Target disease automatic identification method and system based on cloud edge collaboration

Abstract

The invention provides an automatic target disease identification method and system based on cloud-edge collaboration, and the method comprises the steps: collecting an original road surface monitoring data set through an edge calculation node disposed in a road monitoring region, generating a multi-dimensional disease feature set in the edge calculation node, uploading the multi-dimensional disease feature set to a cloud analysis platform, and carrying out the recognition of a target disease through the cloud-edge collaboration. And performing feature matching degree calculation on the multi-dimensional disease feature set and a standard disease pattern in a cloud disease feature library through a cloud analysis platform, generating a target disease type identification result and a corresponding confidence coefficient evaluation parameter, and judging a result according to a relationship between the confidence coefficient evaluation parameter and a preset threshold value. And adjusting a feature extraction strategy of the edge computing node, generating an edge node adaptive optimization instruction set, and triggering feature extraction rule updating and disease recognition model parameter iteration operation for a subsequent monitoring period. According to the method, the disease identification accuracy is improved, and the real-time performance of edge calculation and the global optimization capability of cloud analysis are considered at the same time.

Description

Technical Field

[0001] The present invention relates to the fields of disease detection and deep learning, and specifically to a method and system for automatically identifying target diseases based on cloud-edge collaboration. Background Art

[0002] With the rapid development of smart transportation technologies, automated identification of highway defects has become a critical component in ensuring safe road operations and maintenance. Current mainstream defect identification methods typically rely on centralized cloud-based processing architectures. These methods collect road monitoring data at edge nodes and upload it to the cloud for unified analysis. Defect classification results are generated based on pre-set static feature extraction rules. However, these methods have significant limitations. First, the one-way data processing process results in a lack of dynamic optimization capabilities at edge nodes, making it impossible to adjust feature extraction strategies based on real-time environmental changes. This can lead to feature mismatches under complex climate or traffic load conditions. Second, single-dimensional feature analysis struggles to effectively distinguish structural damage from environmental noise, resulting in high false positive rates. Furthermore, globally unified identification models struggle to adapt to regional variations in pavement material properties, resulting in poor regional adaptability and delayed model updates. Existing technologies lack a closed-loop control mechanism that can coordinate real-time response at the edge with global optimization in the cloud. This makes it difficult to balance identification accuracy and system efficiency, severely hindering the practical application of large-scale road network defect detection. Summary of the Invention

[0003] The present invention provides a method and system for automatic identification of target diseases based on cloud-edge collaboration.

[0004] On the one hand, the present invention provides a method for automatic identification of target defects based on cloud-edge collaboration, comprising the following steps: collecting an original road surface monitoring data set through edge computing nodes deployed in a highway monitoring area, the original road surface monitoring data set including multimodal sensing information and corresponding spatiotemporal location identifiers; in the edge computing node, generating a multidimensional defect feature set based on the original road surface monitoring data set, the multidimensional defect feature set including structural deformation features, surface texture degradation features and environmental interference correlation features; uploading the multidimensional defect feature set to a cloud analysis platform, performing feature matching calculation on the multidimensional defect feature set and a standard defect pattern in a cloud defect feature library through the cloud analysis platform, and generating a target defect type identification result and a corresponding confidence assessment parameter; adjusting the feature extraction strategy of the edge computing node according to the relationship between the confidence assessment parameter and a preset threshold, and generating an edge node adaptive optimization instruction set; synchronizing the target defect type identification result and the edge node adaptive optimization instruction set to the edge computing node, triggering feature extraction rule updates and defect identification model parameter iteration operations for subsequent monitoring cycles.

[0005] On the other hand, the present invention provides an automated disease identification system, including an edge computing node and a cloud analysis platform that communicate with each other, each of the edge computing node and the cloud analysis platform including a processor; and a memory that is communicatively connected to the processor; the memories of the edge computing node and the cloud analysis platform store instructions that can be executed by their corresponding processors, and when the instructions are executed by the corresponding processors respectively, the above-described method is executed.

[0006] The present invention provides an automated target disease identification method based on cloud-edge collaboration. It dynamically extracts multi-dimensional disease features through edge computing nodes and matches them with standard disease patterns on the cloud. It combines confidence assessment parameters to generate edge node feature extraction strategy optimization instructions in real time, and constructs a closed-loop adaptive system from data acquisition, feature optimization to model iteration. This solution can significantly improve the accuracy of disease identification in complex environments, effectively distinguish between structural damage and environmental interference through the dynamic fusion of multi-dimensional spatiotemporal features, and reduce the misjudgment rate. The confidence-driven dynamic optimization mechanism enables the system to have continuous evolution capabilities, and can adapt to changes in pavement material properties and climatic conditions in different regions, thereby enhancing the generalization performance of the model. At the same time, the collaborative mechanism of the cloud-based global error distribution heat map and edge node parameter compensation realizes regionalized precise tuning, significantly reducing the amount of data transmission while ensuring identification efficiency, and reducing system operation and maintenance costs. Through the closed-loop feedback architecture of cloud-edge collaboration, the present invention takes into account the real-time nature of edge computing and the global optimization capabilities of cloud analysis while improving the accuracy of disease identification, providing efficient and reliable technical support for highway maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 A schematic diagram of the architecture of an automatic disease identification system according to an embodiment of the present invention is shown; Figure 2 A flowchart of a method for automatic identification of target diseases based on cloud-edge collaboration according to an embodiment of the present invention is shown; Figure 3 A schematic diagram showing the composition of a computer system according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0008] Figure 1The schematic diagram of the architecture of the automatic disease identification system provided by an embodiment of the present invention is shown. The automatic disease identification system includes one or more edge computing nodes 101, a cloud analysis platform 120, and one or more networks 110 that couple the one or more edge computing nodes 101 to the cloud analysis platform 120. The cloud analysis platform 120 may include one or more general-purpose computers, dedicated server computers (such as PC servers, UNIX servers, mid-range servers), blade servers, mainframe computers, server clusters, or any other appropriate arrangements and / or combinations. The cloud analysis platform 120 may include one or more virtual machines running virtual operating systems, or other computing architectures involving virtualization (such as one or more flexible pools of logical storage devices that can be virtualized to maintain virtual storage devices of servers). In various embodiments, the cloud analysis platform 120 may run one or more services or software applications that provide the functionality described below.

[0009] The automated disease identification system may also include one or more databases 130. In certain embodiments, these databases may be used to store data and other information. Databases 130 may reside in a variety of locations. For example, the databases used by cloud-based analysis platform 120 may be local to cloud-based analysis platform 120, or they may be remote from cloud-based analysis platform 120 and communicate with cloud-based analysis platform 120 via a network-based or dedicated connection.

[0010] Please refer to Figure 2 , an embodiment of the present invention provides a flowchart of a method for automatic identification of target diseases based on cloud-edge collaboration, and the method specifically includes the following steps: Step S100: Collecting a set of original road surface monitoring data through edge computing nodes deployed in the highway monitoring area, wherein the original road surface monitoring data set contains multimodal sensing information and corresponding spatiotemporal location identifiers. The original road surface monitoring data set is a collection of data obtained by monitoring the highway road surface. Multimodal sensing information refers to information about road conditions collected by various types of sensors, such as road vibration information collected by vibration sensors, road image information collected by optical sensors, and road temperature information collected by temperature sensors. The spatiotemporal location identifier is used to mark the specific time and spatial location corresponding to each sensor information. The time identifier can be accurate to a specific moment, and the spatial location identifier can be determined by geographic coordinates and the like.

[0011] In practice, edge computing nodes can be installed at pre-set locations along highways, such as bridges, tunnels, and key road sections. For example, vibration sensors can be installed at a certain depth below the road surface, sensing minute vibrations and acquiring relevant information. Optical sensors can be mounted on roadside lampposts or specialized monitoring brackets to capture road surface images at appropriate angles. Temperature sensors can be installed on the road surface or in shallow depths to measure road surface temperature in real time. To record spatiotemporal location, edge computing nodes can be equipped with high-precision clocks to record the time of data collection and use GPS positioning modules to obtain the geographic coordinates of the collection points. For example, in a monitored section of a highway, an edge computing node collects multimodal sensor information, including vibration, optical, and temperature, every 10 seconds. This information is recorded along with the corresponding geographic coordinates, forming a collection of raw road surface monitoring data.

[0012] Step S200: In the edge computing node, a multi-dimensional disease feature set is generated based on the original pavement monitoring data set. The multi-dimensional disease feature set includes structural deformation features, surface texture degradation features and environmental interference correlation features. The multi-dimensional disease feature set is a feature set that comprehensively reflects the pavement disease situation and describes the pavement disease from multiple different dimensions. The structural deformation feature mainly reflects the deformation of the pavement structure under the influence of various factors, such as the settlement, cracks and other structural changes of the pavement. The surface texture degradation feature focuses on the changes in the surface texture of the pavement. With the influence of factors such as time and vehicle driving, the texture of the pavement surface will gradually wear and degrade. The environmental interference correlation feature refers to the features related to environmental factors that affect pavement diseases, such as the correlation between environmental factors such as temperature changes and humidity changes and pavement diseases.

[0013] In edge computing nodes, raw pavement monitoring data is processed and analyzed to generate a multi-dimensional set of disease signatures. For example, by analyzing vibration information collected by vibration sensors, it is possible to determine whether the pavement structure has deformed. Image processing and analysis of pavement images captured by optical sensors can detect degradation of the pavement surface texture. Furthermore, by combining temperature information collected by temperature sensors with humidity information collected by humidity sensors, it is possible to analyze the relationship between environmental factors and pavement disease.

[0014] As an implementation method, step S200 may specifically include the following steps S210~S250: Step S210: Perform spatiotemporal alignment processing on the original road surface monitoring data set to eliminate the timestamp deviation and spatial coordinate offset of data collected by different sensing devices, and generate a standardized monitoring data sequence.

[0015] Spatiotemporal alignment is the process of unifying data collected by different sensing devices in time and space. Due to their unique characteristics and operating modes, different sensing devices may experience timestamp discrepancies and spatial coordinate offsets. Timestamp discrepancies refer to inconsistencies in the time at which data is collected by different devices, potentially resulting in a certain time delay. Spatial coordinate offset refers to errors in the spatial location identifiers corresponding to data collected by different devices. Standardized monitoring data sequences are those that have undergone spatiotemporal alignment and have unified time and spatial identifiers, facilitating subsequent analysis and processing. To eliminate timestamp discrepancies during spatiotemporal alignment, time synchronization algorithms can be used. For example, using the Network Time Protocol (NTP), the clocks of each sensing device are synchronized with a precise time source, ensuring consistent timestamps for their collected data. To eliminate spatial coordinate offsets, coordinate transformation and calibration methods can be used. First, the installation locations of each sensing device are accurately measured and recorded to obtain their initial spatial coordinates. Then, using tools such as geographic information systems (GIS), the spatial coordinates of the data collected by different devices are transformed into a unified coordinate system and calibrated.

[0016] Step S220: Perform multi-scale feature decomposition processing on the standardized monitoring data sequence, extract the pavement structure response characteristics at different scales, and calculate the energy distribution difference coefficient between adjacent scale features. Multi-scale feature decomposition processing is to decompose the standardized monitoring data sequence at different scales to extract the pavement structure response characteristics at different scales. Different scales can be understood as different resolutions or frequency ranges. At different scales, the response characteristics of the pavement structure are different. The pavement structure response characteristics refer to the response characteristics generated by the pavement when it is subjected to external influences, such as vibration response, deformation response, etc. The energy distribution difference coefficient is a coefficient used to measure the energy distribution difference between adjacent scale features. It can reflect the changes in the pavement structure response characteristics at different scales.

[0017] When performing multi-scale feature decomposition, methods such as wavelet transform can be employed. Taking wavelet transform as an example, a multi-scale analysis framework based on wavelet transform is first constructed, with a mother wavelet function and scale decomposition level parameters matching the highway material properties. Then, a joint time-frequency analysis is performed on the standardized monitoring data series, calculating the energy density distribution function and frequency band correlation index at each scale. Based on the preset pavement structure damage-sensitive frequency range, scale levels containing key vibration modes are selected, and the wavelet coefficient matrix of the corresponding scale is extracted. Singular value decomposition is performed on the wavelet coefficient matrix to obtain the principal component eigenvectors reflecting the overall stiffness changes of the pavement structure. By combining the differences between the principal component eigenvectors and historical health status benchmark data, a pavement structure response characteristic is generated that quantitatively describes the degree of structural deformation. The energy distribution difference coefficient between adjacent scale features can be calculated by taking the difference between the energy density distribution functions at adjacent scales.

[0018] As an implementation method, step S220 may specifically include the following steps S221 to S225: Step S221: Construct a multi-scale analysis framework based on wavelet transform, and set a mother wavelet function and scale decomposition level parameters that match the characteristics of highway materials. The multi-scale analysis framework based on wavelet transform is a mathematical model for multi-scale decomposition of data. Wavelet transform is a time-frequency analysis method that can decompose the signal at different scales and time positions to extract the multi-scale characteristics of the signal. The mother wavelet function is a basic function in the wavelet transform. Different mother wavelet functions have different characteristics. Selecting a mother wavelet function that matches the characteristics of highway materials can more effectively extract the response characteristics of the pavement structure. The scale decomposition level parameters determine the number of decomposition layers of the wavelet transform. Different decomposition layers can obtain characteristics at different scales.

[0019] When constructing a multiscale analysis framework based on wavelet transforms, it is necessary to select an appropriate mother wavelet function based on the characteristics of the road material. For example, for road materials with high elasticity, a mother wavelet function with good smoothness and symmetry, such as the Daubechies wavelet function, can be selected. The parameters for the scale decomposition levels must be considered, considering the complexity of the pavement structure's response characteristics and the required accuracy of the analysis. Generally speaking, the appropriate scale decomposition level parameters can be determined through experimentation and experience.

[0020] Step S222: Perform a time-frequency joint analysis on the standardized monitoring data sequence and calculate the energy density distribution function and frequency band correlation index at each scale. Time-frequency joint analysis is an analysis method that simultaneously considers the time and frequency characteristics of the signal. Through time-frequency joint analysis, we can have a more comprehensive understanding of the distribution of the signal at different times and frequencies. The energy density distribution function is a function that describes the energy distribution of the signal at different frequencies. It can reflect the degree of energy concentration of the signal at different frequencies. The frequency band correlation index is an indicator used to measure the correlation between different frequency bands. It can reflect the degree of association between different frequency bands.

[0021] When performing a joint time-frequency analysis on a standardized monitoring data sequence, the coefficients after wavelet transform can be used to calculate the energy density distribution function and frequency band correlation index. The energy density distribution function can be calculated by summing the squares of the wavelet coefficients at each scale and then dividing by the scale factor. The frequency band correlation index can be calculated by calculating the correlation coefficient of the wavelet coefficients between different frequency bands. For example, after performing a wavelet transform on the standardized monitoring data sequence, the square sum of the wavelet coefficients at each scale is calculated and divided by the scale factor to obtain the energy density distribution function at that scale. Simultaneously, the correlation coefficient of the wavelet coefficients between different frequency bands is calculated to obtain the frequency band correlation index.

[0022] Step S223: Based on the preset pavement structure damage sensitive frequency band range, screen out the scale level containing the key vibration mode and extract the wavelet coefficient matrix of the corresponding scale. The preset pavement structure damage sensitive frequency band range is determined based on a large number of experiments and studies. Within this frequency band range, the vibration response of the pavement structure is closely related to the pavement damage condition. The key vibration mode refers to the vibration mode that shows obvious changes when the pavement structure is damaged. The scale level is the different levels after the wavelet transform decomposition, and each scale level corresponds to a different frequency range. The wavelet coefficient matrix is ​​the coefficient matrix obtained after the wavelet transform, which contains the information of the signal at different scales and time positions.

[0023] When selecting scale levels containing key vibration modes, the first step is to determine the pre-defined frequency range sensitive to pavement structural damage. Then, based on the scale-frequency correspondence of the wavelet transform, the scale levels corresponding to this frequency range are identified. Finally, the wavelet coefficient matrices for these scale levels are extracted. For example, experimental research has determined that the frequency range sensitive to pavement structural damage is 10-50 Hz. Based on the scale-frequency correspondence of the wavelet transform, the corresponding scale levels are found to be levels 3-5. The wavelet coefficient matrices for these three levels are extracted for subsequent analysis.

[0024] Step S224: Perform singular value decomposition on the wavelet coefficient matrix to obtain the principal component eigenvector reflecting the change of the overall stiffness of the pavement structure. Singular value decomposition is a matrix decomposition method that can decompose a matrix into the product of three matrices, namely UΣV T , where U and V are orthogonal matrices, Σ is a diagonal matrix with singular values ​​on the diagonal. Principal component eigenvectors are obtained through singular value decomposition and can reflect the main characteristics and changing trends of the matrix. In pavement structure analysis, principal component eigenvectors can reflect the changes in the overall stiffness of the pavement structure.

[0025] When performing singular value decomposition on the wavelet coefficient matrix, a singular value decomposition algorithm, such as a singular value decomposition algorithm based on QR decomposition, can be used. Through singular value decomposition, the singular values ​​and corresponding eigenvectors of the wavelet coefficient matrix are obtained. Eigenvectors with larger singular values ​​are selected as principal component eigenvectors. These principal component eigenvectors can reflect the main changes in the overall stiffness of the pavement structure. For example, singular value decomposition is performed on the extracted wavelet coefficient matrix to obtain singular values ​​and eigenvectors. The first three eigenvectors with larger singular values ​​are selected as principal component eigenvectors. They can reflect the main changes in the overall stiffness of the pavement structure.

[0026] Step S225: Combine the difference between the principal component eigenvector and the historical health status benchmark data to generate a pavement structure response feature that quantitatively describes the degree of structural deformation. The historical health status benchmark data is relevant data collected when the pavement was in a healthy state, and it can be used as a reference standard. The difference refers to the degree of difference between the principal component eigenvector and the historical health status benchmark data, which can be measured by calculating the distance or similarity between the two. The pavement structure response feature is used to quantitatively describe the degree of deformation of the pavement structure, which can intuitively reflect the health status of the pavement structure.

[0027] When generating a pavement structural response characteristic that quantitatively describes the degree of structural deformation, the first step is to calculate the difference between the principal component eigenvector and the historical health status benchmark data. This difference can be calculated using methods such as the Euclidean distance. Then, based on the magnitude of this difference, the corresponding pavement structural response characteristic is generated. For example, a small difference indicates a small degree of deformation in the pavement structure, and the corresponding pavement structural response characteristic value is also small. A large difference indicates a large degree of deformation in the pavement structure, and the corresponding pavement structural response characteristic value is also large. For example, the Euclidean distance between the principal component eigenvector and the historical health status benchmark data is calculated as the difference, and the corresponding pavement structural response characteristic is generated based on the magnitude of this difference. A larger difference indicates a larger pavement structural response characteristic value, indicating a more severe degree of deformation.

[0028] Step S230: Combine the energy distribution difference coefficient with the preset environmental noise suppression model to filter and optimize the intermediate feature set generated by the multi-scale feature decomposition process to generate a denoised primary feature set. The energy distribution difference coefficient is a coefficient that measures the energy distribution difference between adjacent scale features. It can reflect the changes in the response characteristics of the pavement structure at different scales. The preset environmental noise suppression model is established based on a large number of experiments and data, and is used to suppress the impact of environmental noise on feature extraction. The intermediate feature set is a feature set obtained after the multi-scale feature decomposition process, which may contain interference information such as environmental noise. The denoised primary feature set is a feature set that has been filtered and optimized to remove interference information such as environmental noise.

[0029] During filter optimization, the energy distribution difference coefficient is first used as a weighting factor, and the intermediate feature set is processed in conjunction with a preset environmental noise suppression model. This preset environmental noise suppression model can utilize an adaptive filtering algorithm, such as the Least Mean Squared Error (LMS) algorithm. The LMS algorithm adaptively filters the intermediate feature set based on the energy distribution difference coefficient to remove environmental noise. For example, in a highway monitoring area, the intermediate feature set generated by multi-scale feature decomposition contains interference information such as environmental noise. Combining the energy distribution difference coefficient with the LMS environmental noise suppression model, the intermediate feature set is filtered and optimized to remove environmental noise, generating a denoised primary feature set.

[0030] Step S240: Input the primary feature set into the feature fusion network deployed locally at the edge node, and calculate the feature association weights between different sensor data channels through the cross-modal feature attention mechanism. The primary feature set is a denoised feature set obtained after filtering optimization, which contains feature information of different sensor data channels. The feature fusion network deployed locally at the edge node is a neural network used to fuse features of different sensor data channels, which can effectively fuse different types of features. The cross-modal feature attention mechanism is a mechanism for calculating feature association weights between different sensor data channels, which can automatically adjust the feature weights according to the correlation between different features.

[0031] After the primary feature set is input into the feature fusion network, the cross-modal feature attention mechanism first processes the feature vectors of the different sensor data channels in the primary feature set. For example, the vibration sensing feature vectors, optical sensing feature vectors, and temperature sensing feature vectors are dimensionalized to generate a standardized multimodal feature set with the same vector dimensions. A multimodal feature interaction space is then constructed within the feature fusion network, and each feature vector in the standardized multimodal feature set is projected into a shared embedding space to generate an intermediate feature embedding set that is cross-modally comparable. Attention is then calculated on the different feature vectors in the intermediate feature embedding set, generating an attention weight matrix that reflects the correlation between different features. Finally, the feature vectors in the standardized multimodal feature set are weightedly fused according to the attention weight matrix to obtain the fused features.

[0032] As an embodiment, in step S240, the feature association weights between different sensor data channels are calculated through a cross-modal feature attention mechanism, which may specifically include the following steps S241-S247: Step S241: The vibration sensing feature vector, optical sensing feature vector, and temperature sensing feature vector in the primary feature set are dimensionally normalized to generate a standardized multimodal feature set with the same vector dimensions. The vibration sensing feature vector is a feature vector obtained by processing data collected by the vibration sensor and reflects the vibration of the road surface. The optical sensing feature vector is a feature vector obtained by processing image data collected by the optical sensor and reflects the surface texture of the road surface. The temperature sensing feature vector is a feature vector obtained by processing data collected by the temperature sensor and reflects the temperature of the road surface. Dimensional normalization is the process of converting feature vectors of different dimensions into vectors of the same dimension. The standardized multimodal feature set is a multimodal feature set with the same vector dimensions after dimension normalization.

[0033] When performing dimensionality unification, methods such as feature extraction and dimensionality reduction can be used. For example, for vibration sensing feature vectors, optical sensing feature vectors, and temperature sensing feature vectors, the principal component analysis (PCA) algorithm can be used for dimensionality reduction, unifying their dimensions to the same value. First, each eigenvector is normalized to a mean of 0 and a variance of 1. Then, the covariance matrix of each eigenvector is calculated, and the principal components are obtained by solving the eigenvalues ​​and eigenvectors of the covariance matrix. The first N principal components with the largest eigenvalues ​​are selected as the primary features, and the original eigenvectors are projected onto these principal components to obtain the reduced eigenvectors. Finally, the reduced eigenvectors are combined into a standardized multimodal feature set.

[0034] Step S242: Construct a multimodal feature interaction space in the feature fusion network, project each feature vector in the standardized multimodal feature set into a shared embedding space, and generate an intermediate feature embedding set with cross-modal comparability. The multimodal feature interaction space is a space in the feature fusion network used to promote interaction and fusion between features from different modalities. The shared embedding space is a unified space. Projecting feature vectors from different modalities into this space makes them comparable. The intermediate feature embedding set is a feature set that is cross-modally comparable after projection. When constructing the multimodal feature interaction space, a structure such as a fully connected layer can be used. Each feature vector in the standardized multimodal feature set is input into a fully connected layer, and the fully connected layer transforms them to be projected into the shared embedding space. The weights of the fully connected layer can be determined through training so that the projected feature vectors in the shared embedding space better reflect the relationship between features from different modalities. For example, a fully connected layer can be constructed in the feature fusion network, and each feature vector in the standardized multimodal feature set is input into the fully connected layer. The input dimension of the fully connected layer is the dimension of the feature vector, and the output dimension is the dimension of the shared embedding space. By training the weights of the fully connected layer, the feature vector is projected into the shared embedding space, generating an intermediate feature embedding set that is comparable across modalities.

[0035] Step S243: Perform bidirectional attention calculation on the vibration modal embedding vector and the optical texture embedding vector in the intermediate feature embedding set to generate a first attention weight matrix that reflects the correlation between vibration and optical features. The vibration modal embedding vector is the embedding vector in the intermediate feature embedding set that corresponds to the vibration sensing data, which reflects the representation of the road surface vibration condition in the shared embedding space. The optical texture embedding vector is the embedding vector in the intermediate feature embedding set that corresponds to the optical sensing data, which reflects the representation of the road surface texture condition in the shared embedding space. Bidirectional attention calculation is an attention calculation method that simultaneously considers the mutual influence between two vectors. The first attention weight matrix is ​​a matrix used to reflect the correlation between vibration and optical features, and the elements in the matrix represent the degree of correlation between the vibration modal embedding vector and the optical texture embedding vector.

[0036] When performing bidirectional attention calculations, the dot product attention algorithm in the attention mechanism can be used. First, the dot product of the vibration modal embedding vector and the optical texture embedding vector is calculated to obtain a similarity matrix. Then, the similarity matrix is ​​normalized, for example, using a softmax function, to obtain an attention weight matrix. Each element in the attention weight matrix represents the degree of association between the vibration modal embedding vector and the optical texture embedding vector at the corresponding position. For example, the dot product of the vibration modal embedding vector and the optical texture embedding vector in the intermediate feature embedding set is calculated to obtain a similarity matrix. The similarity matrix is ​​normalized using the softmax function to generate a first attention weight matrix that reflects the correlation between the vibration and optical features.

[0037] Step S244: Perform cross-attention calculation on the temperature distribution embedding vector and the vibration modal embedding vector in the intermediate feature embedding set to generate a second attention weight matrix that reflects the temperature-vibration coupling relationship. The temperature distribution embedding vector is the embedding vector corresponding to the temperature sensor data in the intermediate feature embedding set. It reflects the representation of the road surface temperature in the shared embedding space. Cross-attention calculation is a method for calculating the attention weight between two different modal vectors. The second attention weight matrix is ​​a matrix used to reflect the temperature-vibration coupling relationship. The elements in the matrix represent the degree of coupling between the temperature distribution embedding vector and the vibration modal embedding vector.

[0038] The dot product attention algorithm can also be used when performing cross-attention calculations. First, calculate the dot product of the temperature distribution embedding vector and the vibration mode embedding vector to obtain a similarity matrix. Then, the similarity matrix is ​​normalized, for example, using a softmax function, to obtain a second attention weight matrix. Each element in the second attention weight matrix represents the degree of coupling between the temperature distribution embedding vector and the vibration mode embedding vector at the corresponding position. For example, perform a dot product calculation on the temperature distribution embedding vector and the vibration mode embedding vector in the intermediate feature embedding set to obtain a similarity matrix. Use the softmax function to normalize the similarity matrix to generate a second attention weight matrix that reflects the temperature-vibration coupling relationship.

[0039] Step S245: Construct a multimodal feature fusion coefficient tensor based on the first attention weight matrix and the second attention weight matrix, and perform a tensor convolution operation on each feature vector in the standardized multimodal feature set. The multimodal feature fusion coefficient tensor is a tensor constructed based on the first attention weight matrix and the second attention weight matrix. It is used to represent the fusion coefficient between different modal features. The tensor convolution operation is an operation that performs a convolution operation on a tensor. It can convolve the multimodal feature fusion coefficient tensor with each feature vector in the standardized multimodal feature set to achieve the fusion of different modal features.

[0040] When constructing the multimodal feature fusion coefficient tensor, the first attention weight matrix and the second attention weight matrix can be combined. For example, they can be concatenated in a certain dimension to obtain a three-dimensional multimodal feature fusion coefficient tensor. Then, a tensor convolution operation is performed on each feature vector in the standardized multimodal feature set. This tensor convolution operation can be implemented using a convolution layer in a convolutional neural network (CNN). The multimodal feature fusion coefficient tensor is used as the convolution kernel to convolve each feature vector in the standardized multimodal feature set to obtain a fused feature vector. For example, the first attention weight matrix and the second attention weight matrix are concatenated in the third dimension to obtain a three-dimensional multimodal feature fusion coefficient tensor. A convolution layer is used to perform a tensor convolution operation on each feature vector in the standardized multimodal feature set, using the multimodal feature fusion coefficient tensor as the convolution kernel to obtain a fused feature vector.

[0041] Step S246: The convolved multimodal features are selectively enhanced through a dynamic gating mechanism, retaining cross-modal correlation features that are strongly correlated with pavement defects and suppressing weakly correlated feature components caused by environmental noise. The dynamic gating mechanism selectively enhances or suppresses features based on their importance. The convolved multimodal features are fused features obtained after tensor convolution. Cross-modal correlation features that are strongly correlated with pavement defects are correlation features between different modal features that can directly reflect the pavement defect situation. Weakly correlated feature components caused by environmental noise are feature components that have a weak relationship with pavement defects due to factors such as environmental noise.

[0042] When performing selective enhancement through dynamic gating, structures such as the gated recurrent unit (GRU) can be used. The convolved multimodal features are fed into the GRU, which automatically generates a gating signal based on feature importance. This gating signal controls which features are enhanced and which are suppressed. For example, the gating signal gives greater weight to cross-modal features strongly associated with pavement damage, thereby enhancing them; while the gating signal gives less weight to weakly correlated features caused by environmental noise, thereby suppressing them.

[0043] Step S247: The enhanced cross-modal correlation features are hierarchically spliced ​​with the original unimodal features to generate a fused disease feature set containing complementary information from multiple sources of data. The enhanced cross-modal correlation features are features obtained after selective enhancement by a dynamic gating mechanism, and they contain strong correlation information between features from different modalities. The original unimodal features are features from a single modality that have not been fused in the primary feature set. Hierarchical splicing is the operation of splicing the enhanced cross-modal correlation features and the original unimodal features in a certain dimension. The fused disease feature set is a feature set that contains complementary information from multiple sources of data after hierarchical splicing, and it can more comprehensively reflect the pavement disease situation. When performing hierarchical splicing, the enhanced cross-modal correlation features and the original unimodal features can be spliced ​​in the feature dimension. Through hierarchical splicing, the correlation information between features from different modalities and the feature information of a single modality can be integrated to generate a fused disease feature set containing complementary information from multiple sources of data.

[0044] Step S250: performing weighted fusion processing on the primary feature set according to the feature association weights to generate a multi-dimensional disease feature set including a structural deformation quantitative index, a surface texture degradation map, and an environmental interference association matrix.

[0045] Feature association weights are calculated using a cross-modal feature attention mechanism and reflect the degree of feature association between different sensor data channels. Weighted fusion involves weighted summation of the different features in the primary feature set based on the feature association weights. Structural deformation quantitative indicators are used to quantify the degree of pavement structural deformation and can be calculated based on information such as the pavement structural response characteristics. Surface texture degradation maps are used to describe the degradation of pavement surface textures and can be generated using information such as optical sensor data. The environmental interference correlation matrix is ​​used to describe the correlation between environmental factors and pavement defects and can be calculated based on environmental sensor data such as temperature and humidity, as well as pavement defect characteristics. During weighted fusion, each feature vector in the primary feature set is multiplied by the corresponding feature association weight and then summed to produce a weighted fused feature vector. Based on these weighted fused feature vectors, the structural deformation quantitative indicators, surface texture degradation maps, and environmental interference correlation matrix are calculated. For example, the vibration sensing feature vector, optical sensing feature vector, and temperature sensing feature vector in the primary feature set are multiplied by their corresponding feature association weights and then added together to obtain a weighted fused feature vector. Based on this weighted fused feature vector, combined with the pavement structure response characteristics, a structural deformation quantitative index is calculated. A surface texture degradation map is generated using optical sensing data. An environmental interference correlation matrix is ​​calculated based on environmental sensing data such as temperature and humidity and pavement disease characteristics. Ultimately, a multidimensional disease feature set is generated, including the structural deformation quantitative index, the surface texture degradation map, and the environmental interference correlation matrix.

[0046] Step S300: Upload the multi-dimensional disease feature set to the cloud analysis platform, and use the cloud analysis platform to calculate the feature matching degree between the multi-dimensional disease feature set and the standard disease pattern in the cloud disease feature library to generate the target disease type identification result and the corresponding confidence assessment parameter.

[0047] The multi-dimensional defect feature set is obtained after processing by the edge computing node and contains a feature set of information on various aspects of pavement defects. The cloud analysis platform is a platform with powerful computing and storage capabilities, used to analyze and process uploaded data. The cloud defect feature library is a database that stores various standard defect pattern features. These standard defect pattern features are obtained through a large number of experiments and research. The feature matching calculation is to compare the multi-dimensional defect feature set with the standard defect patterns in the cloud defect feature library and calculate the degree of similarity between them. The target defect type identification result is the pavement defect type determined based on the feature matching calculation result. The confidence assessment parameter is a parameter used to evaluate the credibility of the target defect type identification result.

[0048] After uploading the multidimensional disease feature set to the cloud analysis platform, the cloud analysis platform first preprocesses the multidimensional disease feature set, such as normalizing it, so that it has the same scale as the standard disease pattern in the cloud disease feature library. Then, each feature subset in the multidimensional disease feature set is compared layer by layer with the standard disease pattern of the corresponding dimension, and the feature similarity measurement parameters and the abnormal feature deviation index are calculated. Based on the abnormal feature deviation index, a dynamic weight adjustment function is constructed, and the feature similarity measurement parameters are adaptively weighted to generate a comprehensive matching score set. According to the standard disease pattern identifier corresponding to the highest score in the comprehensive matching score set, the initial disease type recognition result is generated. Finally, the initial disease type recognition result is confidence verified, including calculating the difference ratio between adjacent scores and the stability parameter of the historical recognition results, and generating a two-dimensional confidence evaluation parameter including the credibility level and the error tolerance range.

[0049] As an implementation method, in step S300, the cloud analysis platform calculates the feature matching degree between the multi-dimensional disease feature set and the standard disease pattern in the cloud disease feature library to generate the target disease type identification result and the corresponding confidence assessment parameter. Specifically, the following steps S310 to S350 may be included: Step S310: Deconstruct the feature dimension of the standard disease pattern in the cloud disease feature library, and extract the feature weight distribution vector and key feature constraint conditions corresponding to each standard disease pattern. The standard disease pattern in the cloud disease feature library is a pre-defined feature pattern of various pavement diseases, and each standard disease pattern contains multiple feature dimensions. Feature dimension deconstruction is the process of decomposing the standard disease pattern according to different feature dimensions. The feature weight distribution vector is a vector used to represent the importance of each feature dimension in the standard disease pattern, and each element in the vector corresponds to the weight of a feature dimension. The key feature constraint conditions are restrictions on the value range or relationship of certain key features in the standard disease pattern.

[0050] When deconstructing feature dimensions, the feature data of standard damage patterns can be separated according to different feature dimensions. For example, if a standard damage pattern includes feature dimensions such as structural deformation quantitative indicators, surface texture degradation maps, and environmental interference correlation matrices, these can be extracted separately. To extract feature weight distribution vectors, machine learning algorithms, such as support vector machines (SVMs), can be used to train large amounts of standard damage pattern data to obtain weights for each feature dimension. Key feature constraints can be determined based on actual damage conditions and expert experience. For example, for a certain pavement damage, it is stipulated that the structural deformation quantitative indicators must be within a certain range, and certain feature values ​​of the surface texture degradation map must satisfy preset relationships.

[0051] Step S320: Compare each feature subset in the multi-dimensional defect feature set with the standard defect pattern of the corresponding dimension layer by layer, and calculate the feature similarity measurement parameter and the abnormal feature deviation index. Each feature subset in the multi-dimensional defect feature set corresponds to a feature dimension, such as a structural deformation quantitative index subset, a surface texture degradation map subset, and an environmental interference correlation matrix subset. The standard defect pattern of the corresponding dimension is a standard defect pattern in the cloud defect feature library with the same feature dimension as each feature subset. The feature similarity measurement parameter is a parameter used to measure the degree of similarity between the feature subset and the standard defect pattern of the corresponding dimension. The abnormal feature deviation index is an index used to measure the degree of deviation between the abnormal features in the feature subset and the normal features in the standard defect pattern.

[0052] When performing layer-by-layer comparison, for each feature subset, it is compared with the standard disease pattern of the corresponding dimension. For the calculation of feature similarity measurement parameters, methods such as Euclidean distance and cosine similarity can be used. For example, for the structural deformation quantitative indicator subset and the structural deformation quantitative indicator of the standard disease pattern of the corresponding dimension, the Euclidean distance between them can be calculated. The smaller the distance, the higher the similarity. For the calculation of the abnormal feature deviation index, the value range of the normal features in the standard disease pattern can be determined first, and then the degree of deviation of the abnormal features in the feature subset from the value range is calculated. For example, the structural deformation quantitative indicator subset in the multi-dimensional disease feature set is compared with the structural deformation quantitative indicator of the standard disease pattern of the corresponding dimension, and the feature similarity measurement parameters between them are calculated using Euclidean distance. At the same time, the normal value range of the structural deformation quantitative indicator in the standard disease pattern is determined, and the degree of deviation of the abnormal features in the feature subset from the value range is calculated to obtain the abnormal feature deviation index.

[0053] Step S330: A dynamic weight adjustment function is constructed based on the abnormal feature deviation index, and feature similarity measurement parameters are adaptively weighted to generate a set of comprehensive matching scores. The dynamic weight adjustment function dynamically adjusts feature weights based on the abnormal feature deviation index. Adaptive weighting is the process of weighting feature similarity measurement parameters based on the dynamic weight adjustment function. The set of comprehensive matching scores is the set of comprehensive matching scores between each standard disease pattern and the multi-dimensional disease feature set, obtained after adaptive weighting.

[0054] When constructing a dynamic weight adjustment function, linear functions, exponential functions, and other forms can be used. For example, a linear dynamic weight adjustment function can be constructed: w = 1 - k * d, where w is the adjusted weight, d is the abnormal feature deviation index, and k is a preset constant. Based on the dynamic weight adjustment function, the feature similarity measurement parameters of each feature dimension are weighted, and then the weighted feature similarity measurement parameters are added together to obtain a comprehensive matching score for each standard disease pattern and the multi-dimensional disease feature set. The comprehensive matching scores of all standard disease patterns are combined to form a comprehensive matching score set.

[0055] Step S340: Generate an initial disease type identification result based on the standard disease pattern identifier corresponding to the highest score in the comprehensive matching score set. Each score in the comprehensive matching score set corresponds to a standard disease pattern, and the highest score indicates that the multi-dimensional disease feature set has the highest degree of match with the standard disease pattern. The standard disease pattern identifier is a number or name used to uniquely identify each standard disease pattern. The initial disease type identification result is the pavement disease type determined based on the standard disease pattern identifier corresponding to the highest score. When generating the initial disease type identification result, first find the highest score in the comprehensive matching score set, and then obtain the standard disease pattern identifier corresponding to the highest score. Determine the corresponding pavement disease type based on the standard disease pattern identifier.

[0056] Step S350: Confidence verification of the initial disease type identification result is performed, including calculating the difference ratio between adjacent scores and the stability parameter of the historical identification results, and generating a two-dimensional confidence evaluation parameter including a confidence level and an error tolerance range. Confidence verification is the process of evaluating the credibility of the initial disease type identification result. The difference ratio between adjacent scores refers to the ratio of the difference between the highest score and the second highest score in the comprehensive matching score set to the highest score, which can reflect the degree of dominance of the highest score. The stability parameter of the historical identification result is a parameter determined based on the stability of the identification results under the same or similar circumstances in history, which can reflect the reliability of the identification result. The confidence level is a level used to indicate the credibility of the initial disease type identification result, such as high, medium, low, etc. The error tolerance range refers to the error range allowed for the identification result under a certain credibility.

[0057] When performing confidence verification, first calculate the difference ratio between adjacent scores. For example, if the highest score is 80 points and the second highest score is 60 points, the difference ratio is (80-60) / 80=0.25. Then, based on the stability parameters of the historical recognition results, such as the accuracy and recall rate of the historical recognition results, the difference ratio is comprehensively considered to determine the credibility level. Based on the credibility level, the error tolerance range is determined. For example, if the difference ratio is large and the stability parameters of the historical recognition results are high, the credibility level is high and the error tolerance range is small; conversely, if the difference ratio is small and the stability parameters of the historical recognition results are low, the credibility level is low and the error tolerance range is large. Finally, a two-dimensional confidence evaluation parameter containing the credibility level and the error tolerance range is generated.

[0058] Step S400: Based on the relationship between the confidence assessment parameter and the preset threshold, the edge computing node's feature extraction strategy is adjusted, and an edge node adaptive optimization instruction set is generated. The confidence assessment parameter is a two-dimensional parameter consisting of a confidence level and an error tolerance range. It is used to assess the confidence level of the target disease type identification result. The preset threshold is a pre-set threshold used to determine whether the confidence assessment parameter meets the requirements, including the confidence level threshold and the error tolerance threshold. The edge computing node's feature extraction strategy is the method and rules used by the edge computing node when collecting and processing data. The edge node adaptive optimization instruction set is a set of instructions generated based on the relationship between the confidence assessment parameter and the preset threshold for adjusting the edge computing node's feature extraction strategy. When adjusting the edge computing node's feature extraction strategy based on the relationship between the confidence assessment parameter and the preset threshold, if the confidence level falls below the confidence level threshold, it indicates that the current feature extraction strategy may not be accurately extracting useful features and requires optimization. For example, the feature space expansion mode can be activated to add high-frequency vibration signal acquisition channels to the edge computing node and expand the number of feature mapping layer nodes in the feature fusion network to obtain more feature information. If the error tolerance exceeds the error tolerance threshold, it indicates that the current feature extraction strategy may have large errors and requires verification. For example, a multi-source data verification mechanism is triggered, and a cross-validation process for LiDAR point cloud data and visible light image data is introduced into the edge computing node to improve the accuracy of feature extraction. At the same time, a strategy optimization trend prediction model is generated based on the historical feature matching calculation results, and the data sampling frequency of the edge computing node and the computational complexity parameters of the feature extraction algorithm are adjusted. Finally, the feature space expansion mode, multi-source data verification mechanism, and computational complexity adjustment parameters are integrated into a configurable instruction code block to generate an adaptive optimization instruction set that includes execution condition judgment logic and parameter update rules.

[0059] As an implementation method, step S400 may specifically include the following steps S410 to S450: Step S410: Determine the optimization priority and parameter adjustment range of the current feature extraction strategy based on the credibility level division result in the confidence assessment parameter. The credibility level division result in the confidence assessment parameter is to divide the credibility level into different levels, such as high, medium, low, etc. The optimization priority refers to the priority of different optimization measures when adjusting the feature extraction strategy of the edge computing node. The parameter adjustment range refers to the range and degree of parameter adjustment when adjusting the parameters of the feature extraction strategy.

[0060] When determining optimization priorities and parameter adjustment ranges based on the confidence level, a low confidence level indicates significant issues with the current feature extraction strategy and requires prioritization. For example, prioritizing the activation of the feature space expansion mode, increasing the number of high-frequency vibration signal acquisition channels, and expanding the number of feature mapping layer nodes in the feature fusion network can be beneficial. Furthermore, parameter adjustments can be made more significantly to rapidly improve feature extraction accuracy. A medium confidence level indicates issues with the current feature extraction strategy and requires appropriate optimization. For example, the data sampling frequency and the computational complexity parameters of the feature extraction algorithm can be appropriately adjusted. Parameter adjustments can be moderate. A high confidence level indicates that the current feature extraction strategy is relatively effective and can be maintained or slightly optimized.

[0061] Step S420: When the credibility level is lower than the first preset threshold, the feature space expansion mode is activated, a high-frequency vibration signal acquisition channel is added to the edge computing node, and the number of feature mapping layer nodes of the feature fusion network is expanded. The first preset threshold is a pre-set threshold for judging whether the credibility level requires activation of the feature space expansion mode. The feature space expansion mode is a mode for expanding the feature extraction capability of the edge computing node. The high-frequency vibration signal acquisition channel is a channel for collecting high-frequency vibration signals of the road surface. Increasing the high-frequency vibration signal acquisition channel can obtain more road vibration information. The number of feature mapping layer nodes of the feature fusion network is the number of nodes of the feature mapping layer in the feature fusion network. Expanding the number of feature mapping layer nodes can improve the feature fusion capability.

[0062] When the confidence level falls below the first preset threshold, the current feature extraction strategy fails to meet the requirements and the feature space expansion mode needs to be activated. Adding high-frequency vibration signal acquisition channels to the edge computing node can be achieved by installing more high-frequency vibration sensors. For example, in addition to the existing one high-frequency vibration sensor, two more high-frequency vibration sensors can be installed to increase the number of high-frequency vibration signal acquisition channels. Expanding the number of feature mapping layer nodes in the feature fusion network can be achieved by modifying the structure of the feature fusion network.

[0063] Step S430: When the error tolerance range exceeds the second preset threshold, the multi-source data verification mechanism is triggered, and a cross-validation processing flow of the lidar point cloud data and the visible light image data is introduced in the edge computing node. The second preset threshold is a pre-set threshold for judging whether the error tolerance range needs to trigger the multi-source data verification mechanism. The multi-source data verification mechanism is a mechanism that uses data from multiple different sources for mutual verification. The lidar point cloud data is three-dimensional point cloud data about the road surface collected by the lidar device, which can provide three-dimensional structural information of the road surface. The visible light image data is road surface image data collected by the visible light camera, which can provide information such as the surface texture and color of the road surface. The cross-validation processing flow is a process of comparing and verifying the lidar point cloud data with the visible light image data.

[0064] When the error tolerance range exceeds the second preset threshold, it means that the current feature extraction strategy has a large error, and the multi-source data verification mechanism needs to be triggered. Introducing the cross-validation processing flow of lidar point cloud data and visible light image data in the edge computing node, first, it is necessary to install the lidar equipment and visible light camera to collect lidar point cloud data and visible light image data. Then, the collected lidar point cloud data and visible light image data are preprocessed, such as alignment and filtering. Next, the preprocessed lidar point cloud data and visible light image data are compared and verified to check their consistency. If inconsistencies are found, further analysis and processing are required to improve the accuracy of feature extraction.

[0065] Step S440: Generate a strategy optimization trend prediction model based on the historical feature matching calculation results, and adjust the edge computing node's data sampling frequency and the computational complexity parameters of the feature extraction algorithm. The historical feature matching calculation results are the result of calculating the feature matching between the multi-dimensional disease feature set and the standard disease patterns in the cloud-based disease feature library during past monitoring processes. The strategy optimization trend prediction model is built based on the historical feature matching calculation results to predict the optimization trend of the feature extraction strategy. The data sampling frequency is the frequency at which the edge computing node collects data. Adjusting the data sampling frequency can change the amount of collected data and its timeliness. The computational complexity parameters of the feature extraction algorithm are used to control the computational complexity of the feature extraction algorithm. Adjusting the computational complexity parameters can improve the computational efficiency and accuracy of the feature extraction algorithm. When generating the strategy optimization trend prediction model based on the historical feature matching calculation results, time series analysis methods such as the autoregressive integrated moving average (ARIMA) model can be used. By analyzing and modeling the historical feature matching calculation results, future optimization trends of the feature extraction strategy can be predicted. Based on the prediction results of the strategy-optimized trend prediction model, the data sampling frequency of the edge computing node and the computational complexity parameters of the feature extraction algorithm are adjusted. For example, if the prediction results indicate that more feature information will be needed in the future, the data sampling frequency can be increased; if the prediction results indicate that the accuracy of feature extraction needs to be improved, the computational complexity parameters of the feature extraction algorithm can be increased.

[0066] Step S450: Integrate the feature space expansion mode, multi-source data verification mechanism, and computational complexity adjustment parameters into a configurable instruction code block to generate an adaptive optimization instruction set containing execution condition judgment logic and parameter update rules. The feature space expansion mode is used to expand the feature extraction capabilities of edge computing nodes, including increasing the high-frequency vibration signal acquisition channel and expanding the number of feature mapping layer nodes in the feature fusion network. The multi-source data verification mechanism utilizes data from multiple sources for mutual verification, including introducing a cross-validation process between lidar point cloud data and visible light image data. The computational complexity adjustment parameters are used to adjust the computational complexity of the feature extraction algorithm. The configurable instruction code block integrates the feature space expansion mode, multi-source data verification mechanism, and computational complexity adjustment parameters into a single configurable code block for easy execution by edge computing nodes. The execution condition judgment logic is used to determine when to execute the adaptive optimization instruction set, for example, when the confidence level falls below a first preset threshold or when the error tolerance exceeds a second preset threshold. Parameter update rules are used to update parameters within the adaptive optimization instruction set. For example, they update the data sampling frequency and computational complexity parameters of the feature extraction algorithm based on the prediction results of the strategy optimization trend prediction model. When integrating the feature space expansion mode, multi-source data verification mechanism, and computational complexity adjustment parameters into a configurable instruction code block, these elements are first written into code. Then, the execution conditional judgment logic and parameter update rules are added.

[0067] Step S500: Synchronize the target disease type identification result and the edge node adaptive optimization instruction set to the edge computing node, triggering the feature extraction rule update and disease identification model parameter iteration operation for the subsequent monitoring cycle. The target disease type identification result is the pavement disease type obtained by calculating the feature matching degree between the multi-dimensional disease feature set and the standard disease pattern in the cloud disease feature library through the cloud analysis platform. The edge node adaptive optimization instruction set is generated based on the relationship between the confidence assessment parameter and the preset threshold, and is an instruction set for adjusting the feature extraction strategy of the edge computing node. Feature extraction rule update refers to updating the feature extraction rules of the edge computing node according to the edge node adaptive optimization instruction set to improve the accuracy and efficiency of feature extraction. The disease identification model parameter iteration operation refers to iteratively updating the parameters of the disease identification model of the edge computing node according to the edge node adaptive optimization instruction set to improve the accuracy of disease identification.

[0068] When synchronizing the target disease type identification results and the edge node adaptive optimization instruction set to the edge computing node, the data can be transmitted to the edge computing node via a network communication protocol, such as the HTTP protocol. After receiving the data, the edge computing node first updates the feature extraction rules according to the edge node adaptive optimization instruction set. For example, if the instruction set requires the addition of a high-frequency vibration signal acquisition channel, the edge computing node will install the corresponding sensor and modify the feature extraction program to collect high-frequency vibration signals. The edge computing node then iteratively updates the parameters of the disease identification model according to the edge node adaptive optimization instruction set. For example, if the instruction set requires adjusting the computational complexity parameters of the feature extraction algorithm, the edge computing node will modify the corresponding parameters in the disease identification model. By updating the feature extraction rules and iterating the disease identification model parameters, the edge computing node's disease identification capability in subsequent monitoring cycles can be improved.

[0069] As an implementation method, in step S500, the feature extraction rule update and disease identification model parameter iteration operation for the subsequent monitoring cycle are triggered, which may specifically include the following steps S510~S550: Step S510: Establish a feature extraction rule version control library in the edge computing node to record the feature dimension change information and algorithm performance indicators after each strategy adjustment. The feature extraction rule version control library is a database for managing the feature extraction rule version of the edge computing node. It can record the changes in the feature extraction rules after each strategy adjustment. Feature dimension change information refers to information such as the increase, decrease or modification of feature dimensions in the feature extraction rules after the strategy adjustment. Algorithm performance indicators are indicators used to measure the performance of feature extraction algorithms, such as accuracy, recall rate, F1 value, etc.

[0070] When establishing a version control repository for feature extraction rules in edge computing nodes, you can use a database management system, such as MySQL. After each policy adjustment, record feature dimension changes and algorithm performance metrics in the version control repository. For example, if a high-frequency vibration signal acquisition channel is added to the feature space expansion mode, the addition of high-frequency vibration signals to the feature dimension will be recorded. Simultaneously, performance metrics such as accuracy, recall, and F1 score of the adjusted feature extraction algorithm are calculated and recorded in the version control repository. Establishing a version control repository for feature extraction rules facilitates the management and backtracking of feature extraction rules, allowing for visibility into the effectiveness of each policy adjustment.

[0071] Step S520: According to the execution condition judgment logic in the adaptive optimization instruction set, the current feature extraction rules are incrementally updated, and the feature channel configuration associated with the historical high-confidence recognition results is retained. The execution condition judgment logic in the adaptive optimization instruction set is the logic used to determine when to execute the adaptive optimization instruction set, such as when the credibility level is lower than the first preset threshold or the error tolerance range exceeds the second preset threshold. Incremental update refers to a partial update based on the current feature extraction rules according to the adaptive optimization instruction set, rather than a complete replacement. The feature channel configuration associated with the historical high-confidence recognition results refers to the configuration information of the feature channel that can obtain high-confidence recognition results during the historical monitoring process.

[0072] When incrementally updating the current feature extraction rules based on the execution condition judgment logic in the adaptive optimization instruction set, the system first checks whether the execution condition is met. If so, the current feature extraction rules are updated according to the specific instructions in the adaptive optimization instruction set. For example, if the instructions require adding a high-frequency vibration signal acquisition channel, the corresponding acquisition procedure is added to the current feature extraction rules. At the same time, the feature channel configuration associated with historical high-confidence recognition results is retained to ensure that valid feature information is not lost when updating the feature extraction rules.

[0073] Step S530: Construct a fault identification model parameter optimization queue on the cloud-based analysis platform and generate a global model error distribution heat map based on the confidence assessment parameters uploaded by multiple edge computing nodes. The fault identification model parameter optimization queue is used to manage fault identification model parameter optimization tasks. It can sort and schedule parameter optimization tasks for multiple edge computing nodes. The global model error distribution heat map is a visual chart that displays the spatial distribution of fault identification model errors across multiple edge computing nodes.

[0074] When building a fault identification model parameter optimization queue in the cloud-based analysis platform, you can use a queue data structure, such as the Queue module in Python. Add the fault identification model parameter optimization tasks for each edge computing node to the queue and sort and schedule them based on their priority. To generate a global model error distribution heatmap based on the confidence assessment parameters uploaded by multiple edge computing nodes, first calculate the fault identification model error for each edge computing node based on the confidence assessment parameters. Then, map the location information and error value of each edge computing node onto a map, and use a heatmap visualization tool, such as the imshow function in Matplotlib, to generate a global model error distribution heatmap.

[0075] Step S540: Based on the regional outlier nodes identified in the global model error distribution heatmap, model parameter compensation instructions are issued to the corresponding edge computing nodes to adjust the classification boundary conditions and feature weight distribution ratios of the local disease recognition model. Regional outlier nodes are edge computing nodes whose cumulative error values ​​exceed the statistical mean of adjacent regions and meet a preset persistence period in the global model error distribution heatmap. Model parameter compensation instructions are used to adjust the parameters of the local disease recognition model at the edge computing node. They can adjust the classification boundary conditions and feature weight distribution ratios of the local disease recognition model based on the specific conditions of the regional outlier nodes. Classification boundary conditions refer to the boundary conditions used to distinguish different disease types in the disease recognition model. Adjusting these classification boundary conditions can improve the model's recognition accuracy for different disease types. Feature weight distribution ratios refer to the weight ratios assigned to different features in the disease recognition model. Adjusting these weight ratios can change the model's emphasis on different features.

[0076] When identifying regional outlier nodes based on the global model error distribution heatmap, the first step is to determine the statistical mean of the adjacent regions and the preset duration period. The statistical mean of the adjacent regions can be calculated by calculating the mean error of the regions adjacent to the node. The preset duration period can be set based on actual conditions, for example, three consecutive monitoring cycles. When the cumulative error value of an edge computing node exceeds the statistical mean of the adjacent regions and meets the preset duration period, it is identified as a regional outlier node. Then, corresponding model parameter compensation instructions are generated for each regional outlier node. The process of generating model parameter compensation instructions is relatively complex, and the specific steps are detailed below.

[0077] As an implementation method, step S540 may specifically include the following steps S541 to S547: Step S541: Analyze the spatial anomaly distribution pattern of the global model error distribution heat map, identify the target abnormal node set whose error accumulation value exceeds the statistical mean of the adjacent regions and meets the preset duration period, and extract the multi-dimensional environment-related feature vector corresponding to the target abnormal node set. The spatial anomaly distribution pattern of the global model error distribution heat map refers to the spatial characteristics of the error distribution in the heat map, such as whether there are local high-error areas, whether the error distribution has directionality, etc. The target abnormal node set refers to the set of edge computing nodes whose error accumulation value exceeds the statistical mean of the adjacent regions and meets the preset duration period. The multi-dimensional environment-related feature vector refers to the feature vector of multiple environmental factors corresponding to the target abnormal node set, such as the characteristics of environmental factors such as temperature, humidity, and light intensity.

[0078] Image processing and data analysis techniques can be used to analyze spatial anomaly patterns in the global model error distribution heatmap. For example, an image segmentation algorithm can be used to segment the heatmap into distinct regions, and then the error distribution in each region can be analyzed. By calculating the cumulative error value for each edge computing node and comparing it with the statistical mean of the adjacent regions, and checking for compliance with a preset persistence period, a target set of anomaly nodes can be identified. For each target anomaly node set, a corresponding multi-dimensional environmental correlation feature vector is extracted. By querying the environmental sensor data of the edge computing nodes, values ​​for environmental factors such as temperature, humidity, and light intensity can be obtained, and these values ​​are combined into a multi-dimensional environmental correlation feature vector. For example, in a highway monitoring project, the K-means clustering algorithm was used to segment the global model error distribution heatmap into high-error, medium-error, and low-error regions. By calculating the cumulative error value for each edge computing node, it was found that five nodes had cumulative error values ​​exceeding the statistical mean of the adjacent regions for four consecutive monitoring periods. These five nodes were identified as the target anomaly node set. The environmental sensor data of these five nodes are queried to obtain the values ​​of environmental factors such as temperature, humidity, and light intensity, and then combined into a five-dimensional multi-dimensional environmental correlation feature vector.

[0079] Step S542: Perform feature space mapping on the multi-dimensional environment-related feature vectors and the regional environment adaptation templates in the cloud-based disease feature library, and select a reference disease pattern set that matches the pavement material properties, climate conditions, and traffic load characteristics of the target abnormal node set. The regional environment adaptation template is a pre-defined template in the cloud-based disease feature library that matches different regional environmental conditions. It contains disease pattern characteristics under different pavement material properties, climate conditions, and traffic load characteristics. Feature space mapping is the process of matching and mapping the multi-dimensional environment-related feature vectors and the regional environment adaptation templates in the feature space. The reference disease pattern set is a set of disease patterns that match the pavement material properties, climate conditions, and traffic load characteristics of the target abnormal node set.

[0080] When performing feature space mapping, similarity calculation methods, such as cosine similarity, can be used. The cosine similarity between the multi-dimensional environmental association feature vector and each regional environmental adaptation template is calculated. A higher similarity indicates a better match. Regional environmental adaptation templates with similarity greater than a preset threshold are selected. Corresponding reference disease patterns are extracted from these templates to form a reference disease pattern set.

[0081] Step S543: Perform feature weight decoupling analysis on each damage pattern in the reference damage pattern set to separate the core feature weight set directly related to pavement structure damage and the interference feature weight set coupled with environmental noise. Feature weight decoupling analysis is the process of decomposing and analyzing the feature weights of each damage pattern in the reference damage pattern set, with the aim of separating the core feature weight set directly related to pavement structure damage and the interference feature weight set coupled with environmental noise. The core feature weight set refers to the weight set of features in the damage pattern that are directly related to pavement structure damage. These features can directly reflect the condition of the pavement. The interference feature weight set refers to the weight set of features coupled with environmental noise. These features may be affected by environmental factors and interfere with the identification of pavement damage.

[0082] When performing feature weight decoupling analysis, methods such as principal component analysis (PCA) and factor analysis can be used. First, the feature weights of each damage pattern in the reference damage pattern set are standardized to the same scale. Then, PCA is used to reduce the dimensionality of the standardized feature weights and extract the principal components. Based on the principal component contribution rate and load matrix, it is determined which features are directly related to pavement structural damage and which are coupled to ambient noise. The weights of features directly related to pavement structural damage are combined into a core feature weight set, while the weights of features coupled to ambient noise are combined into an interference feature weight set.

[0083] Step S544: A parameter compensation benchmark model is constructed based on the core feature weight set. The orthogonal projection residual matrix is ​​calculated between the local fault identification model parameters of each edge computing node in the target abnormal node set and the benchmark model parameters in the feature space. The parameter compensation benchmark model is constructed based on the core feature weight set and is used to compensate for the local fault identification model parameters of edge computing nodes. The orthogonal projection residual matrix is ​​used to measure the difference in feature space between the local fault identification model parameters of each edge computing node in the target abnormal node set and the benchmark model parameters.

[0084] When constructing a parameter compensation benchmark model based on the core feature weight set, a linear regression model can be used. The core feature weight set is used as the input feature and trained using a large amount of historical data to obtain a parameter compensation benchmark model. For the local disease recognition model parameters of each edge computing node in the target abnormal node set, they are projected into the feature space where the benchmark model parameters are located, and the difference between the projected vector and the benchmark model parameter vector is calculated to form an orthogonal projection residual matrix. For example, using the core feature weight set as the input feature and using 1000 historical data to train a linear regression model, a parameter compensation benchmark model is obtained. For the local disease recognition model parameters of an edge computing node in the target abnormal node set, they are projected into the feature space where the benchmark model parameters are located, and the difference between the projected vector and the benchmark model parameter vector is calculated. This difference is used as a row of the orthogonal projection residual matrix.

[0085] Step S545: Generate a spatially constrained model parameter compensation matrix based on the orthogonal projection residual matrix. A matrix dimension alignment algorithm is used to eliminate the tensor structure differences between the compensation matrix and the local model parameters of each edge computing node. The spatially constrained model parameter compensation matrix is ​​generated based on the orthogonal projection residual matrix, taking spatial factors into account. This matrix is ​​used to compensate for the local defect identification model parameters of the edge computing nodes. The matrix dimension alignment algorithm is used to eliminate the tensor structure differences between the compensation matrix and the local model parameters of each edge computing node.

[0086] When generating a spatially constrained model parameter compensation matrix, spatial factors must first be considered. Based on the spatial distribution of the target abnormal node set, a spatial weight can be assigned to each node. The orthogonal projection residual matrix is ​​then multiplied by the corresponding spatial weight to obtain a spatially constrained model parameter compensation matrix. Differences in the tensor structure between the compensation matrix and the local model parameters of each edge computing node can be eliminated using a matrix dimension alignment algorithm. For example, using the nearest neighbor interpolation algorithm, the dimensions of the compensation matrix can be adjusted to match those of the local model parameters of each edge computing node.

[0087] Step S546: Encapsulate the model parameter compensation matrix as structured instruction code to generate a model parameter compensation instruction containing feature weight adjustment rules and parameter validation conditions. Structured instruction code represents the model parameter compensation matrix in a structured manner, making it easier for edge computing nodes to parse and execute it. Feature weight adjustment rules are rules used to guide edge computing nodes on how to adjust the feature weights of the local disease identification model. Parameter validation conditions refer to the conditions under which the model parameter compensation instruction takes effect, such as at the beginning of a monitoring cycle.

[0088] When encapsulating the model parameter compensation matrix as a structured instruction code, the JSON format can be used. The model parameter compensation matrix, feature weight adjustment rules, and parameter validation conditions are encapsulated in the form of a JSON object.

[0089] Step S547: Send model parameter compensation instructions to each edge computing node in the identified target abnormal node set through an encrypted transmission channel, and perform instruction parsing and parameter injection operations in each edge computing node, while updating the version identification information of the local model to trigger feature extraction strategy adaptation in subsequent monitoring cycles. The encrypted transmission channel is a channel for securely transmitting model parameter compensation instructions, which can prevent the instructions from being stolen or tampered with during transmission. Instruction parsing refers to the edge computing node parsing the received model parameter compensation instructions to extract information such as the model parameter compensation matrix, feature weight adjustment rules, and parameter effectiveness conditions. The parameter injection operation refers to applying the parsed model parameter compensation matrix and feature weight adjustment rules to the local disease recognition model to adjust the parameters of the model. Version identification information is information used to identify the local model version. Updating the version identification information can trigger feature extraction strategy adaptation in subsequent monitoring cycles.

[0090] When issuing model parameter compensation instructions to each edge computing node in the target abnormal node set via an encrypted transmission channel, the instructions can be encrypted using the SSL / TLS protocol. Upon receiving the instructions, the edge computing node first parses the instructions to extract information such as the model parameter compensation matrix, feature weight adjustment rules, and parameter validation conditions. Then, the feature weights of the local disease recognition model are adjusted according to the feature weight adjustment rules. The model parameter compensation matrix is ​​applied to the parameters of the local disease recognition model, completing the parameter injection operation. Finally, the local model version identifier is updated, for example, by incrementing the version number by 1, triggering feature extraction strategy adaptation in subsequent monitoring cycles. For example, the model parameter compensation instruction can be encrypted and transmitted to an edge computing node in the target abnormal node set using the SSL / TLS protocol. Upon receiving the instruction, the node parses the instruction using a JSON parsing library to extract information such as the model parameter compensation matrix, feature weight adjustment rules, and parameter validation conditions. Based on the feature weight adjustment rules, the weight of feature 1 in the local disease recognition model is increased by 0.1, and the weight of feature 2 is decreased by 0.05. The model parameter compensation matrix is ​​applied to the parameters of the local disease recognition model, completing the parameter injection operation. Update the version number of the local model from 1.0 to 1.1, triggering the adaptation of the feature extraction strategy in subsequent monitoring cycles.

[0091] Step S550: Establish a two-way verification mechanism between the edge node and the cloud platform, and perform a cross-node recognition result consistency check after each parameter iteration, so that the feature extraction rules after the model update meet the multi-region collaborative recognition accuracy requirements. The two-way verification mechanism is a mechanism for mutual verification between the edge node and the cloud platform, which can ensure the accuracy and consistency of the data. The cross-node recognition result consistency check refers to the comparison and verification of the disease recognition results of different edge nodes to check whether they are consistent. The multi-region collaborative recognition accuracy requirement refers to the requirement for recognition accuracy when collaborative disease recognition is performed in multiple regions. When establishing a two-way verification mechanism between the edge node and the cloud platform, it is first necessary to determine the content and method of verification. The content of the verification may include the accuracy of the data uploaded by the edge node, the correctness of the instructions issued by the cloud platform, etc. The verification method may adopt methods such as data comparison and check code. After each parameter iteration, a cross-node recognition result consistency check is performed.

[0092] As an implementation method, step S550 may specifically include the following steps S551 to S557: Step S551: Generate a standardized verification case set covering different climate conditions and traffic load scenarios in the cloud platform, the standardized verification case set includes multiple verification cases, and each verification case includes a manually labeled disease type label and corresponding multimodal sensor data features. The standardized verification case set is a set of use cases used to verify the accuracy of the edge node disease recognition model, which covers different climate conditions and traffic load scenarios. The manually labeled disease type label is a label that is manually labeled for the disease type, which can be used as a reference standard for verification. Multimodal sensor data features refer to data features collected by multiple sensors corresponding to the disease type, such as vibration sensor features, optical sensor features, temperature sensor features, etc.

[0093] When generating a standardized set of verification cases on the cloud platform, multimodal sensor data is first collected under different climate conditions and traffic load scenarios. Sensors can be installed on roads in different regions to collect multimodal sensor data at different times and under different environmental conditions. Professionals then manually annotate the collected data to determine the damage type labels. These manually annotated damage type labels and the corresponding multimodal sensor data features are combined to form a standardized set of verification cases. For example, multimodal sensor data from roads is collected under different climate conditions (such as sunny, rainy, and snowy days) and traffic load scenarios (such as peak and off-peak hours). Professionals manually annotate this data to determine the damage type labels, such as cracks and potholes. These manually annotated damage type labels and the corresponding multimodal sensor data features are combined to form a standardized set of verification cases, which contains multiple verification cases.

[0094] Step S552: Distribute the standardized set of verification cases to each edge computing node, triggering the local defect recognition model to process the verification cases in parallel, and collect the recognition result sets and corresponding feature extraction process metadata output by each node. When distributing the standardized set of verification cases to each edge computing node, a network communication protocol, such as HTTP, can be used to transmit data to each edge computing node. After receiving the standardized set of verification cases, each edge computing node triggers the local defect recognition model to process the verification cases in parallel. Parallel processing can improve processing efficiency and shorten verification time. During processing, feature extraction process metadata, such as the execution time and intermediate results of the feature extraction algorithm, is recorded. After processing is complete, the recognition result sets and corresponding feature extraction process metadata output by each node are collected. For example, the standardized set of verification cases is distributed to 10 edge computing nodes using HTTP. After receiving the data, each edge computing node triggers the local defect recognition model to process the verification cases in parallel. During processing, feature extraction process metadata, such as the execution time and intermediate results of the feature extraction algorithm, is recorded. After processing is complete, the recognition result sets and corresponding feature extraction process metadata output by each node are collected.

[0095] Step S553: ​​Perform cross-node consistency analysis on the set of recognition results, calculate the recognition result difference index for the same verification case at different edge nodes, and detect anomalous verification case groups whose difference exceeds a preset threshold. Cross-node consistency analysis is the process of comparing and analyzing the recognition results of different edge nodes to verify the consistency of the recognition results. The recognition result difference index is used to measure the degree of difference in the recognition results of the same verification case at different edge nodes. An anomalous verification case group is a set of verification cases whose recognition result difference exceeds a preset threshold.

[0096] When performing cross-node consistency analysis, the recognition result set must first be organized and preprocessed. The recognition results for the same verification case at different edge nodes are aligned to facilitate comparison. Next, a difference index for the recognition results is calculated. This difference index can be calculated using methods such as Hamming distance and Euclidean distance. For example, for a verification case, the recognition results for different edge nodes are "crack disease," "pothole disease," and "crack disease," respectively. The difference between these can be calculated using Hamming distance. Finally, abnormal verification case groups whose difference exceeds a preset threshold are detected. The preset threshold can be set based on actual conditions, such as 0.2. When the difference in the recognition results exceeds the preset threshold, the verification case is marked as an abnormal verification case. For example, after organizing and preprocessing the recognition result set, the Hamming distance is used to calculate the difference index for the recognition results for the same verification case at different edge nodes. If three verification cases are found to have a difference exceeding the preset threshold of 0.2, these three verification cases are marked as an abnormal verification case group.

[0097] As an implementation method, step S553 may specifically include the following steps S5531 to S5537: Step S5531: Generate a spatial association relationship map of edge computing nodes based on the highway network topology structure, and divide nodes with adjacent physical locations and overlapping monitoring ranges into associated node groups. The highway network topology structure refers to the topological structure of the highway network, which describes the connection relationship and position relationship between highways. The spatial association relationship map of edge computing nodes is a map used to represent the spatial association relationship between edge computing nodes, which can intuitively display the adjacent relationship between nodes and the overlap of monitoring ranges. An associated node group refers to a collection of edge computing nodes with adjacent physical locations and overlapping monitoring ranges.

[0098] Graph theory methods can be used to generate a spatial association relationship map of edge computing nodes based on the highway network topology. Each edge computing node is considered a node in the graph, and the connection relationship between nodes represents their adjacency. Based on the highway network topology and the installation location of the edge computing nodes, the connection relationship between the nodes is determined, and a spatial association relationship map is generated. Then, based on the spatial association relationship map, nodes with adjacent physical locations and overlapping monitoring ranges are divided into associated node groups. For example, in a highway network, there are 10 edge computing nodes. Based on the highway network topology and the installation location of the nodes, a spatial association relationship map is generated. By analyzing the map, it is found that there are three nodes with adjacent physical locations and overlapping monitoring ranges. These three nodes are divided into an associated node group.

[0099] Step S5532: Perform feature space mapping on the recognition results of the associated node group under the same verification case to generate a corresponding defect feature distribution vector for each node. Feature space mapping is the process of mapping the recognition results of the associated node group under the same verification case into the feature space. The defect feature distribution vector represents the distribution of defect features for each node in the feature space and reflects the node's recognition characteristics for the verification case.

[0100] When mapping the recognition results of a group of associated nodes under the same verification use case into a feature space, methods such as principal component analysis (PCA) can be used. First, the recognition results for each node are converted into a feature vector. For example, the disease type label in the recognition result is converted into a numerical code to form a feature vector. Then, PCA is used to reduce the dimensionality of the feature vector and map it into the feature space, generating a disease feature distribution vector corresponding to each node. For example, for three nodes in a group of associated nodes, the recognition results under the same verification use case are "crack disease," "pothole disease," and "crack disease." These recognition results are converted into feature vectors, such as [1, 0, 0], [0, 1, 0], and [1, 0, 0]. PCA is used to reduce the dimensionality of these feature vectors and map them into a two-dimensional feature space, generating a disease feature distribution vector corresponding to each node.

[0101] Step S5533: Calculate the cosine similarity set between each disease feature distribution vector within the associated node group, extract abnormal similarity data points that fall below the historical similarity baseline value, and mark the target associated node group containing the abnormal similarity data points. The historical similarity baseline value is a baseline value for the similarity between disease feature distribution vectors within the associated node group, calculated based on historical data statistics. Abnormal similarity data points are data points whose cosine similarity falls below the historical similarity baseline value. The target associated node group is the associated node group containing the abnormal similarity data points.

[0102] When calculating the cosine similarity set between the disease feature distribution vectors within an associated node group, the cosine similarity calculation formula can be used. For each node pair within the associated node group, the cosine similarity of their disease feature distribution vectors is calculated. The cosine similarities of all node pairs are combined into a cosine similarity set. Then, each data point in the cosine similarity set is compared with a historical similarity benchmark value, and abnormal similarity data points that fall below the historical similarity benchmark value are extracted. Target associated node groups containing abnormal similarity data points are marked.

[0103] Step S5534: Perform multi-dimensional status diagnosis on the target associated node group to obtain the corresponding node's sensor calibration records, data sampling integrity index, and model parameter update time series. Multi-dimensional status diagnosis is the process of performing multi-dimensional status analysis and diagnosis on the target associated node group, with the goal of identifying the causes of discrepancies in recognition results. Sensor calibration records refer to the calibration records of the edge computing node's sensors, which can reflect sensor accuracy. The data sampling integrity index is used to measure the data sampling integrity of the edge computing node and can reflect the data quality. The model parameter update time series refers to the time series of parameter updates of the edge computing node's disease recognition model, which can reflect the model's updates.

[0104] When performing multi-dimensional status diagnosis on a target group of associated nodes, the first step is to obtain the corresponding node's sensor calibration records, data sampling integrity indicators, and model parameter update time series. This information can be obtained by querying the edge computing node's log files or database. For example, query the log file of a node in the target group of associated nodes to obtain sensor calibration records, such as the most recent calibration time and results. Calculate the node's data sampling integrity indicators, such as the sampling rate and missing data ratio. Obtain the node's model parameter update time series, including the time of each update and the updated parameters.

[0105] Step S5535: Determine the root cause of the abnormality based on the multi-dimensional status diagnosis results: When there is a deviation between the sensor calibration record and the current environmental conditions, a hardware calibration instruction is generated and the data re-collection process is triggered; when the model parameter update time series lags behind the cloud collaborative training cycle, the regional federated learning parameter synchronization task is initiated. The root cause of the abnormality refers to the type of cause that causes the abnormal identification result of the target associated node group, mainly including sensor calibration deviation and model parameter update lag. The hardware calibration instruction is an instruction for calibrating the sensors of the edge computing node. The data re-collection process refers to the process of re-collecting data from the edge computing node. The regional federated learning parameter synchronization task refers to the task of performing federated learning within the region and synchronizing the disease identification model parameters of the edge computing nodes.

[0106] When determining the root cause of an anomaly based on the results of multi-dimensional status diagnosis, first check whether there is a deviation between the sensor calibration record and the current environmental conditions. If there is a deviation, generate a hardware calibration instruction and trigger the data re-collection process. Then, check whether the model parameter update time series lags behind the cloud collaborative training cycle. If there is a lag, initiate the regional federated learning parameter synchronization task. For example, through multi-dimensional status diagnosis, it is found that the sensor calibration record of a node in the target associated node group deviates from the current environmental conditions, generate a hardware calibration instruction and trigger the data re-collection process. At the same time, it is found that the model parameter update time series of another node lags behind the cloud collaborative training cycle, and initiate the regional federated learning parameter synchronization task.

[0107] Step S5536: Dynamically adjust the feature weight aggregation strategy within the regional federated learning parameter synchronization task, prioritizing the integration of pavement structure feature dimensions associated with the anomaly verification use case group. The feature weight aggregation strategy is used to aggregate edge computing node feature weights within the regional federated learning parameter synchronization task. Pavement structure feature dimensions refer to those related to the pavement structure, such as quantitative indicators of structural deformation and surface texture degradation maps.

[0108] When dynamically adjusting the feature weight aggregation strategy in the regional federated learning parameter synchronization task, the characteristics of the abnormal verification case group are first analyzed. The pavement structure feature dimensions associated with the abnormal verification case group are identified. Then, the feature weights of these feature dimensions are prioritized for fusion. Methods such as weighted averaging can be used to aggregate feature weights. For example, by analyzing the abnormal verification case group, it was found that two pavement structure feature dimensions, the structural deformation quantitative index and the surface texture degradation map, were highly correlated. In the regional federated learning parameter synchronization task, the feature weights of these two feature dimensions are prioritized for fusion, and the weighted averaging method is used to aggregate the feature weights of these two feature dimensions for each edge computing node.

[0109] Step S5537: Monitor the cosine similarity trends of the target associated node group during parameter synchronization in real time. When the cosine similarity returns to the historical baseline fluctuation range, terminate the synchronization task and update the abnormal node status flag in the cloud platform. During the execution of the regional federated learning parameter synchronization task, monitor the cosine similarity trends of the target associated node group in real time. This can be done by regularly calculating the cosine similarity between the disease feature distribution vectors within the associated node group and observing its changes. When the cosine similarity returns to the historical baseline fluctuation range, it indicates that the parameter synchronization task has achieved good results, and the synchronization task is terminated. Simultaneously, the abnormal node status flag in the cloud platform is updated, marking the abnormal node as a normal node. For example, during the execution of the regional federated learning parameter synchronization task, calculate the cosine similarity between the disease feature distribution vectors within the target associated node group every hour. After a period of synchronization, if the cosine similarity returns to the historical baseline fluctuation range, terminate the synchronization task, and update the abnormal node status flag in the cloud platform.

[0110] Step S554: Generate a collaborative calibration task instruction set for the edge node combination associated with the abnormal verification use case group. The instruction set specifies the role assignments for the lead node and auxiliary node, as well as the data interaction protocol. The collaborative calibration task instruction set is a set of instructions used to guide the edge node combination associated with the abnormal verification use case group to perform collaborative calibration. The lead node is the node that plays a leading role in the collaborative calibration process and is responsible for coordinating and managing the calibration process. The auxiliary node is the node that assists the lead node in calibration during the collaborative calibration process. The data interaction protocol is a protocol used to standardize data interaction between edge nodes, which can ensure the accurate transmission and sharing of data.

[0111] When generating a collaborative calibration task instruction set, first determine the edge node combination associated with the abnormal verification use case group. Then, specify the role assignments of the dominant node and the auxiliary node based on the node performance and resource conditions. For example, select a node with strong computing power and large data storage capacity as the dominant node. At the same time, formulate a data interaction protocol to specify the format, frequency, and security mechanism of data transmission between nodes. Include information such as role assignment and data interaction protocol in the collaborative calibration task instruction set. For example, for an edge node combination associated with an abnormal verification use case group, select node A as the dominant node, and node B and node C as auxiliary nodes. Formulate a data interaction protocol to specify that data be transmitted between nodes in JSON format, every 10 minutes, and encrypted using the SSL / TLS protocol. Include this information in the collaborative calibration task instruction set.

[0112] Step S555: A federated learning-based model parameter exchange channel is established between the lead node and the auxiliary nodes, enabling targeted optimization of model parameters through encrypted transmission of intermediate feature gradients. This federated learning-based model parameter exchange channel is used to exchange model parameters between the lead node and auxiliary nodes, enabling model parameter sharing and optimization between nodes. Intermediate feature gradients are calculated during model training and reflect the direction of model parameter updates. Encrypted transmission encrypts the intermediate feature gradients before transmission to ensure data security.

[0113] When establishing a federated learning-based model parameter exchange channel between the leading and supporting nodes, an encrypted communication protocol, such as SSL / TLS, can be used. The leading and supporting nodes perform model training locally and calculate intermediate feature gradients. These gradients are then encrypted and transmitted to the other node via the model parameter exchange channel. Upon receiving the intermediate feature gradients, the other node decrypts them and updates its local model parameters based on the gradient information. This enables targeted optimization of model parameters. For example, the leading and supporting nodes perform model training locally using the same training data and calculate intermediate feature gradients. These gradients are encrypted using SSL / TLS and transmitted to the other node via the model parameter exchange channel. Upon receiving the encrypted intermediate feature gradients, the other node decrypts them using the corresponding key and updates its local model parameters based on the gradient information, achieving targeted optimization of model parameters.

[0114] Step S556: Perform regional adaptive adjustment on the model parameters of the dominant node according to the local data distribution characteristics of the auxiliary node, generate a set of updated parameters with spatial adaptability, and synchronize them to all edge nodes participating in the collaborative calibration. The local data distribution characteristics of the auxiliary node refer to the spatial and feature distribution of the data collected by the auxiliary node, which can reflect the characteristics of the region where the node is located. Regional adaptive adjustment is to adjust the model parameters of the dominant node according to the local data distribution characteristics of the auxiliary node to make it more suitable for the situation in the region. The set of updated parameters with spatial adaptability is a set of model parameters obtained after regional adaptive adjustment, which can improve the recognition accuracy of the model in different regions.

[0115] When regionally adapting the model parameters of the dominant node based on the local data distribution characteristics of the auxiliary nodes, the local data distribution characteristics of the auxiliary nodes are first analyzed. For example, the characteristic distribution and spatial distribution of the data are analyzed. Then, based on the analysis results, the model parameters of the dominant node are adjusted. Adjustments can be made using methods such as weighted averaging and linear transformation. The adjusted model parameters are combined into an updated parameter set with spatial adaptability. Finally, the updated parameter set is synchronized to all edge nodes participating in the collaborative calibration. For example, by analyzing the local data distribution characteristics of the auxiliary nodes, it is found that the pavement disease in this area is mainly manifested as crack disease. Based on this characteristic, the model parameters of the dominant node are adjusted to increase the feature weights related to crack disease. The adjusted model parameters are combined into an updated parameter set with spatial adaptability and synchronized to all edge nodes participating in the collaborative calibration.

[0116] Step S557: After parameter synchronization is complete, the edge node re-executes the recognition task for the verification case and transmits the updated recognition results back to the cloud platform for use in correcting the regional weight coefficients of the global model error distribution heatmap. After parameter synchronization is complete, the edge node re-executes the recognition task for the verification case. Because the model parameters have been adjusted and optimized, the updated recognition results should be more accurate. The updated recognition results are transmitted back to the cloud platform. After receiving the updated recognition results, the cloud platform corrects the regional weight coefficients of the global model error distribution heatmap based on these results. Regional weight coefficients represent the importance of different regions in the global model error distribution. Correcting the regional weight coefficients can make the global model error distribution heatmap more accurately reflect the actual situation. For example, after parameter synchronization is complete, the edge node re-executes the recognition task for the verification case and obtains updated recognition results. The updated recognition results are transmitted back to the cloud platform via the network communication protocol. Based on these results, the cloud platform adjusts the weight coefficients of the corresponding regions in the global model error distribution heatmap, so that the heatmap more accurately reflects the model error conditions in different regions.

[0117] It is understandable that the various algorithms involved in the introduction of the embodiments of the present invention, such as the cosine distance algorithm, the nearest neighbor interpolation algorithm, etc., can be learned from the relevant content in the prior art. In order to save space, they will not be expanded too much in the embodiments of the present invention. In addition, when implementing the scheme of the present invention, those skilled in the art can supplement the details according to the common knowledge in this field. For example, according to the common knowledge in this field, normalization can be used to eliminate dimensional conflicts before feature fusion, interpolation can be used to eliminate dimensional differences, and historical data, experience or business scenario requirements can be combined to reasonably set the threshold, and the model can be trained based on the general model training method, etc. The present invention will no longer provide redundant introductions to the overly detailed implementation process.

[0118] In the automated disease identification system of the embodiments of the present invention, both the edge computing node and the cloud-based analysis platform include a processor and a memory communicatively connected to the processor. The memories of the edge computing node and the cloud-based analysis platform store instructions executable by their respective processors. When these instructions are executed by their respective processors, the automated target disease identification method based on cloud-edge collaboration provided by the embodiments of the present invention is implemented.

[0119] The following is a schematic diagram of the structure of a computer system for implementing edge computing nodes and / or cloud analysis platforms provided by an embodiment of the present invention. Figure 3 The computer system includes a computing unit 1001, which can perform various appropriate actions and processes according to a computer program stored in a ROM (read-only memory) 1002 or a computer program loaded from a storage unit 1008 into a RAM (random access memory) 1003. RAM 1003 may also store various programs and data required for the operation of the cloud-based analysis platform 120. The computing unit 1001, ROM 1002, and RAM 1003 are interconnected via a bus 1004. An I / O interface (input / output interface) 1005 is also connected to bus 1004.

[0120] Several components are connected to the I / O interface 1005, including an input unit 1006, an output unit 1007, a storage unit 1008, and a communication unit 1009. The input unit 1006 can be any type of device capable of inputting information into the cloud-based analysis platform 120. The input unit 1006 can receive input numeric or character information and generate key signal inputs related to user settings and / or function control of the server. The output unit 1007 can be any type of device capable of presenting information. The storage unit 1008 can include, but is not limited to, a magnetic disk or an optical disk. The communication unit 1009 allows the cloud-based analysis platform 120 to exchange information / data with other devices via computer networks such as the Internet and / or various telecommunications networks. The computing unit 1001 can be any general-purpose and / or specialized processing component with processing and computing capabilities. The computing unit 1001 can include, but is not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various specialized artificial intelligence (AI) computing chips, various computing units that run machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 1001 performs the various methods and processes described above, such as the target disease automated identification method based on cloud-edge collaboration. For example, in some embodiments, the target disease automated identification method based on cloud-edge collaboration can be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit 1008. In some embodiments, part or all of the computer program can be loaded and / or installed on the cloud analysis platform 120 via the ROM 1002 and / or the communication unit 1009. When the computer program is loaded into the RAM 1003 and executed by the computing unit 1001, one or more steps of the target disease automated identification method based on cloud-edge collaboration described above can be performed. Alternatively, in other embodiments, the computing unit 1001 can be configured to execute the target disease automated identification method based on cloud-edge collaboration in any other appropriate manner (for example, by means of firmware).

Claims

1. A method for automatic identification of target diseases based on cloud-edge collaboration, characterized in that: The following steps are involved: Collecting a set of raw road surface monitoring data by edge computing nodes deployed in a highway monitoring area, wherein the raw road surface monitoring data set includes multimodal sensing information and corresponding spatiotemporal location identifiers; In the edge computing node, a multi-dimensional disease feature set is generated based on the original road surface monitoring data set, wherein the multi-dimensional disease feature set includes structural deformation features, surface texture degradation features, and environmental interference correlation features; Uploading the multi-dimensional disease feature set to a cloud analysis platform, performing feature matching calculations on the multi-dimensional disease feature set and standard disease patterns in a cloud disease feature library through the cloud analysis platform, and generating target disease type identification results and corresponding confidence assessment parameters; According to the result of the relationship between the confidence assessment parameter and the preset threshold, the feature extraction strategy of the edge computing node is adjusted to generate an edge node adaptive optimization instruction set; The target disease type identification result and the edge node adaptive optimization instruction set are synchronized to the edge computing node to trigger the feature extraction rule update and disease identification model parameter iteration operation for subsequent monitoring cycles.

2. The method according to claim 1, wherein In the edge computing node, generating a multi-dimensional disease feature set based on the original road surface monitoring data set includes: Performing spatiotemporal alignment processing on the original road surface monitoring data set to eliminate time stamp deviation and spatial coordinate offset of data collected by different sensor devices, and generating a standardized monitoring data sequence; Performing multi-scale feature decomposition processing on the standardized monitoring data sequence, extracting pavement structure response characteristics at different scales, and calculating energy distribution difference coefficients between adjacent scale features; Combining the energy distribution difference coefficient with a preset environmental noise suppression model, filtering and optimizing the intermediate feature set generated by the multi-scale feature decomposition process to generate a denoised primary feature set; Inputting the primary feature set into a feature fusion network deployed locally at the edge node, and calculating feature association weights between different sensor data channels through a cross-modal feature attention mechanism; The primary feature set is weightedly fused according to the feature association weights to generate a multi-dimensional disease feature set including a structural deformation quantitative index, a surface texture degradation map and an environmental interference association matrix.

3. The method according to claim 2, wherein The cloud analysis platform calculates the feature matching degree between the multi-dimensional disease feature set and the standard disease pattern in the cloud disease feature library to generate the target disease type identification result and the corresponding confidence assessment parameter, including: Deconstructing the characteristic dimensions of the standard disease patterns in the cloud disease feature library, and extracting the characteristic weight distribution vector and key characteristic constraint conditions corresponding to each standard disease pattern; Comparing each feature subset in the multi-dimensional disease feature set with the standard disease pattern of the corresponding dimension layer by layer, and calculating feature similarity measurement parameters and abnormal feature deviation index; Constructing a dynamic weight adjustment function based on the abnormal feature deviation index, performing adaptive weighting processing on the feature similarity measurement parameters, and generating a comprehensive matching score set; generating an initial disease type identification result according to the standard disease pattern identifier corresponding to the highest score in the comprehensive matching score set; The confidence level of the initial disease type identification result is verified, including calculating the difference ratio between adjacent scores and the stability parameter of the historical identification results, and generating a two-dimensional confidence evaluation parameter including a credibility level and an error tolerance range.

4. The method according to claim 3, wherein The step of adjusting the feature extraction strategy of the edge computing node based on the result of determining the relationship between the confidence evaluation parameter and the preset threshold, and generating an edge node adaptive optimization instruction set, includes: Determining the optimization priority and parameter adjustment range of the current feature extraction strategy based on the credibility level classification result in the confidence assessment parameter; When the credibility level is lower than a first preset threshold, activating a feature space expansion mode, adding a high-frequency vibration signal acquisition channel in the edge computing node, and expanding the number of feature mapping layer nodes in the feature fusion network; When the error tolerance range exceeds a second preset threshold, a multi-source data verification mechanism is triggered, and a cross-validation processing flow of the lidar point cloud data and the visible light image data is introduced into the edge computing node; Generate a strategy optimization trend prediction model based on the historical feature matching calculation results, and adjust the data sampling frequency of the edge computing node and the computational complexity parameters of the feature extraction algorithm; The feature space expansion mode, multi-source data verification mechanism and computational complexity adjustment parameters are integrated into a configurable instruction code block to generate an adaptive optimization instruction set including execution condition judgment logic and parameter update rules.

5. The method according to claim 4, wherein The triggering of feature extraction rule update and disease identification model parameter iteration for subsequent monitoring cycles includes: Establish a feature extraction rule version control library in the edge computing node to record feature dimension change information and algorithm performance indicators after each strategy adjustment; Incrementally updating the current feature extraction rules according to the execution condition judgment logic in the adaptive optimization instruction set, retaining the feature channel configuration associated with the historical high-confidence recognition results; Build a disease identification model parameter optimization queue in the cloud analysis platform and generate a global model error distribution heat map based on the confidence assessment parameters uploaded by multiple edge computing nodes; According to the regional abnormal nodes identified in the global model error distribution heat map, model parameter compensation instructions are issued to the corresponding edge computing nodes to adjust the classification boundary conditions and feature weight distribution ratio of the local disease identification model; A two-way verification mechanism is established between the edge nodes and the cloud platform, and a cross-node recognition result consistency check is performed after each parameter iteration to ensure that the feature extraction rules after the model update meet the multi-region collaborative recognition accuracy requirements.

6. The method according to claim 2, wherein The multi-scale feature decomposition processing is performed on the standardized monitoring data sequence to extract the pavement structure response characteristics at different scales, including: Construct a multi-scale analysis framework based on wavelet transform, set the mother wavelet function and scale decomposition level parameters that match the characteristics of highway materials; Performing a time-frequency joint analysis on the standardized monitoring data sequence to calculate the energy density distribution function and frequency band correlation index at each scale; According to the preset pavement structure damage sensitive frequency band range, the scale level containing the key vibration mode is screened out, and the wavelet coefficient matrix of the corresponding scale is extracted; Performing singular value decomposition on the wavelet coefficient matrix to obtain a principal component eigenvector reflecting the overall stiffness change of the pavement structure; The difference between the principal component eigenvector and the historical health status benchmark data is combined to generate a pavement structure response characteristic that quantitatively describes the degree of structural deformation.

7. The method according to claim 2, wherein The feature association weights between different sensor data channels are calculated through the cross-modal feature attention mechanism, including: Performing dimensionality unification processing on the vibration sensing feature vector, the optical sensing feature vector, and the temperature sensing feature vector in the primary feature set to generate a standardized multimodal feature set with the same vector dimension; Constructing a multimodal feature interaction space in the feature fusion network, projecting each feature vector in the standardized multimodal feature set into a shared embedding space, and generating an intermediate feature embedding set with cross-modal comparability; performing a bidirectional attention calculation on the vibration modal embedding vector and the optical texture embedding vector in the intermediate feature embedding set to generate a first attention weight matrix reflecting the correlation between the vibration and optical features; Performing cross-attention calculation on the temperature distribution embedding vector and the vibration mode embedding vector in the intermediate feature embedding set to generate a second attention weight matrix reflecting the temperature-vibration coupling relationship; Constructing a multimodal feature fusion coefficient tensor based on the first attention weight matrix and the second attention weight matrix, and performing a tensor convolution operation on each feature vector in the standardized multimodal feature set; The convolutional multimodal features are selectively enhanced through a dynamic gating mechanism, retaining cross-modal correlation features that are strongly correlated with pavement defects and suppressing weakly correlated feature components caused by environmental noise. The enhanced cross-modal correlation features are hierarchically spliced ​​with the original unimodal features to generate a fused disease feature set containing complementary information from multiple source data.

8. The method according to claim 5, wherein The step of sending a model parameter compensation instruction to a corresponding edge computing node based on the regional abnormal node identified in the global model error distribution heat map specifically includes: Analyze the spatial anomaly distribution pattern of the global model error distribution heat map, identify the target abnormal node set whose error accumulation value exceeds the statistical mean of the adjacent regions and meets the preset duration period, and extract the multi-dimensional environment association feature vector corresponding to the target abnormal node set; Perform feature space mapping on the multi-dimensional environment-related feature vector and the regional environment adaptation template in the cloud-based disease feature library to screen out a reference disease pattern set that matches the pavement material properties, climate conditions, and traffic load characteristics of the target abnormal node set; Performing feature weight decoupling analysis on each disease pattern in the reference disease pattern set to separate a core feature weight set directly associated with pavement structure damage and an interference feature weight set coupled with environmental noise; Constructing a parameter compensation benchmark model based on the core feature weight set, and calculating the orthogonal projection residual matrix of the local disease recognition model parameters of each edge computing node in the target abnormal node set and the benchmark model parameters in the feature space; Generate a model parameter compensation matrix with spatial constraints based on the orthogonal projection residual matrix, and eliminate the tensor structure differences between the compensation matrix and the local model parameters of each edge computing node through a matrix dimension alignment algorithm; Encapsulating the model parameter compensation matrix into a structured instruction code to generate a model parameter compensation instruction including feature weight adjustment rules and parameter effectiveness conditions; The model parameter compensation instruction is sent to each edge computing node in the identified target abnormal node set through an encrypted transmission channel, and instruction parsing and parameter injection operations are performed in each edge computing node. At the same time, the version identification information of the local model is updated to trigger the feature extraction strategy adaptation of the subsequent monitoring cycle.

9. The method according to claim 5, wherein The establishment of a two-way authentication mechanism between the edge node and the cloud platform specifically includes: Generate a standardized set of verification cases covering different climate conditions and traffic load scenarios on a cloud platform, wherein the standardized verification case set includes multiple verification cases, each of which includes a manually annotated disease type label and corresponding multimodal sensor data features; Distribute the standardized verification case set to each edge computing node, trigger the local disease recognition model to process the verification case in parallel, and collect the recognition result set output by each node and the corresponding feature extraction process metadata; Performing cross-node consistency analysis on the recognition result set, calculating the difference index of the recognition results of the same verification case at different edge nodes, and detecting abnormal verification case groups whose difference exceeds a preset threshold; For the edge node combination associated with the abnormal verification case group, a collaborative calibration task instruction set is generated, in which the role allocation and data interaction protocol of the leading node and the auxiliary node are specified; Establish a model parameter exchange channel based on federated learning between the leading node and the auxiliary node, and achieve targeted optimization of model parameters by encrypting the transmission of intermediate feature gradients; The model parameters of the dominant node are regionally adapted according to the local data distribution characteristics of the auxiliary nodes to generate a spatially adaptive updated parameter set, which is then synchronized to all edge nodes participating in the collaborative calibration. After completing the parameter synchronization, the recognition task of the verification use case is re-executed, and the updated recognition result is transmitted back to the cloud platform for correcting the regional weight coefficient of the global model error distribution heat map.

10. An automatic disease identification system, characterized in that: The method comprises an edge computing node and a cloud analysis platform that communicate with each other, each of the edge computing node and the cloud analysis platform comprising a processor; and a memory that is communicatively connected to the processor; the memories of the edge computing node and the cloud analysis platform store instructions that can be executed by their corresponding processors, and when the instructions are respectively executed by the corresponding processors, the method according to any one of claims 1 to 9 is executed.

Images