Bridge cable state evaluation method based on unsupervised single classification analysis

Abstract

The invention relates to the technical field of bridge health monitoring, in particular to a bridge cable state evaluation method based on unsupervised single classification analysis. The method comprises the following steps: acquiring bridge data of a plurality of target bridges, preprocessing all the acquired bridge data to obtain a plurality of preprocessed data, creating an evaluation model, inputting the preprocessed data into the evaluation model, extracting image feature vectors of the target bridges based on the preprocessed data through the evaluation model, and evaluating the target bridges according to the image feature vectors. The method comprises the steps of extracting an HOG feature vector of a target bridge, performing single classification model training based on the image feature vector to obtain a trained evaluation model, finally collecting real-time data of a real-time bridge, inputting the real-time data into the trained evaluation model, and judging the cable state of the real-time bridge through the trained evaluation model. And single-classification analysis learning is performed based on the HOG feature vector, so that the trained evaluation model can quickly judge the cable state according to the real-time data of the real-time bridge, and the detection efficiency of the cable state is improved.

Description

Technical Field

[0001] This invention relates to the field of bridge health monitoring technology, and more specifically, to a method for assessing the condition of bridge cables based on unsupervised single-classification analysis. Background Technology

[0002] The cable system of long-span bridges is the main load-bearing structure of suspension bridges, cable-stayed bridges, and arch suspension bridges. Because they span rivers, lakes, and seas and bear dynamic and static loads, and are constantly exposed to wind, rain, humidity, and polluted air, they are susceptible to stress corrosion, fatigue corrosion, and metal corrosion. The cables and anchorages are all prone to corrosion damage. Therefore, to ensure the safety of the bridge and cable structure, corrosion protection for the cables is essential.

[0003] Chinese Patent No. CN119043520B discloses a method and apparatus for identifying the health status of a cable. This method involves acquiring cable temperature data and ambient temperature data; determining whether the cable body is damaged based on the cable temperature data; determining the temperature response coefficient and response hysteresis coefficient of the cable based on the cable temperature data and the ambient temperature data; and determining whether the cable clamp is loose based on the temperature response coefficient and the response hysteresis coefficient. However, existing technologies often require consideration of multiple cable states, leading to low efficiency in determining abnormal states and a high risk of errors. Summary of the Invention

[0004] The purpose of this invention is to address the problems existing in the background technology by proposing a bridge cable condition assessment method based on unsupervised single classification analysis.

[0005] The technical solution of the present invention: On the one hand, this application provides a method for assessing the condition of bridge cables based on unsupervised single classification analysis, including: Bridge data from multiple target bridges is collected, and all collected bridge data is preprocessed to obtain multiple preprocessed data sets. Create an evaluation model; The preprocessed data is input into the evaluation model. The evaluation model extracts the image feature vector of the target bridge based on the preprocessed data, and trains a single classification model based on the image feature vector to obtain the trained evaluation model. Real-time data of the bridge is collected and input into a trained evaluation model. The trained evaluation model is then used to determine the cable status of the bridge in real time.

[0006] Preferably, bridge data from multiple target bridges is collected, and all collected bridge data is preprocessed to obtain multiple preprocessed data sets, including: Create a bridge database; For multiple target bridges, set the collection parameters of each target bridge respectively; Based on the collection parameters, data collection is performed on each target bridge to obtain multiple bridge data and image data, and all collected data are put into a bridge database; the bridge data includes cable force and cable dynamic characteristics; An image data is randomly selected from the bridge database; The image data is denoised to obtain a preprocessed image; An image data is randomly selected from the bridge database, and multiple preprocessed data are obtained.

[0007] Preferably, the preprocessed data is input into an evaluation model, the image feature vector of the target bridge is extracted based on the preprocessed data through the evaluation model, and a single classification model training is performed based on the image feature vector to obtain a trained evaluation model, including: All preprocessed data is divided into a training set and a test set according to a random ratio; The training set is input into the evaluation model, the image feature vector of each cable image is extracted through the evaluation model, a feature matrix is constructed based on the image feature vector, a single classification model training is performed in combination with the feature matrix, and thus a trained evaluation model is obtained; the trained evaluation model has the ability to automatically extract image features according to the input cable image and judge the cable state of the target bridge based on the image features; The test set is input into the trained evaluation model to determine whether the trained evaluation model is trained.

[0008] Preferably, the training set is input into the evaluation model, the image feature vector of each cable image is extracted through the evaluation model, a feature matrix is constructed based on the image feature vector, a single classification model training is performed in combination with the feature matrix, and thus a trained evaluation model is obtained, including: A cable image is randomly selected from the training set; The cable image is grayed to obtain a cable grayscale image; The image HOG feature vector is extracted based on the cable grayscale image; A cable image is randomly selected from the training set until all cable images in the training set are selected to obtain the image feature vector of each cable image; A feature matrix is established based on the image feature vector; each row of the feature matrix is an image feature vector of a cable image.

[0009] Preferably, the image HOG feature vector is extracted based on the cable grayscale image, including: A size standard is set, and the size of the cable grayscale image is set to the standard size to obtain a standard grayscale image; Obtain the gradient number of the direction, and divide the standard gray image into a plurality of cells; Combine the plurality of cells into a unit block, and perform normalization processing on the unit block to obtain a normalized image; Slide the normalized image, calculate the HOG feature of each position of the normalized image, and obtain a plurality of HOG features; Serially connect the plurality of HOG features to obtain a HOG feature vector.

[0010] Preferably, real-time data of the real-time bridge is collected, the real-time data is input into the trained evaluation model, and the cable state of the real-time bridge is judged by the trained evaluation model, including: Collecting real-time data of a real-time bridge; the real-time data includes real-time bridge data and real-time image data; Input the real-time data into the trained evaluation model, and perform clustering analysis on the real-time bridge based on the bridge data by the trained evaluation model to obtain the target bridge corresponding to the real-time bridge; the target bridge corresponding to the real-time bridge is recorded as a real-time target; Extract the real-time feature vector of the real-time bridge, and judge the cable state of the real-time bridge based on the real-time feature vector and the feature matrix of the real-time target.

[0011] Preferably, the real-time data is input into the trained evaluation model, and the clustering analysis is performed on the real-time bridge based on the bridge data by the trained evaluation model to obtain the target bridge corresponding to the real-time bridge; the target bridge corresponding to the real-time bridge is recorded as a real-time target, including: Randomly select K target bridges, record the selected K target bridges as initial cluster centers, and randomly distribute the remaining target bridges and the real-time bridge to the K initial cluster centers to form K clusters; Set the iteration number threshold; For each cluster, the distance from each point in the cluster to the cluster center is calculated, and the point corresponding to the average distance is recorded as a new cluster center; Determine whether the iteration number is greater than or equal to the iteration number threshold; If the iteration number is greater than or equal to the iteration number threshold, the target bridge corresponding to the cluster center obtained in the last iteration is recorded as the real-time target.

[0012] Preferably, the real-time feature vector of the real-time bridge is extracted, and the cable state of the real-time bridge is judged based on the real-time feature vector and the feature matrix of the real-time target, including: Set the difference threshold; Obtain the real-time feature vector of the real-time bridge based on the real-time image data by the trained evaluation model, and calculate the difference between the real-time feature vector and the feature matrix of the real-time target; judging whether the difference between the real-time feature vector and the feature matrix of the real-time target is greater than or equal to the difference threshold value; If the difference between the real-time feature vector and the feature matrix of the real-time target is greater than or equal to the difference threshold value, it is judged that the cable state of the real-time bridge is abnormal.

[0013] On the other hand, the application also provides a bridge cable state evaluation system based on unsupervised single classification analysis, which is applied to the bridge cable state evaluation method based on unsupervised single classification analysis in any of the preceding aspects, comprising a collection component and an evaluation component, the bridge data of the target bridge is collected through the collection component, the evaluation component is in communication connection with the collection component, the image features of the target bridge are extracted based on the bridge data through the evaluation component, and the cable state of the template bridge is judged in combination with the image features.

[0014] Preferably, the collection component comprises a data collection module and an image collection module, the bridge data of the target bridge is collected through the data collection module, and the bridge image of the target bridge is collected through the image collection module.

[0015] Compared with the prior art, the above technical scheme of the application has the following beneficial technical effects: By collecting bridge data of a plurality of target bridges, all collected bridge data is preprocessed to obtain a plurality of preprocessed data, then an evaluation model is created, and the preprocessed data is input into the evaluation model, the image feature vector of the target bridge is extracted based on the preprocessed data through the evaluation model, and single classification model training is performed based on the image feature vector to obtain a trained evaluation model, finally real-time data of a real-time bridge is collected, the real-time data is input into the trained evaluation model, and the cable state of the real-time bridge is judged through the trained evaluation model, the HOG feature vector of the target bridge is extracted, and single classification analysis learning is performed based on the HOG feature vector, so that the trained evaluation model can quickly judge the cable state according to the real-time data of the real-time bridge, and the detection efficiency of the cable state is improved. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 A flowchart of a bridge cable state evaluation method based on unsupervised single classification analysis is provided for the application; Figure 2 A principle block diagram of a bridge cable state evaluation system based on unsupervised single classification analysis is provided for the application; REFERENCE NUMERALS: 100, collection component; 101, data collection module; 102, image collection module; 200, evaluation component. DETAILED DESCRIPTION

[0017] In example one, as Figure 1As shown, the bridge cable state evaluation method based on unsupervised single classification analysis is provided, which comprises: S100, collecting bridge data of multiple target bridges, preprocessing all collected bridge data to obtain multiple preprocessing data; Specifically, the bridge data collected in the application is the data of the bridge cable under normal state, so as to facilitate subsequent single classification analysis of the evaluation model; S200, creating an evaluation model; S300, inputting the preprocessing data into the evaluation model, extracting the image feature vector of the target bridge based on the preprocessing data through the evaluation model, and performing single classification model training based on the image feature vector to obtain the trained evaluation model; S400, collecting real-time data of a real-time bridge, inputting the real-time data into the trained evaluation model, and judging the cable state of the real-time bridge through the trained evaluation model.

[0018] In the application, by collecting bridge data of multiple target bridges, preprocessing all collected bridge data to obtain multiple preprocessing data, then creating an evaluation model, inputting the preprocessing data into the evaluation model, extracting the image feature vector of the target bridge based on the preprocessing data through the evaluation model, and performing single classification model training based on the image feature vector to obtain the trained evaluation model, finally collecting real-time data of a real-time bridge, inputting the real-time data into the trained evaluation model, and judging the cable state of the real-time bridge through the trained evaluation model, the HOG feature vector of the target bridge is extracted, and single classification analysis learning is performed based on the HOG feature vector, so that the trained evaluation model can quickly judge the cable state according to the real-time data of the real-time bridge, and the detection efficiency of the cable state is improved.

[0019] In an optional embodiment, the S100 comprises: S110, creating a bridge database; S120, setting the collection parameters of each target bridge respectively for multiple target bridges; S130, collecting data of each target bridge based on the collection parameters to obtain multiple bridge data and image data, and putting all collected data into the bridge database; the bridge data comprises cable force and cable dynamic characteristics; Specifically, the cable dynamic characteristics comprise natural frequency, amplitude, vibration mode and damping ratio, and the dynamic characteristic data is collected because the cable will vibrate under the action of wind and traffic flow, so that the cable is damaged by fatigue, and the dynamic characteristics can also reflect the change of cable force or structure connection state; S140, randomly selecting an image data from the bridge database; S150, denoising the image data to obtain a preprocessed image; S160, returning to randomly selecting one image data from the bridge database to obtain a plurality of preprocessed data.

[0020] It should be noted that, since in the daily monitoring of the bridge, the periodic detection of the state of the bridge cable is one of the important means to ensure the safety of the bridge, therefore, it is necessary to collect the bridge data of the target bridge and the image data corresponding to each bridge data at the same time, so as to complete the evaluation of the state of the bridge cable based on these data.

[0021] After the data is collected, in order to ensure the integrity and reliability of the data, the image data is denoised to improve the quality of the image data. The reason for using image data as training samples is that image data is a non-contact detection method that can be applied to high-altitude detection, thereby improving the applicability of the present application. In addition, image data can also detect early defects that sensors are difficult to capture, so as to discover cable state abnormalities in time.

[0022] In an optional embodiment, the S300 comprises: S310, dividing all preprocessed data into a training set and a test set according to a random ratio; S320, inputting the training set into the evaluation model, extracting the image feature vector of each cable image through the evaluation model, and constructing a feature matrix based on the image feature vector, and training a single classification model combined with the feature matrix, thereby obtaining a trained evaluation model; the trained evaluation model has the ability to automatically extract image features according to the input cable image, and to judge the state of the cable of the target bridge based on the image features; S330, inputting the test set into the trained evaluation model to determine whether the trained evaluation model is trained.

[0023] It should be noted that when dividing the training set and the test set, the division ratio of the training set should be greater than that of the test set, so as to increase the number of training samples contained in the training set. When verifying whether the trained evaluation model is trained through the test set, the output accuracy and / or response time of the trained evaluation model can be used as a judgment standard.

[0024] By inputting the training samples in the training set into the evaluation model in turn, the evaluation model completes the data of the training samples based on the single classification analysis method, which is essentially a thought of using a data-driven method to identify and evaluate the abnormal state of the cable. Single classification analysis is a method of establishing a model that can describe the "normal state" based on a large number of "normal" samples, and then judging whether a new sample belongs to the normal category. If it does not belong to the normal category, it is considered to be an abnormal or unknown state. Since single classification analysis only trains normal sample data, it does not need to collect abnormal samples and is suitable for abnormal detection tasks.

[0025] Since the bridge cable is in a normal state most of the time during use, it is difficult to collect data of the abnormal state of the bridge cable, and the sample size is small. Therefore, the evaluation model is trained by the single classification analysis method, a baseline model of the health state of the cable is constructed based on the normal state data, and the abnormal state deviating from the normal state can be identified by collecting real-time cable data.

[0026] In an optional embodiment, the S320 comprises: S321, randomly selecting a cable image from the training set; S322, performing grayscale processing on the cable image to obtain a cable grayscale image; S323, extracting an image HOG feature vector based on the cable grayscale image; Specifically, HOG is a histogram of oriented gradients, which is an image feature description method. The method divides the image into cell units to extract gradient direction distribution, and normalizes the block level to enhance the adaptability to light changes. S324, returning to randomly selecting a cable image from the training set until all cable images in the training set are selected, and obtaining an image feature vector of each cable image; S325, establishing a feature matrix based on the image feature vector. Each row of the feature matrix is an image feature vector of a cable image.

[0027] It should be noted that the HOG feature vector of each image data is collected, so that the HOG feature vector reflects the characteristics of the bridge cable corresponding to the image data. Since the image data is all normal bridge cable images, each HOG feature vector collected can represent that the bridge cable is in a normal state. Therefore, when the difference between the HOG feature vector of the image data of the target bridge and the feature matrix is too large, it means that the state of the cable of the target bridge is abnormal, thereby improving the efficiency of the state detection of the cable of the target bridge.

[0028] In an optional embodiment, the S323 comprises: S3231, Set the size standard and set the cable grayscale image size to the standard size to obtain the standard grayscale image; S3232, obtains the directional gradient number and divides the standard grayscale image into multiple cells; S3233 combines multiple cells into a cell block and normalizes the cell block to obtain a normalized image; S3234, slide the normalized image, calculate the HOG features at each position of the normalized image, and obtain multiple HOG features; S3235 concatenates multiple HOG features to obtain the HOG feature vector.

[0029] It should be noted that when extracting HOG features from an image, the reason for setting the cable grayscale image to the same standard size is to ensure that the number of cell blocks is consistent in the subsequent division. Before dividing the cell, the gradient of the standard grayscale image in the X and Y directions needs to be calculated using the Sobel operator, and the gradient magnitude and direction of each pixel need to be calculated in order to divide the cell.

[0030] In an optional embodiment, S400 includes: S410, Collect real-time data of the bridge; the real-time data includes real-time bridge data and real-time image data; S420: Input real-time data into the trained evaluation model. The trained evaluation model performs cluster analysis on the real-time bridges based on the real-time bridge data to obtain the target bridges corresponding to the real-time bridges. The target bridges corresponding to the real-time bridges are recorded as real-time targets. S430 extracts the real-time feature vector of the bridge and determines the cable status of the bridge based on the real-time feature vector and the feature matrix of the real-time target.

[0031] It should be noted that by feeding the real-time data of the real-time bridge into the trained evaluation model, the trained evaluation model can then filter out the real-time target that is closest to the real-time bridge from the database based on the real-time bridge data. Since the database stores target bridges in normal state, the real-time target is the target bridge in normal state that is closest to the real-time bridge. Therefore, if the real-time feature vector of the real-time bridge differs significantly from the feature vector of the target bridge in normal state that is closest to it, the cable state of the real-time bridge can be considered abnormal.

[0032] In an optional embodiment, S420 includes: S421, randomly select K target bridges, record the selected K target bridges as the initial cluster centers, and randomly assign the remaining target bridges and real-time bridges to the K initial cluster centers to form K clusters; S422, Set the threshold for the number of iterations; S423, For each cluster, calculate the distance from each point in the cluster to the cluster center, and record the point corresponding to the average distance as the new cluster center; S424, determine whether the number of iterations is greater than or equal to the iteration number threshold; S425, if the number of iterations is greater than or equal to the iteration number threshold, then the target bridge corresponding to the cluster center obtained in the last iteration is recorded as the real-time target; Specifically, if the number of iterations is less than the iteration threshold, return to step S423 until the number of iterations is greater than or equal to the iteration threshold.

[0033] It should be noted that this application uses the K-means clustering method to cluster real-time bridges, thereby selecting the real-time targets that are closest to the real-time bridges. Then, the cable status of the real-time bridges can be determined based on the feature vectors of the real-time targets, thus improving the detection efficiency of the cable status of the target bridges.

[0034] K-means clustering is a widely used clustering algorithm, primarily used to divide data into K predefined clusters. This method iteratively assigns data points to the nearest centroids (i.e., "centroids") and updates the positions of these centroids until a stopping condition is met.

[0035] In an optional embodiment, S430 includes: S431, Set the difference threshold; S432, the real-time feature vector of the bridge is obtained based on real-time image data through the trained evaluation model, and the difference between the real-time feature vector and the feature matrix of the real-time target is calculated. S433, determine whether the difference between the real-time feature vector and the feature matrix of the real-time target is greater than or equal to the difference threshold; S434, if the difference between the real-time feature vector and the feature matrix of the real-time target is greater than or equal to the difference threshold, then the cable status of the real-time bridge is judged to be abnormal.

[0036] It should be noted that by inputting real-time image data of the real-time bridge into the trained evaluation model, the real-time feature vector of the real-time bridge can be obtained. Then, the difference between the feature matrix of the real-time target and the real-time feature vector of the real-time bridge is calculated. If the difference is too large, it means that the cable status of the real-time bridge is abnormal. Therefore, the difference threshold is actually a descriptive value for judging whether the data difference between the real-time bridge and the target bridge is normal. So, the larger the difference threshold is set, the more lenient the judgment range of the cable status of the real-time bridge is, and the smaller the difference threshold is set, the more precise the judgment range of the cable status of the real-time bridge is.

[0037] As Figure 2 The application also provides a bridge cable state evaluation system based on unsupervised single classification analysis, which is applied to the bridge cable state evaluation method based on unsupervised single classification analysis in any of the preceding embodiments, and comprises a collection component and an evaluation component. The bridge data of the target bridge is collected by the collection component. The evaluation component is in communication connection with the collection component. The image features of the target bridge are extracted based on the bridge data by the evaluation component, and the cable state of the template bridge is judged in combination with the image features.

[0038] It should be noted that the bridge data and the image data of the target bridge are collected by the collection component, and all the collected data are input into the evaluation component. The cable state of the target bridge is judged by the evaluation component in combination with the collected data, so that the problem cable can be found in time.

[0039] In an optional embodiment, the collection component comprises a data collection module and an image collection module. The bridge data of the target bridge is collected by the data collection module, and the bridge image of the target bridge is collected by the image collection module.

[0040] It should be noted that the bridge data of the target bridge is collected by the data collection module. The data collection module comprises a plurality of different types of sensors, so as to collect a plurality of different types of bridge data. The bridge image of the target bridge is collected by the image collection module. The image collection module can be set as an infrared camera.

[0041] The embodiments of the application are described in detail above in combination with the drawings, but the application is not limited thereto. Various changes can be made within the knowledge of those skilled in the art without departing from the spirit of the application.

Claims

1. A bridge cable condition assessment method based on unsupervised single classification analysis, characterized in that, The method comprises the following steps: Collecting bridge data of multiple target bridges, preprocessing all collected bridge data to obtain multiple preprocessing data; Creating an evaluation model; Inputting the preprocessing data into the evaluation model, extracting the image feature vector of the target bridge based on the preprocessing data through the evaluation model, and training a single classification model based on the image feature vector to obtain a trained evaluation model; Collecting real-time data of a real-time bridge, inputting the real-time data into the trained evaluation model, and determining the cable state of the real-time bridge through the trained evaluation model.

2. The bridge cable condition assessment method based on unsupervised single classification analysis according to claim 1, characterized in that, Collecting bridge data of multiple target bridges, preprocessing all collected bridge data to obtain multiple preprocessing data, comprising: Creating a bridge database; For multiple target bridges, set the collection parameters of each target bridge; Based on the collection parameters, data collection is performed on each target bridge to obtain multiple bridge data and image data, and all collected data is put into the bridge database; the bridge data includes cable force and cable dynamic characteristics; Randomly selecting an image data from the bridge database; Denoising the image data to obtain a preprocessed image; Return to randomly select an image data from the bridge database to obtain multiple preprocessing data.

3. The bridge cable condition assessment method based on unsupervised single classification analysis according to claim 2, characterized in that, Inputting the preprocessing data into the evaluation model, extracting the image feature vector of the target bridge based on the preprocessing data through the evaluation model, and training a single classification model based on the image feature vector to obtain a trained evaluation model, comprising: Divide all preprocessing data into training set and test set according to random proportion; Input the training set into the evaluation model, extract the image feature vector of each cable image through the evaluation model, construct a feature matrix based on the image feature vector, and train a single classification model combined with the feature matrix to obtain a trained evaluation model; the trained evaluation model has the ability to automatically extract image features according to the input cable image and determine the cable state of the target bridge based on the image features; Input the test set into the trained evaluation model to determine whether the trained evaluation model is trained.

4. The bridge cable condition assessment method based on unsupervised single classification analysis according to claim 3, characterized in that, Input the training set into the evaluation model, extract the image feature vector of each cable image through the evaluation model, construct a feature matrix based on the image feature vector, and train a single classification model combined with the feature matrix to obtain a trained evaluation model, comprising: Randomly select a cable image from the training set; Grayscale processing of the cable image to obtain a cable grayscale image; Extracting image HOG feature vector based on the cable grayscale image; Return to randomly select a cable image from the training set until all cable images in the training set are selected to complete, and obtain the image feature vector of each cable image; Establishing a feature matrix based on the image feature vector; each row of the feature matrix is an image feature vector of a cable image.

5. The bridge cable condition assessment method based on unsupervised single classification analysis according to claim 4, characterized in that, Extracting image HOG feature vector based on the cable grayscale image, comprising: Setting a size standard and setting the size of the cable grayscale image to the standard size to obtain a standard grayscale image; Obtaining the direction gradient number and dividing the standard grayscale image into multiple cells; Combine multiple cells into a unit block, and normalize the unit block to obtain a normalized image; Slide the normalized image, calculate the HOG feature of each position of the normalized image, and obtain multiple HOG features; Serially connect the multiple HOG features to obtain a HOG feature vector.

6. The bridge cable condition assessment method based on unsupervised single classification analysis according to claim 5, wherein, Collect real-time data of the real-time bridge, input the real-time data into the trained evaluation model, and determine the cable state of the real-time bridge through the trained evaluation model, including: Collecting real-time data of the real-time bridge; the real-time data includes real-time bridge data and real-time image data; Input the real-time data into the trained evaluation model, and perform clustering analysis on the real-time bridge based on the bridge data through the trained evaluation model to obtain the target bridge corresponding to the real-time bridge; the target bridge corresponding to the real-time bridge is recorded as a real-time target; Extract the real-time feature vector of the real-time bridge, and determine the cable state of the real-time bridge based on the real-time feature vector and the feature matrix of the real-time target.

7. The bridge cable condition assessment method based on unsupervised single classification analysis according to claim 6, characterized in that, Input the real-time data into the trained evaluation model, and perform clustering analysis on the real-time bridge based on the bridge data through the trained evaluation model to obtain the target bridge corresponding to the real-time bridge; Record the target bridge corresponding to the real-time bridge as a real-time target, including: Randomly select K target bridges, record the selected K target bridges as initial cluster centers, and randomly distribute the remaining target bridges and the real-time bridge to the K initial cluster centers to form K clusters; Set an iteration threshold; For each cluster, calculate the distance from each point in the cluster to the cluster center, and record the point corresponding to the average distance as a new cluster center; Determine whether the iteration number is greater than or equal to the iteration threshold; If the iteration number is greater than or equal to the iteration threshold, record the target bridge corresponding to the cluster center obtained in the last iteration as the real-time target.

8. The bridge cable condition assessment method based on unsupervised single classification analysis according to claim 7, characterized in that, Extract the real-time feature vector of the real-time bridge, and determine the cable state of the real-time bridge based on the real-time feature vector and the feature matrix of the real-time target, including: Set a difference threshold; Obtain the real-time feature vector of the real-time bridge based on the real-time image data through the trained evaluation model, and calculate the difference between the real-time feature vector and the feature matrix of the real-time target; Determine whether the difference between the real-time feature vector and the feature matrix of the real-time target is greater than or equal to the difference threshold; If the difference between the real-time feature vector and the feature matrix of the real-time target is greater than or equal to the difference threshold, determine that the cable state of the real-time bridge is abnormal.

9. A bridge cable condition assessment system based on unsupervised single classification analysis, comprising the bridge cable condition assessment method based on unsupervised single classification analysis according to any one of claims 1-8, characterized in that, Including: A collection component, through which bridge data of a target bridge is collected; An evaluation component, which is in communication connection with the collection component, through which image features of the target bridge are extracted based on the bridge data, and the cable state of the template bridge is determined in combination with the image features.

10. The bridge cable condition assessment system based on unsupervised single classification analysis according to claim 9, wherein, The collection component includes a data collection module and an image collection module, through which bridge data of a target bridge is collected, and through which bridge images of a target bridge are collected.

Images