when the electromagnetic waves emitted by a ground-penetrating radar system propagate through a medium and encounter interfaces with different dielectric constants.

[0020] Laser point cloud refers to a set of three-dimensional spatial points obtained through laser scanning equipment, where each point contains its coordinate information in space.

[0021] Environmental conditions refer to external environmental factors that affect data collection and feature representation when identifying defects on airport roads, such as temperature, humidity, pavement material type, and pavement service life. These factors may cause changes in radar echo and laser point cloud characteristics, thereby affecting the accuracy of defect identification.

[0022] This method includes the following main steps: First, acquire radar echoes, laser point clouds, and environmental conditions, and extract three types of features; then, construct a disease feature template. First, the correlation values ​​between radar, point cloud, and template are calculated. Second, the fusion weights of the three types of features are dynamically adjusted based on the correlation values, and then fused to obtain the fused features. Finally, identify latent diseases. The detailed instructions are as follows: First, the steps involve acquiring radar echoes, laser point clouds, and corresponding environmental conditions of the target area along the airport road. Radar echoes can be collected using ground-penetrating radar (GPR) equipment. For example, a GPR antenna can be placed on a testing vehicle, which travels along the airport road at a fixed speed, continuously emitting electromagnetic waves and receiving reflected signals. Laser point clouds can be acquired using a 3D laser scanner. For example, a laser scanner can be mounted on the top of the testing vehicle, performing high-density scanning of the road surface during travel to generate 3D point cloud data. Environmental condition data can be acquired using environmental sensors installed on the testing vehicle, such as thermometers and hygrometers.

[0023] After acquiring the raw data, feature extraction needs to be performed on the radar echo, laser point cloud, and environmental conditions respectively. For example, for radar echo, basic physical quantities such as amplitude and attenuation rate can be extracted as features; for laser point cloud, geometric features such as micro-settlement and surface roughness can be extracted; for environmental conditions, raw values ​​such as temperature and humidity can be directly used as features. Thus, radar echo features, laser point cloud features, and environmental condition features are obtained.

[0024] It should be noted that, due to the different dimensions of various features, after the initial feature extraction, features of different dimensions can be mapped to the same high-dimensional space. Since this is a common practice in this field, it will not be elaborated here.

[0025] Secondly, calculate the first correlation value between radar echo features and disease feature templates, and the second correlation value between laser point cloud features and disease feature templates.

[0026] The construction of disease feature templates is the key to this embodiment. For example, radar echoes and laser point clouds of a large number of known disease samples (such as slab bottom voids, base layer cavities, etc.) can be collected in advance, and inherent features unaffected by the environment can be extracted from them. Inherent characteristics are unique and distinctive features of various pavement defects, possessing strong stability and high distinguishability. Eliminating external interference factors, they accurately reflect the structural damage characteristics of the defect itself. Types of inherent characteristics include, for example, radar-related inherent characteristics: echo amplitude, wave velocity, attenuation gradient, and phase axis distortion; point cloud-related inherent characteristics: micro-settlement, settlement gradient, and compactness of the deformation area; and structural inherent characteristics of defects: damage location, depth, extent, and morphology. These inherent characteristics can include energy distribution within a specific frequency range, reflection intensity patterns at specific depths, etc.

[0027] Then, by weighting and averaging these inherent characteristics, inherent characteristics representing various diseases are formed. And further aggregate to obtain disease characteristic templates. In practical applications, the defect feature template can be stored in a database and retrieved during detection. Subsequently, the radar echo features extracted from the current target area are compared with the defect feature template to calculate a first correlation value, which measures the similarity between the current radar features and known defect patterns. Similarly, a second correlation value is calculated between the laser point cloud features and the defect feature template.

[0028] Here, the vector mean of the fused features of disease samples of the same category is used to obtain the category standard features. Then, the samples of each category are weighted and aggregated to obtain the global baseline features. This serves as a unified reference for disease feature matching. Then, the degree of matching between the current feature and the standard disease template is quantified based on the correlation value, providing an objective basis for dynamic fusion weights, realizing adaptive weighting of multimodal features, and improving the rationality of fusion and the robustness of disease identification.

[0029] Secondly, if the first correlation value between the radar echo features and the disease feature template is high, it indicates that the radar data may more reliably indicate the disease, and the fusion weight value of the radar echo features can be increased accordingly. Conversely, if the second correlation value of the laser point cloud features is high, the fusion weight value of the laser point cloud features can be increased. The weight value of the environmental condition features can be preset or empirically adjusted according to their influence on the disease features. After obtaining the dynamically adjusted fusion weight values, the radar echo features, laser point cloud features, and environmental condition features are fused based on these weight values ​​to obtain the fused features. For example, the fusion process can adopt a simple linear weighted summation method, multiplying each modal feature by its corresponding weight value and then adding them together to form a comprehensive fused feature vector.

[0030] Finally, latent diseases in the target area are identified based on the fusion features. For example, the obtained fusion features can be input into a pre-trained classification model, such as a support vector machine (SVM), neural network, or decision tree. By learning a large number of labeled fusion feature samples, the classification model can identify the type or severity of the disease represented by the fusion features. Thus, it can output whether latent diseases exist in the target area and the specific type and location of the diseases.

[0031] The following example will provide a more detailed explanation of the above technical solution: Suppose that a hidden defect detection is required on a section of runway at airport location A. There is no obvious visible damage on the runway surface, but historical data suggests a potential risk of underslab delamination or base layer voids in the area.

[0032] First, a detection vehicle equipped with ground-penetrating radar, a 3D laser scanner, and environmental sensors drives over the target area of ​​the runway. The ground-penetrating radar continuously emits and receives electromagnetic waves to acquire radar echo data of the area. The 3D laser scanner simultaneously scans the runway surface, generating high-precision laser point cloud data. At the same time, the environmental sensors record environmental condition data such as temperature and humidity in real time. Subsequently, the data acquisition unit processes this raw data, extracting radar echo features from the radar echo, such as abnormal change patterns in echo amplitude; extracting laser point cloud features from the laser point cloud, such as micro-settlement and its gradient; and extracting environmental condition features from the environmental condition data, such as the current temperature and humidity values.

[0033] Next, a pre-established disease feature template is invoked. This template is built upon a large amount of historical disease data and includes inherent features of various known diseases (such as slab bottom voids and base layer cavities). The currently extracted radar echo features are compared with the disease feature template to calculate a first correlation value. For example, if the radar echo features are highly similar to the inherent features of slab bottom voids, the first correlation value will be higher. Similarly, a second correlation value is calculated between the laser point cloud features and the disease feature template. If the laser point cloud features show a local settlement pattern related to base layer cavities, the second correlation value will be higher.

[0034] Subsequently, based on the calculated first and second correlation values, the fusion weights of radar echo features, laser point cloud features, and environmental condition features are dynamically adjusted. For example, if the first correlation value is much higher than the second correlation value, it indicates that radar data may contribute more to disease identification in the current context, and the fusion weight of the radar echo features will be increased accordingly; conversely, the lower the correlation value, the lower the environmental condition feature weight. The weight of environmental condition features is also taken into consideration. For example, in high temperature and high humidity environments, the reliability of some features may decrease, and their weights may be adjusted appropriately. After the weights are determined, the feature fusion unit fuses the radar echo features, laser point cloud features, and environmental condition features based on these dynamically adjusted fusion weights to generate a comprehensive fusion feature. This fusion feature contains effective information from different modalities and environmental factors, enabling a more comprehensive characterization of the potential disease status of the target area.

[0035] Finally, a pre-trained disease identification model is run internally. The model analyzes the fused features and outputs the identification results. For example, the model may identify the presence of a hidden disease, such as a void at the bottom of the slab, in the target area and provide its approximate location and severity assessment.

[0036] Through the above process, the method can comprehensively utilize multi-source information such as radar echoes, laser point clouds, and environmental conditions, and by dynamically adjusting the feature fusion weights, it enables different modal data to play their respective advantages in defect identification, thereby achieving accurate identification of hidden defects in airport roads.

[0037] This method acquires data from two different modalities—radar echo and laser point cloud—and extracts their respective features, achieving comprehensive capture of information on the internal structure and surface deformation of airport pavements. This avoids the limitations of single-modal data in capturing both internal and external information, significantly improving the accuracy and completeness of defect identification. Furthermore, this method introduces the concept of defect feature templates and dynamically adjusts feature fusion weights based on correlation values. The fusion weights are dynamically adjusted according to the first and second correlation values, intelligently allocating the contribution of different modal features based on the degree of matching between the current data and the defect template. Compared to the simple linear superposition or fixed-weight fusion methods in existing technologies, this achieves a deeper level of feature synergy, making the fused features more representative and robust, ultimately improving the accuracy and stability of latent defect identification. Therefore, this method effectively addresses the technical requirements of high precision and high adaptability in the detection of latent defects on airport pavements.

[0038] In some embodiments of this application, the fusion weight values ​​of the corresponding radar echo features, laser point cloud features, and environmental condition features are dynamically adjusted based on the first correlation value and the second correlation value, including: (1); (2); (3); in, These are radar echo characteristics Laser point cloud features The fusion weight value of environmental operating condition characteristics, For the number of iterations, This is the first correlation value between radar echo characteristics and disease characteristic templates. This is the second correlation value between the laser point cloud features and the disease feature template; The process of calculating the correlation value between any feature and the disease feature template includes: (4); (5); (6); in, It can be any one of radar echo characteristics, laser point cloud characteristics, and environmental condition characteristics. and They are respectively The minimum and maximum values ​​of the global numerical boundary. and They are respectively The minimum and maximum values ​​of the global numerical boundary. for and The inner product between them for and The inner product between them for The vector 2 norm, for The vector 2 norm.

[0039] The fusion weight values ​​are coefficients used to measure the importance of different modal features in the final fused features. These weight values ​​can be dynamically adjusted to reflect the contribution of different features to disease identification. The fusion weight values ​​can be normalized to ensure that the sum of all weight values ​​is 1.

[0040] Number of iterations This refers to the number of times the algorithm repeatedly executes the weight update step during the dynamic adjustment of the fusion weight values. Through multiple iterations, the fusion weight values ​​can gradually converge to a better state, thus more accurately reflecting the actual contribution of each modality feature. The iteration process can terminate based on a preset maximum number of iterations, or when the change in weight values ​​is less than a certain preset threshold.

[0041] Associated values It is used to quantify a specific feature. (Such as radar echo characteristics, laser point cloud characteristics, or environmental condition characteristics) and disease characteristic templates Similarity or correlation indices between features are used to measure the similarity or correlation between features and disease feature templates. The higher the correlation value, the more similar the features are to the disease feature templates, and the greater the potential contribution to disease identification. The correlation value can be calculated using various methods. For example, in addition to cosine similarity, Pearson correlation coefficient, the inverse of Euclidean distance, or similarity measurement methods based on neural network learning can also be used.

[0042] Normalization characteristics This refers to the characteristics Disease characteristic template The feature representation after scaling the numerical range to a standard interval helps to eliminate the differences in units and numerical ranges between different features, and avoids certain features with larger values ​​from dominating the calculation of correlation values, thereby ensuring that all features are comparable when calculating correlation values. In addition to linear scaling (Min-Max normalization), normalization methods can also use Z-score standardization (mean-standard deviation normalization) or Sigmoid function normalization, etc.

[0043] The inner product is a multiplication operation between two vectors, and its result is a scalar that reflects the similarity between the two vectors in direction. In the context of normalized features, a larger inner product usually indicates that the two feature vectors are closer in direction, i.e., the higher their similarity. The vector 2-norm, also known as the Euclidean norm, is a measure of the length or magnitude of a vector. For a vector, its 2-norm is the square root of the sum of the squares of its components. When calculating cosine similarity, the vector 2-norm is used to normalize the inner product, thereby eliminating the influence of vector length on the similarity calculation, so that the similarity depends only on the angle between the vectors.

[0044] This embodiment addresses the challenge of accurately quantifying the contribution of each modality feature to disease identification during multimodal feature fusion by introducing an iterative fusion weight adjustment mechanism and a standardized correlation value calculation method. Specifically, after acquiring radar echo features, laser point cloud features, and environmental condition features, these features, along with a pre-constructed disease feature template, are first normalized to obtain... and This is done to eliminate the dimensional differences between different features; subsequently, the normalized feature vectors are calculated. With the normalized disease feature template Cosine similarity between them, i.e., correlation value To objectively measure the matching degree between each modal feature and the disease feature template, the method adopts an iterative update approach during the dynamic adjustment of the fusion weight values. In the middle, based on the correlation values ​​between the current modal features and the disease feature templates... and The radar echo features are updated by summing the association values ​​of all features with the disease feature template. and laser point cloud features Fusion weights and The fusion weights of environmental operating condition characteristics The weights are determined by ensuring that the sum of all weights is 1. This iterative update mechanism allows the weight values ​​to be adaptively adjusted according to the actual correlation between each modal feature and the disease feature template. Features with high correlation receive greater weights, and vice versa. Through multiple iterations, the weight values ​​gradually converge, so that the fused features can more accurately reflect the essential characteristics of the disease. This effectively avoids the problem of decreased recognition accuracy caused by subjective weight setting or improper weight adjustment. This strategy of dynamically adjusting weights based on correlation allows different modal features to play their due role in the fusion process, thereby improving the accuracy and robustness of airport road hidden disease identification.

[0045] In some embodiments of this application, fused features are obtained by fusing radar echo features, laser point cloud features, and environmental condition features based on fusion weight values, including: (7); in, As a feature of fusion, This represents the linear correlation strength coefficient between radar echo characteristics and laser point cloud characteristics. This represents the linear correlation strength coefficient between laser point cloud features and environmental condition features. This represents the linear correlation strength coefficient between radar echo characteristics and environmental condition characteristics. For constraint coefficients, This represents the maximum number of iterations.

[0046] Fusion features It is a comprehensive feature representation that integrates radar echo features, laser point cloud features, and environmental condition features. It aims to provide more comprehensive and discriminative information than single-modal features by combining multi-source information, thereby more effectively characterizing the hidden defects in the target area.

[0047] Linear correlation strength coefficient Used to quantify the degree of linear correlation between radar echo features and laser point cloud features; similarly, the linear correlation strength coefficient. This is used to quantify the degree of linear correlation between laser point cloud features and environmental condition features, while the linear correlation strength coefficient... These coefficients are used to quantify the degree of linear correlation between radar echo characteristics and environmental condition characteristics. The larger the absolute value of these coefficients, the stronger the linear relationship between the corresponding feature pairs. These coefficients can be obtained by calculating the Pearson correlation coefficient or by other statistical methods such as analysis of covariance.

[0048] constraint coefficient It is a parameter between 0 and 1, whose function is to balance the fusion features. The weighted sum of individual characteristics and the contribution of the interaction term between characteristics. Constraint coefficients. The value can be optimized on the training data through machine learning hyperparameter tuning methods such as cross-validation and grid search to achieve the best disease identification performance, or it can be preset based on empirical values ​​or domain knowledge.

[0049] This embodiment introduces a more refined feature fusion mechanism to overcome the limitations of traditional weighted fusion in capturing complex feature interactions. Specifically, after obtaining the fusion weight values ​​dynamically adjusted after the maximum number of iterations T, this method fuses radar echo features, laser point cloud features, and environmental condition features. The fusion process considers not only the independent contribution of each feature (i.e., ...) The linear correlation strength coefficient between feature pairs is introduced to characterize the interaction between features (i.e., This fusion method enables the fusion features to more comprehensively and deeply characterize the defects of airport roads, especially for latent defects where single-modal signals are not obvious but multimodal interactions are significant, providing stronger discrimination capabilities.

[0050] In some embodiments of this application, the calculation process of the linear correlation strength coefficient includes: (8); (9); (10); in, For feature dimension, The first of the radar echo characteristics Numerical values ​​of 1D feature components The first feature in laser point cloud Numerical values ​​of 1D feature components The first of the environmental operating conditions characteristics Numerical values ​​of 1D feature components The mean value of the features obtained for all dimensions of the radar echo characteristics. The feature mean is calculated for all dimensions of the laser point cloud feature. The mean value of features is obtained for all dimensions of environmental operating conditions.

[0051] This embodiment calculates the linear correlation strength coefficient between radar echo characteristics and laser point cloud characteristics. The linear correlation strength coefficient between laser point cloud features and environmental condition features And the linear correlation strength coefficient between radar echo characteristics and environmental operating condition characteristics. This quantifies the degree of linear correlation between different modal features. Specifically, for each pair of features, such as radar echo features and laser point cloud features, firstly, the feature component values ​​of each feature are obtained across all feature dimensions D. and Next, the eigenvalue of each feature is calculated. and The first step is to center the feature components. Then, using the Pearson correlation coefficient formula, the linear correlation strength coefficient is obtained by summing the products of the centered feature components and dividing by the product of their respective standard deviations. Similarly, it can be calculated that and The introduction of these linear correlation strength coefficients allows for consideration of not only the correlation between each modal feature and the disease feature template, but also the interaction between modal features, such as synergistic enhancement or redundant information, during subsequent feature fusion. By incorporating these correlation strength coefficients into the fusion formula, the latent disease information of the target area can be characterized more comprehensively and accurately, thereby improving the accuracy and robustness of disease identification.

[0052] In some embodiments of this application, the environmental operating condition characteristics include at least one environmental operating condition correction factor; Before calculating the first correlation value between radar echo features and the disease feature template, and the second correlation value between laser point cloud features and the disease feature template, the following steps are also included: The environmental condition correction coefficient is multiplied by the radar echo characteristics and the laser point cloud characteristics respectively to correct the radar echo characteristics and the laser point cloud characteristics.

[0053] This embodiment introduces a step of correcting the radar echo features and laser point cloud features based on environmental condition correction coefficients before calculating the first correlation value between radar echo features and the disease feature template, and the second correlation value between laser point cloud features and the disease feature template.

[0054] Specifically, after acquiring radar echoes, laser point clouds, and corresponding environmental conditions of the target area on the airport road, the system first extracts radar echo features, laser point cloud features, and environmental condition features. Then, based on the current environmental condition features, corresponding environmental condition correction coefficients are determined. These correction coefficients are applied to the radar echo features and laser point cloud features respectively, adjusting the original features through multiplication operations to compensate for or eliminate interference from environmental factors (such as temperature and humidity). For example, if high humidity leads to increased radar echo signal attenuation, a humidity correction coefficient will be applied to adjust the signal. The radar echo features are standardized to a state unaffected by humidity. The corrected radar echo features and laser point cloud features will more accurately reflect the true physical characteristics of hidden defects on airport roads. Based on this, the first correlation value between the corrected radar echo features and the defect feature template, as well as the second correlation value between the corrected laser point cloud features and the defect feature template, are calculated. This pre-correction process ensures the accuracy of subsequent correlation value calculations, thus providing a more reliable basis for dynamically adjusting the fusion weight values ​​and the final fusion features, significantly improving the robustness and accuracy of the entire defect identification method.

[0055] In some embodiments of this application, the environmental operating conditions characteristics include at least one of the following: temperature correction factor, humidity correction factor, pavement age correction factor, and material correction factor.

[0056] In some embodiments of this application, radar echo characteristics include at least one of the following: echo amplitude, attenuation coefficient gradient, in-phase axis texture complexity, and electromagnetic wave propagation speed. Laser point cloud features include at least one of the following: micro-settlement amount, settlement gradient, and settlement region compactness.

[0057] For ease of understanding, one embodiment of this application provides a method for identifying latent defects in airport concrete pavements based on multimodal methods, comprising the following steps: Step S210: Build an integrated vehicle-mounted synchronous inspection platform to complete the synchronous acquisition and spatiotemporal registration of three types of modal data, laying the foundation for subsequent feature extraction. The specific acquisition content is as follows: (1) High-frequency ground-penetrating radar (1.5GHz-2.0GHz) collects radar echo signals of underground media to capture the physical characteristics of hidden defects (bottom voids, base voids, etc.) inside the pavement; (2) Millimeter-level three-dimensional lidar collects high-density point cloud data of the pavement to capture the related deformation features such as surface micro-settlement and slab height difference corresponding to hidden diseases; (3) The environmental perception module collects parameters such as on-site temperature and humidity, pavement age, and pavement material type to correct environmental interference and improve the stability of feature extraction and model discrimination.

[0058] Step S220, data processing; Based on the service characteristics of airport pavements, latent defects specifically refer to defects where the surface has no obvious damage or only minor damage, but the main hidden dangers exist inside the pavement or beneath the surface. These are further divided into five core types: Internal cracks in the pavement (latent cracks) are mainly distributed within the pavement surface layer and base layer. They have no obvious cracks on the surface or only have fine hairline cracks, which cannot be directly identified by the naked eye. They are easy to gradually expand and cause damage to the pavement structure, and are one of the most common hidden defects of airport pavements.

[0059] Voiding at the bottom of the slab mainly occurs between the cement concrete pavement slab and the base layer. Due to erosion and aging of the base layer, the bottom of the slab separates from the base layer, forming a void. Although there is no obvious damage on the surface, the load-bearing capacity is greatly reduced, which can easily lead to pavement slab breakage. It is a core hidden danger for airport runways and taxiways.

[0060] Voids in the base layer / subgrade are hollow areas located inside the pavement base layer or subgrade, caused by groundwater erosion, soil settlement, pipeline leakage, etc. They are highly concealed and can easily lead to sudden pavement settlement and damage, endangering flight safety.

[0061] Interlayer separation occurs between different structural layers of the pavement (such as surface layer and base layer, base layer and subgrade). It is caused by the failure of interlayer bonding due to aging of bonding materials and construction defects. Although there is no obvious damage on the surface, it will lead to a decrease in the overall structure of the pavement and easily cause surface peeling and crack expansion.

[0062] The base layer is loose. The density of the pavement base material has decreased due to weathering, water erosion, and insufficient compaction. Although there is no obvious damage on the surface, the load-bearing capacity is insufficient, which can easily induce secondary diseases such as voids and cracks at the bottom of the slab.

[0063] (1) Extraction of radar echo features; Based on the thickness parameters of the standard structural layers (surface layer, base layer, and subgrade) of the airport pavement, the radar B-scan two-dimensional map is layered and extracted to extract the radar echo signal of each structural layer, so as to avoid the deep defects signal being masked by the upper medium signal. Radar echo characteristics: ;in: Echo amplitude; The attenuation coefficient gradient; For the texture complexity of the same phase axis; The propagation speed of electromagnetic waves. Four types of characteristics were identified through correlation value calculations (…). All of them are highly correlated with latent diseases and participate in the subsequent fusion; at the same time, the core features are normalized and the feature values ​​are mapped to the 0~1 range to ensure the rationality of the fusion.

[0064] (2) Extraction of laser point cloud features; Based on the normal pavement elevation distribution pattern, a quadratic polynomial is used to fit the dynamic pavement reference surface. To avoid the problem that a fixed reference plane cannot adapt to the overall slope of the pavement, based on effective deformation points, a feature vector specific to the point cloud mode is first constructed, and then each feature term is extracted separately, as follows: Laser point cloud features: ;in: The formula for calculating the micro-settlement is as follows: ( For dynamic reference surface elevation, (Elevation of the measured point); Settlement gradient; For the compactness of the settlement area.

[0065] (3) Extraction of environmental operating condition characteristics; Based on the collected temperature and humidity pavement age pavement material The testing conditions are divided into four categories; Environmental modal feature vectors: ;in: This is a temperature correction factor; This is the humidity correction factor; This is the pavement age correction factor; This is the material correction factor.

[0066] Step S230: Correct the radar echo characteristics and laser point cloud characteristics based on environmental conditions; Because the measured values ​​of radar and point cloud features are affected by environmental conditions (such as temperature, humidity, material, and age), systematic deviations (feature anomalies not caused by defects) may occur. Features can be used as quantitative compensation parameters for environmental interference.

[0067] Radar echo characteristics are multiplied by the corresponding environmental correction coefficients (amplitude) Multiply by humidity correction factor Wave speed Multiply by temperature correction factor attenuation coefficient gradient In-phase axis texture complexity Multiply by material correction factor Laser point cloud features and temperature correction coefficients, respectively. Material correction factor Multiplication, through coefficient weighting, achieves feature correction and improves the stability of features under different operating conditions.

[0068] Step S240, Feature fusion; First, feature dimension unification: the corrected radar echo features (4D) Laser point cloud features (3D), through a feature mapping network, uniformly mapped to the same high-dimensional feature space (16D), ensuring the feasibility of fusion; environmental operating condition features (4-dimensional) synchronous mapping to the 16-dimensional space to achieve the unification of the three-modal feature dimensions.

[0069] Initial weights of radar echo features ; Initial weights of laser point cloud features ; Initial weights of environmental operating condition characteristics ; Adaptive adjustment of modal feature weights is achieved through correlation value calculation and dynamic weight iteration: First, mutual information value calculation: Calculate the correlation value between each type of modal feature (corrected radar, point cloud, and environmental features) and the preset disease feature template. The correlation value represents the degree of correlation between a single modal feature and the nature of the disease. The higher the value, the greater the contribution of the modal feature to the identification of the disease at the current location, and the higher the weight should be assigned.

[0070] The calculation formula is detailed in the formula above and will not be elaborated here.

[0071] The data was pre-generated using an offline training method. The specific steps were as follows: First, a large amount of measured tri-modal data of airport pavement was collected, covering samples from five core hidden defects and defect-free areas, encompassing different pavement materials, working conditions, and defect severity. Second, feature extraction, environmental correction, and simplified fusion were performed on each sample, and the corresponding defect type and core feature parameters were labeled. Third, the K-means clustering algorithm was used to cluster the fused features of all labeled samples, extracting the inherent features of each defect and defect-free area. Finally, all inherent features were integrated into a pre-set defect feature template with the same dimensions as the fused feature vector (16 dimensions). It is embedded into the adaptive attention module and can be directly called during online detection without repeated training.

[0072] Based on the initial weights and correlation values, the weights are iteratively optimized using formulas (1) to (6) to obtain the final weights. Then, the three-modal features are weighted to obtain the final 16-dimensional high-dimensional fusion feature vector, which can be found in formula (7). This not only considers the independent contribution of each feature (i.e., ...) The linear correlation strength coefficient between feature pairs is introduced to characterize the interaction between features (i.e., This fusion method enables the fusion features to more comprehensively and deeply characterize the defects of airport roads, especially for latent defects where single-modal signals are not obvious but multimodal interactions are significant, providing stronger discrimination capabilities.

[0073] Step S250, output the result; A 1×1 convolution kernel is used to compress the 16-dimensional fused features to 8 dimensions, eliminating redundant features and reducing computational consumption while retaining core defect features. A dedicated residual branch for airport pavement defects is embedded to enhance the expression of specific features based on the characteristic differences of different types of latent defects (such as underslab delamination and base layer voids), solving the problem of deep feature loss. A dual-output structure is adopted to separately determine the presence or absence of defects and classify defect types, outputting two types of results to meet detection requirements. Output 1: Presence or absence of disease (binary classification): Output the discrimination probability through the sigmoid activation function. , It was determined to be diseased. Determined to be free of disease; Output 2: Disease Type (Multi-classification): For the four core latent diseases, namely, slab bottom void, base layer void, interlayer separation, and roadbed looseness, the probability of each disease is output through the softmax activation function, and the maximum probability is taken as the final disease type.

[0074] Loss function design: A hybrid loss function combining binary classification loss and multi-class classification loss is adopted to optimize the model's discrimination accuracy. The loss function is as follows: (12); in: The cross-entropy loss is used for binary classification to determine whether a disease exists or not. Multi-class cross-entropy loss for disease type classification; (Balancing the weights of the two types of discrimination tasks).

[0075] Model training and optimization: The gradient descent algorithm is used to iteratively train the model, while an early stopping strategy is introduced to avoid overfitting. During the training process, the model parameters are fine-tuned in a targeted manner based on the mechanism of airport pavement defects to improve the model's ability to distinguish between minor and deep defects.

[0076] like Figure 2 One embodiment of this application provides a multimodal-based device for identifying latent defects in airport concrete pavements. The device includes: The data acquisition unit 1001 is used to acquire radar echoes, laser point clouds and corresponding environmental conditions of the target area in the airport road, and extract the features of radar echoes, laser point clouds and environmental conditions respectively to obtain radar echo features, laser point cloud features and environmental condition features. The relation calculation unit 1002 is used to calculate the first correlation value between radar echo features and the defect feature template, and the second correlation value between laser point cloud features and the defect feature template; wherein, the calculation process of the defect feature template includes: ; ; in, As a template for disease characteristics, This represents the total number of disease categories. For the first The total number of inherent characteristics contained in a disease class. For the first The inherent characteristics of this type of disease For the first The first type of disease corresponding to An inherent characteristic; an inherent characteristic is a characteristic that is not affected by environmental operating conditions; The feature fusion unit 1003 is used to dynamically adjust the fusion weight values ​​of the corresponding radar echo features, laser point cloud features and environmental condition features according to the first correlation value and the second correlation value, and fuse the radar echo features, laser point cloud features and environmental condition features based on the fusion weight values ​​to obtain the fused features; The disease prediction unit 1004 is used to identify latent diseases in the target area based on the fusion features.

[0077] It should be noted that the multimodal airport concrete road hidden defect identification device provided in this embodiment is based on the same inventive concept as the multimodal airport concrete road hidden defect identification method described above. Therefore, the content of the multimodal airport concrete road hidden defect identification device described in this embodiment is also applicable to the content of the multimodal airport concrete road hidden defect identification method described above, and will not be repeated here.

[0078] like Figure 3 One embodiment of this application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described method for identifying latent defects in airport concrete pavements based on multimodal computing. The electronic device includes: At least one battery; At least one memory; At least one processor; At least one program; The program is stored in memory, and the processor executes at least one program to implement the above-described method for identifying latent defects in airport concrete pavements based on multimodality.

[0079] This electronic device can be any smart terminal, including mobile phones, tablets, personal digital assistants (PDAs), and in-vehicle computers.

[0080] The electronic devices according to embodiments of this application will now be described in detail.

[0081] The processor 1600 can be implemented using a general-purpose central processing unit (CPU), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this disclosure. The memory 1700 can be implemented as a read-only memory (ROM), static storage device, dynamic storage device, or random access memory (RAM). The memory 1700 can store the operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 1700 and is called and executed by the processor 1600 to perform a multimodal method for identifying latent defects in airport concrete pavements according to an embodiment of this disclosure.

[0082] The input / output interface 1800 is used to implement information input and output. The communication interface 1900 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.). Bus 2000 transmits information between various components of the device (e.g., processor 1600, memory 1700, input / output interface 1800, and communication interface 1900); The processor 1600, memory 1700, input / output interface 1800 and communication interface 1900 are connected to each other within the device via bus 2000.

[0083] This disclosure also provides a storage medium, which is a computer-readable storage medium storing computer-executable instructions for causing a computer to execute the above-described method for identifying latent defects in airport concrete pavements based on multimodality.

[0084] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0085] The embodiments described in this disclosure are for the purpose of more clearly illustrating the technical solutions of this disclosure and do not constitute a limitation on the technical solutions provided by this disclosure. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by this disclosure are also applicable to similar technical problems.

[0086] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this disclosure, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0087] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0088] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0089] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0090] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0091] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0092] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0093] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0094] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause an electronic device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0095] The above is a detailed description of the preferred embodiments of this application. However, the embodiments of this application are not limited to the above-described implementation methods. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the embodiments of this application. All such equivalent modifications or substitutions are included within the scope defined by the claims of the embodiments of this application.

Claims

1. A method for identifying latent defects in airport concrete pavements based on multimodal analysis, characterized in that, The method includes: The radar echo, laser point cloud, and corresponding environmental conditions of the target area in the airport road are obtained, and the features of the radar echo, laser point cloud, and environmental conditions are extracted respectively to obtain radar echo features, laser point cloud features, and environmental condition features. Calculate a first correlation value between the radar echo features and the disease feature template, and calculate a second correlation value between the laser point cloud features and the disease feature template; wherein, the calculation process of the disease feature template includes: ； ； in, As a template for disease characteristics, This represents the total number of disease categories. For the first The total number of inherent characteristics contained in a disease class. For the first The inherent characteristics of this type of disease, For the first The first type of disease corresponding to An inherent characteristic; the inherent characteristic is a characteristic that is not affected by the environmental conditions. The fusion weight values ​​corresponding to the radar echo features, laser point cloud features and environmental condition features are dynamically adjusted based on the first correlation value and the second correlation value, and the radar echo features, laser point cloud features and environmental condition features are fused based on the fusion weight values ​​to obtain fused features; The latent diseases in the target area are identified based on the fusion features.

2. The method for identifying latent defects in airport concrete pavements based on multimodal analysis according to claim 1, characterized in that, The fusion weight values ​​corresponding to the radar echo features, the laser point cloud features, and the environmental condition features are dynamically adjusted based on the first correlation value and the second correlation value, including: ； ； ； in, These are radar echo characteristics Laser point cloud features The fusion weight value of environmental operating condition characteristics, For the number of iterations, This is the first correlation value between radar echo characteristics and disease characteristic templates. This is the second correlation value between the laser point cloud features and the disease feature template; The process of calculating the correlation value between any feature and the disease feature template includes: ； ； ； in, It can be any one of the radar echo characteristics, the laser point cloud characteristics, and the environmental condition characteristics. and They are respectively The minimum and maximum values ​​of the global numerical boundary. and They are respectively The minimum and maximum values ​​of the global numerical boundary. for and The inner product between for and The inner product between for The vector 2 norm, for The vector 2 norm.

3. The method for identifying latent defects in airport concrete pavements based on multimodal analysis according to claim 2, characterized in that, The process of fusing the radar echo features, the laser point cloud features, and the environmental condition features based on the fusion weight value to obtain fused features includes: ； in, As a feature of fusion, This represents the linear correlation strength coefficient between radar echo characteristics and laser point cloud characteristics. This represents the linear correlation strength coefficient between laser point cloud features and environmental condition features. This represents the linear correlation strength coefficient between radar echo characteristics and environmental condition characteristics. For constraint coefficients, This represents the maximum number of iterations.

4. The method for identifying latent defects in airport concrete pavements based on multimodal analysis according to claim 3, characterized in that, The calculation process of the linear correlation strength coefficient includes: ； ； ； in, For feature dimension, The first of the radar echo characteristics Numerical values ​​of 1D feature components The first feature in laser point cloud Numerical values ​​of 1D feature components The first of the environmental operating conditions characteristics Numerical values ​​of 1D feature components The mean value of the features obtained for all dimensions of the radar echo characteristics. The feature mean is calculated for all dimensions of the laser point cloud feature. The mean value of features is obtained for all dimensions of environmental operating conditions.

5. The method for identifying latent defects in airport concrete pavements based on multimodal analysis according to claim 1, characterized in that, The environmental operating condition characteristics include at least one environmental operating condition correction factor; Before calculating the first correlation value between the radar echo features and the disease feature template and the second correlation value between the laser point cloud features and the disease feature template, the method further includes: The environmental condition correction coefficient is multiplied by the radar echo feature and the laser point cloud feature respectively to correct the radar echo feature and the laser point cloud feature.

6. The method for identifying latent defects in airport concrete pavements based on multimodal analysis according to claim 5, characterized in that, The environmental operating conditions characteristics include at least one of the following: temperature correction factor, humidity correction factor, pavement age correction factor, and material correction factor.

7. The method for identifying latent defects in airport concrete pavements based on multimodal analysis according to claim 6, characterized in that, The radar echo characteristics include at least one of the following: echo amplitude, attenuation coefficient gradient, in-phase axis texture complexity, and electromagnetic wave propagation speed. The laser point cloud features include at least one of the following: micro-settlement amount, settlement gradient, and settlement region compactness.

8. A multimodal-based device for identifying latent defects in airport concrete pavements, characterized in that, The device includes: The data acquisition unit is used to acquire radar echoes, laser point clouds, and corresponding environmental conditions of the target area in the airport road, and extract the features of the radar echoes, laser point clouds, and environmental conditions respectively to obtain radar echo features, laser point cloud features, and environmental condition features. A relation calculation unit is used to calculate a first correlation value between the radar echo features and the disease feature template, and a second correlation value between the laser point cloud features and the disease feature template; wherein, the calculation process of the disease feature template includes: ； ； in, As a template for disease characteristics, This represents the total number of disease categories. For the first The total number of inherent characteristics contained in a disease class. For the first The inherent characteristics of each type of disease. For the first The first type of disease corresponding to An inherent characteristic; the inherent characteristic is a characteristic that is not affected by the environmental conditions. The feature fusion unit is used to dynamically adjust the fusion weight values ​​corresponding to the radar echo feature, the laser point cloud feature and the environmental condition feature according to the first association value and the second association value, and fuse the radar echo feature, the laser point cloud feature and the environmental condition feature based on the fusion weight values ​​to obtain the fused feature; The disease prediction unit is used to identify latent diseases in the target area based on the fused features.

9. An electronic device, characterized in that, It includes at least one controller and a memory for communicatively connecting with the controller; the memory stores instructions that can be executed by the at least one controller, the instructions being executed by the at least one controller to cause the at least one controller to perform a multimodal method for identifying latent defects in airport concrete pavements as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions for causing a computer to perform a method for identifying latent defects in airport concrete pavements based on multimodality as described in any one of claims 1 to 7.

Images