Artificial intelligence-based visual two-dimension deflection detection system

Abstract

This invention discloses an artificial intelligence-based visual dual-dimensional deflection detection system, belonging to the field of engineering quality inspection technology. It includes a visual image acquisition module, a deflection feature extraction module, an intelligent error compensation module, a multi-dimensional data fusion module, a dynamic evolution prediction module, a deflection fault early warning module, and a self-learning and optimization module. This AI-based visual dual-dimensional deflection detection system overcomes the shortcomings of traditional systems, such as single data source, manual feature extraction, and poor robustness in dynamic environments, by employing adaptive feature extraction, a deflection prediction model, and a self-learning and optimization module. The system can not only correct detection deviations in real time by fusing multi-dimensional visual, stress, and displacement information, improving detection accuracy and stability, but also possesses autonomous optimization and updating capabilities, maintaining efficient and accurate deflection prediction during long-term monitoring, thereby effectively ensuring the health and safety of structures.

Description

Technical Field

[0001] This invention relates to the field of engineering quality inspection technology, specifically to a visual two-dimensional deflection detection system based on artificial intelligence. Background Technology

[0002] In the health monitoring of modern infrastructure, deflection detection is one of the important means to assess structural safety and stability. Traditional deflection detection mainly relies on single sensors (such as displacement sensors, stress sensors, etc.) or simple visual monitoring methods. These methods are usually limited to single-dimensional data acquisition and analysis, resulting in low detection accuracy and the inability to detect local and subtle structural deformations in a timely manner. In addition, traditional deflection detection systems often require manual intervention for data analysis and feature extraction, leading to low efficiency and susceptibility to human factors. In complex detection environments, the multi-dimensional characteristics of deflection often exhibit dynamic complexity, and existing technologies lack the ability to effectively extract and fuse these complex features, resulting in a significant reduction in the accuracy and robustness of the system in long-term monitoring.

[0003] The disadvantages of existing technologies are as follows:

[0004] Limitations of a single data source: Traditional deflection detection systems often rely on a single vision, displacement, or stress sensor. This single-modal data acquisition method cannot comprehensively capture the multidimensional deflection information of complex structures, resulting in low and unstable detection accuracy, especially prone to errors in changing external environments.

[0005] The inefficiency of manual feature extraction: Most traditional deflection detection systems require manual setting of feature templates and rely on manual analysis of image or sensor data, making it difficult to handle complex and variable deflection features. Furthermore, the efficiency and accuracy of feature extraction are easily affected by human factors.

[0006] Lack of multi-source data fusion capability: Existing deflection detection systems often focus only on single-type sensor data, failing to effectively utilize the combined effects of stress, displacement, and visual information. This limitation of lacking multi-dimensional data fusion makes it difficult for the system to perform accurate real-time correction and global prediction in practical applications.

[0007] Poor robustness in dynamic environments: Under changing environmental conditions, traditional systems are difficult to dynamically adjust the model, making the detection results susceptible to noise interference and lacking sufficient robustness and stability. Summary of the Invention

[0008] To address the shortcomings of existing technologies, this invention provides a visual two-dimensional deflection detection system based on artificial intelligence, in order to solve the problems mentioned in the background.

[0009] To achieve the above objectives, the present invention provides the following technical solution: a visual dual-dimensional deflection detection system based on artificial intelligence, comprising a visual image acquisition module, a deflection feature extraction module, an intelligent error compensation module, a multi-dimensional data fusion module, a dynamic evolution prediction module, a deflection fault early warning module, and a self-learning and optimization module;

[0010] The visual image acquisition module acquires two-dimensional images or point cloud data of the structure in real time based on a high-resolution camera or lidar.

[0011] The deflection feature extraction module automatically extracts deflection features from the image using a deep learning model based on a convolutional neural network.

[0012] The intelligent error compensation module, based on a dynamic error compensation algorithm, compensates for errors in real time through the identification and tracking of image feature points, time series analysis, and Kalman filtering, ensuring the accuracy of deflection detection results.

[0013] The multidimensional data fusion module deeply integrates dual-dimensional visual information with sensor data, such as stress sensor and displacement sensor data, to form a multimodal data analysis framework. By fusing detection data from different dimensions, it corrects deviations in visual deflection detection in real time.

[0014] The dynamic evolution prediction module dynamically predicts deflection changes through time series analysis and long short-term memory networks.

[0015] The deflection fault early warning module detects deflection anomalies and issues an alarm to the operator when the deflection exceeds a set safety threshold.

[0016] The self-learning and optimization module, based on reinforcement learning and adaptive algorithms, continuously optimizes the system's performance according to actual detection data.

[0017] To further optimize this technical solution, the visual image acquisition module uses a camera or lidar to avoid the influence of lighting or weather factors in complex environments and obtain high-quality two-dimensional images or point cloud data.

[0018] Cameras or lidar acquire two-dimensional images or point cloud data from different angles by synchronizing multiple sensors, enabling comprehensive monitoring of the structure.

[0019] This module also makes real-time adjustments based on changes in the external environment and automatically calibrates parameters to improve the accuracy and clarity of two-dimensional images or point cloud data.

[0020] To further optimize this technical solution, the deflection feature extraction module is trained using a large amount of labeled structural deformation data to automatically identify and segment the deflection region.

[0021] This module uses an adaptive method to handle various complex deformation forms without the need for preset feature templates. It also dynamically updates the deflection feature library based on historical detection data to perform self-optimization and improve the accuracy of feature extraction.

[0022] To further optimize this technical solution, the adaptive method includes the following steps:

[0023] Adaptive based on multi-scale feature extraction: Employs a multi-scale convolutional neural network to simultaneously extract subtle and large-scale deflection features in convolutional layers at different scales;

[0024] Morphological adaptation and geometric invariants: By combining geometric invariants and morphological feature extraction methods, geometric invariant features such as curvature and angle are selected to ensure the robustness of feature extraction; morphological operations such as dilation and erosion are used to extract different types of boundary features to cope with the nonlinear deformation of the structure.

[0025] Adaptive transfer learning in deep learning: adapting learned deflection features to new structures;

[0026] Adaptive feature selection based on self-attention mechanism: During the feature extraction process, the region most important for deflection detection is automatically identified and background noise is ignored. Different regions are assigned different weights to adapt to different structural deflection features.

[0027] The dynamic update of the deflection feature library includes the following process:

[0028] New deflection features are extracted from historical detection data periodically, and the feature library is continuously expanded and updated based on the current detection situation;

[0029] Automatic labeling is performed by detecting the internal correlations of the data, and the labeled data is used to train and optimize the model;

[0030] New data from each detection is added to the model's training set in real time. The currently detected deflection features are compared with the historical feature library, and the model parameters are updated in real time when new changes or patterns are discovered.

[0031] To further optimize this technical solution, the multi-dimensional data fusion module utilizes a multi-sensor data fusion algorithm to find the optimal synergistic relationship among multiple sources of information and construct a global deflection prediction model.

[0032] Suppose we predict the time of a certain structure ( Deflection at time () It utilizes the output data from multiple sensors, such as visual data, stress, and displacement, to perform fusion modeling;

[0033] The deflection prediction model is shown below.

[0034]

[0035] in,

[0036] ( ) for in time ( The predicted deflection value;

[0037] ( The deflection is the result of extracting two-dimensional visual information obtained from visual detection.

[0038] ( ) is the stress sensor in time ( The stress data provided;

[0039] ( ) is the displacement sensor in time ( The displacement data provided;

[0040] ( ) are weighting coefficients, representing the contribution of different data sources to the model;

[0041] ( ) is a time decay function used to consider the influence of historical deflection on the current deflection, reflecting the temporal change of deflection;

[0042] ( ) represents Gaussian noise, indicating measurement error or unpredictable randomness.

[0043] To further optimize this technical solution, in the deflection prediction model, the change in deflection is time-dependent, and the deformation of the structure gradually intensifies over time, based on the time decay function ( To analyze the impact of historical deflection on current deflection, the function uses the following exponential decay form:

[0044]

[0045] in,( () is the time decay constant, indicating that the influence of historical deflection gradually weakens as time increases;

[0046] By using a time decay function, the long-term deflection evolution of the structure is captured, thus avoiding predictions that rely solely on instantaneous data.

[0047] To further optimize this technical solution, the deflection prediction model is optimized and tuned based on Bayesian inference and information entropy. The Bayesian inference is used for weight optimization, and the information entropy is used for uncertainty analysis.

[0048] The Bayesian inference formula is as follows:

[0049]

[0050] in,( ) is the sensor (given historical data) The weighted distribution probability of ); ) is the likelihood of the data given the weights; ) is the prior probability of the weight, ( ) represents the total probability of the data;

[0051] The formula for calculating information entropy is as follows:

[0052]

[0053] in,( ) represents the distribution of sensor data, ( () represents the probability distribution of sensor states; by calculating the data entropy values ​​of different sensors, weights are intelligently allocated to reduce interference from high-entropy data.

[0054] To further optimize this technical solution, the dynamic evolution prediction module learns the historical deflection change trend of the structure and predicts the future deflection in advance, so as to prevent the structure from reaching a harmful deformation state.

[0055] It updates online based on real-time detection data, performs deflection evolution analysis based on the stress-strain relationship of the structure, and visualizes the prediction results as an intuitive trend chart.

[0056] To further optimize this technical solution, the deflection fault early warning module has a built-in AI-based anomaly detection algorithm. By comparing historical data and current data of structural deflection, it accurately identifies potential anomalies and, in conjunction with the influence of environmental factors, determines whether the structure is in a dangerous state. It then sends early warning signals through various means such as mobile apps and computer terminals.

[0057] To further optimize this technical solution, the self-learning and optimization module also uploads the detection data to the cloud based on edge computing and cloud collaborative analysis, compares and learns with other detection cases in the cloud, optimizes model parameters, and automatically identifies repeatability errors during long-term detection and makes optimization adjustments.

[0058] Compared with existing technologies, this invention provides a visual two-dimensional deflection detection system based on artificial intelligence, which has the following beneficial effects:

[0059] This AI-based visual dual-dimensional deflection detection system overcomes the shortcomings of traditional systems, such as single data source, manual feature extraction, and poor robustness in dynamic environments, by employing adaptive feature extraction, deflection prediction models, and self-learning and optimization modules. The system can not only correct detection deviations in real time by integrating multi-dimensional visual, stress, and displacement information, thus improving detection accuracy and stability, but also has the ability to autonomously optimize and update. It can maintain efficient and accurate deflection prediction during long-term monitoring, thereby effectively ensuring the health and safety of the structure. Attached Figure Description

[0060] Figure 1 This is a schematic diagram of the structure of a visual two-dimensional deflection detection system based on artificial intelligence proposed in this invention;

[0061] Figure 2 This is a flowchart illustrating an adaptive method in a visual two-dimensional deflection detection system based on artificial intelligence proposed in this invention.

[0062] Figure 3 This is a schematic diagram illustrating the process of dynamically updating the deflection feature library in a visual two-dimensional deflection detection system based on artificial intelligence proposed in this invention.

[0063] Figure 4 This is a flowchart illustrating the deflection prediction model in a visual two-dimensional deflection detection system based on artificial intelligence proposed in this invention. Detailed Implementation

[0064] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Example

[0065] Please see Figure 1 A visual dual-dimensional deflection detection system based on artificial intelligence includes a visual image acquisition module, a deflection feature extraction module, an intelligent error compensation module, a multi-dimensional data fusion module, a dynamic evolution prediction module, a deflection fault early warning module, and a self-learning and optimization module.

[0066] The visual image acquisition module acquires two-dimensional images or point cloud data of the structure in real time, based on a high-resolution camera or lidar.

[0067] In this embodiment, the visual image acquisition module uses a camera or lidar to avoid the influence of lighting or weather factors in complex environments and obtain high-quality two-dimensional images or point cloud data.

[0068] Cameras or lidar acquire two-dimensional images or point cloud data from different angles by synchronizing multiple sensors, enabling comprehensive monitoring of the structure.

[0069] This module also makes real-time adjustments based on changes in the external environment and automatically calibrates parameters to improve the accuracy and clarity of two-dimensional images or point cloud data.

[0070] The deflection feature extraction module automatically extracts deflection features from images using a deep learning model based on a convolutional neural network.

[0071] In this embodiment, the deflection feature extraction module is trained using a large amount of labeled structural deformation data to automatically identify and segment the deflection region;

[0072] This module uses an adaptive method to handle various complex deformation forms without the need for preset feature templates. It also dynamically updates the deflection feature library based on historical detection data to perform self-optimization and improve the accuracy of feature extraction.

[0073] like Figure 2 As shown, the adaptive method includes the following steps:

[0074] Adaptive based on multi-scale feature extraction: Employs a multi-scale convolutional neural network to simultaneously extract subtle and large-scale deflection features in convolutional layers at different scales;

[0075] Morphological adaptation and geometric invariants: By combining geometric invariants and morphological feature extraction methods, geometric invariant features such as curvature and angle are selected to ensure the robustness of feature extraction; morphological operations such as dilation and erosion are used to extract different types of boundary features to cope with the nonlinear deformation of the structure.

[0076] Adaptive transfer learning in deep learning: adapting learned deflection features to new structures;

[0077] Adaptive feature selection based on self-attention mechanism: During the feature extraction process, the region most important for deflection detection is automatically identified and background noise is ignored. Different regions are assigned different weights to adapt to different structural deflection features.

[0078] like Figure 3 As shown, the dynamic update of the deflection feature library includes the following process:

[0079] New deflection features are extracted from historical detection data periodically, and the feature library is continuously expanded and updated based on the current detection situation;

[0080] Automatic labeling is performed by detecting the internal correlations of the data, and the labeled data is used to train and optimize the model;

[0081] New data from each detection is added to the model's training set in real time. The currently detected deflection features are compared with the historical feature library, and the model parameters are updated in real time when new changes or patterns are discovered.

[0082] Simultaneously, it possesses anomaly detection capabilities, able to identify characteristic anomalies occurring during the detection process, such as false deflection signals caused by equipment vibration. When an anomaly is detected, the system automatically corrects the model's parameters and uses a feedback mechanism to exclude these anomalies from future detections, thereby reducing false alarms.

[0083] The intelligent error compensation module addresses the potential for data acquisition errors during deflection detection due to external factors such as camera shake and lighting changes. By employing a dynamic error compensation algorithm, the module eliminates these influences through the identification and tracking of image feature points.

[0084] In this embodiment, time series analysis and Kalman filtering are used to compensate for errors in real time, ensuring the accuracy of deflection detection results.

[0085] The multidimensional data fusion module deeply integrates dual-dimensional visual information with sensor data, such as stress sensor and displacement sensor data, to form a multimodal data analysis framework. By fusing detection data from different dimensions, it corrects deviations in visual deflection detection in real time.

[0086] In this embodiment, the multi-dimensional data fusion module uses a multi-sensor data fusion algorithm to find the best synergistic relationship among multi-source information and construct a global deflection prediction model.

[0087] Suppose we predict the time of a certain structure ( Deflection at time () It utilizes the output data from multiple sensors, such as visual data, stress, and displacement, to perform fusion modeling;

[0088] The deflection prediction model is shown below.

[0089]

[0090] in,

[0091] ( ) for in time ( The predicted deflection value;

[0092] ( The deflection is the result of extracting two-dimensional visual information obtained from visual detection.

[0093] ( ) is the stress sensor in time ( The stress data provided;

[0094] ( ) is the displacement sensor in time ( The displacement data provided;

[0095] ( ) are weighting coefficients, representing the contribution of different data sources to the model;

[0096] ( ) is a time decay function used to consider the influence of historical deflection on the current deflection, reflecting the temporal change of deflection;

[0097] ( ) represents Gaussian noise, indicating measurement error or unpredictable randomness.

[0098] Visual inspection data ( This data comes from two-dimensional visual analysis, acquiring images of the structural surface via cameras or LiDAR to extract deflection features. Stress sensor data ( The potential impact of deflection is assessed by detecting changes in internal stress within the structure. Displacement sensors ( It provides information on changes in the structural position, which is often an important signal during the process of large-scale or gradual deformation of the structure.

[0099] Each data source is weighted by a coefficient ( Weighting is applied to balance the contribution of data from different sources to the overall deflection prediction. The weights can be adjusted based on the relevance of features in historical data. For example, in some scenarios, visual information may be more relevant, while in others, stress or displacement data may dominate deflection prediction.

[0100] In the deflection prediction model, the change in deflection is time-dependent, and the deformation of the structure gradually intensifies over time, based on a time decay function ( To analyze the impact of historical deflection on current deflection, the function uses the following exponential decay form:

[0101]

[0102] in,( () is the time decay constant, indicating that the influence of historical deflection gradually weakens as time increases;

[0103] By using a time decay function, the long-term deflection evolution of the structure is captured, thus avoiding predictions that rely solely on instantaneous data.

[0104] Noise term in the model ( Assuming a Gaussian distribution, it simulates various measurement errors and randomness in real-world systems. The presence of a noise term makes the model more robust and prevents overfitting of predictions to a specific set of data. Through Bayesian inference, the noise term ( The accuracy of the prediction results can be ensured by dynamically adjusting the probability distribution through updates.

[0105] The deflection prediction model is optimized and tuned based on Bayesian inference and information entropy. The Bayesian inference is used for weight optimization, and the information entropy is used for uncertainty analysis.

[0106] The Bayesian inference formula is as follows:

[0107]

[0108] in,( ) is the sensor (given historical data) The weighted distribution probability of ); ) is the likelihood of the data given the weights; ) is the prior probability of the weight, ( ) represents the total probability of the data.

[0109] Through the Bayesian inference described above, the system can dynamically adjust the weights during real-time detection, so that the synergistic effect between the data from each sensor can be optimized.

[0110] The formula for calculating information entropy is as follows:

[0111]

[0112] in,( ) represents the distribution of sensor data, ( () represents the probability distribution of sensor states; by calculating the data entropy values ​​of different sensors, weights are intelligently allocated to reduce interference from high-entropy data.

[0113] like Figure 4 As shown, in practical applications, the deflection prediction model can be applied through the following steps:

[0114] Real-time data acquisition and fusion: The system acquires real-time data from various sensors (such as cameras, stress sensors, and displacement sensors) and inputs it into the deflection prediction model.

[0115] Multidimensional data weighted calculation: The system dynamically adjusts the weights of each sensor based on Bayesian inference and information entropy calculation results. ), and then merge these data.

[0116] Deflection evolution prediction: using the time decay function ( The system comprehensively considers current and historical deflection data to predict future deflection changes.

[0117] Model updates and optimizations: Through continuous detection feedback and data accumulation, the system dynamically adjusts model parameters to ensure the accuracy of the prediction model in long-term operation.

[0118] The dynamic evolution prediction module dynamically predicts deflection changes through time series analysis and long short-term memory networks.

[0119] In this embodiment, the dynamic evolution prediction module learns the historical deflection change trend of the structure and predicts the future deflection in advance to prevent the structure from reaching a harmful deformation state.

[0120] It updates online based on real-time detection data, performs deflection evolution analysis based on the stress-strain relationship of the structure, and visualizes the prediction results as an intuitive trend chart.

[0121] The deflection fault early warning module detects deflection anomalies and issues an alarm to the operator when the deflection exceeds a set safety threshold.

[0122] In this embodiment, the deflection fault early warning module has a built-in AI-based anomaly detection algorithm. By comparing the historical data of structural deflection with the current data, it accurately identifies potential anomalies and, in conjunction with the influence of environmental factors, determines whether the structure is in a dangerous state. It then sends early warning signals through various means such as a mobile app and a computer terminal.

[0123] The self-learning and optimization module, based on reinforcement learning and adaptive algorithms, continuously optimizes the system's performance according to actual detection data.

[0124] In this embodiment, the self-learning and optimization module also uploads the detection data to the cloud based on edge computing and cloud collaborative analysis, compares and learns with other detection cases in the cloud, optimizes model parameters, and automatically identifies repeatability errors during long-term detection and makes optimization adjustments.

[0125] The beneficial effects of this invention are:

[0126] This AI-based visual dual-dimensional deflection detection system overcomes the shortcomings of traditional systems, such as single data source, manual feature extraction, and poor robustness in dynamic environments, by employing adaptive feature extraction, deflection prediction models, and self-learning and optimization modules. The system can not only correct detection deviations in real time by integrating multi-dimensional visual, stress, and displacement information, thus improving detection accuracy and stability, but also has the ability to autonomously optimize and update. It can maintain efficient and accurate deflection prediction during long-term monitoring, thereby effectively ensuring the health and safety of the structure.

[0127] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0128] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An artificial intelligence based visual two-dimensional deflection detection system, characterized in that, It includes a visual image acquisition module, a deflection feature extraction module, an intelligent error compensation module, a multi-dimensional data fusion module, a dynamic evolution prediction module, a deflection fault early warning module, and a self-learning and optimization module; The visual image acquisition module acquires two-dimensional images or point cloud data of the structure in real time based on a high-resolution camera or lidar. The deflection feature extraction module automatically extracts deflection features from the image using a deep learning model based on a convolutional neural network. The intelligent error compensation module, based on a dynamic error compensation algorithm, compensates for errors in real time through the identification and tracking of image feature points, time series analysis, and Kalman filtering, ensuring the accuracy of deflection detection results. The multi-dimensional data fusion module deeply integrates dual-dimensional visual information with sensor data, including stress sensor and displacement sensor data, to form a multi-modal data analysis framework. By fusing detection data from different dimensions, it corrects deviations in visual deflection detection in real time. The multi-dimensional data fusion module uses a multi-sensor data fusion algorithm to find the best synergistic relationship among multiple sources of information and construct a global deflection prediction model. Suppose we predict the time of a certain structure ( Deflection at time () It utilizes the output data from multiple sensors, including visual data, stress, and displacement, to perform fusion modeling; The deflection prediction model is shown below: ； in, ) is the deflection prediction value at time (t ).​ ) deflection extracted from the two-dimensional visual information obtained by visual inspection;​ ( ) stress data provided by the stress sensor over time ( ) ( ) displacement data provided by the displacement sensor at time ( ) ) is a weighted coefficient, indicating the contribution of different data sources to the model;​ ) is a time decay function, which is used to consider the influence of historical deflection on current deflection, embodying the time sequence change of deflection;​ ) is a Gaussian noise term representing measurement error or unpredictable randomness;​ The dynamic evolution prediction module dynamically predicts deflection changes through time series analysis and long short-term memory networks. The deflection fault early warning module detects deflection anomalies and issues an alarm to the operator when the deflection exceeds a set safety threshold. The self-learning and optimization module, based on reinforcement learning and adaptive algorithms, continuously optimizes the system's performance according to actual detection data.

2. The visual two-dimensional deflection detection system based on artificial intelligence according to claim 1, wherein, The visual image acquisition module uses a camera or lidar to avoid the influence of lighting or weather factors in complex environments and obtain high-quality two-dimensional images or point cloud data. Cameras or lidar acquire two-dimensional images or point cloud data from different angles by synchronizing multiple sensors, enabling comprehensive monitoring of the structure. This module also makes real-time adjustments based on changes in the external environment and automatically calibrates parameters to improve the accuracy and clarity of two-dimensional images or point cloud data.

3. The visual two-dimensional deflection detection system based on artificial intelligence according to claim 1, wherein, The deflection feature extraction module is trained using a large amount of labeled structural deformation data to automatically identify and segment the deflection region. This module uses an adaptive method to handle various complex deformation forms without the need for preset feature templates. It also dynamically updates the deflection feature library based on historical detection data to perform self-optimization and improve the accuracy of feature extraction.

4. The visual two-dimensional deflection detection system based on artificial intelligence according to claim 3, characterized in that, The adaptive method The process includes the following steps: Adaptive based on multi-scale feature extraction: Employs a multi-scale convolutional neural network to simultaneously extract subtle and large-scale deflection features in convolutional layers at different scales; Morphological adaptation and geometric invariants: By combining geometric invariants and morphological feature extraction methods, geometric invariant features, including curvature and angle, are selected to ensure the robustness of feature extraction; morphological operations, including dilation and erosion, are used to extract different types of boundary features to cope with the nonlinear deformation of the structure. Adaptive transfer learning in deep learning: adapting learned deflection features to new structures; Adaptive feature selection based on self-attention mechanism: During the feature extraction process, the region most important for deflection detection is automatically identified and background noise is ignored. Different regions are assigned different weights to adapt to different structural deflection features. The dynamic update of the deflection feature library includes the following process: New deflection features are extracted from historical detection data periodically, and the feature library is continuously expanded and updated based on the current detection situation; Automatic labeling is performed by detecting the internal correlations of the data, and the labeled data is used to train and optimize the model; New data from each detection is added to the model's training set in real time. The currently detected deflection features are compared with the historical feature library, and the model parameters are updated in real time when new changes or patterns are discovered.

5. The visual two-dimensional deflection detection system based on artificial intelligence according to claim 1, wherein, In the deflection prediction model, the change of deflection has time dependence, the deformation of the structure gradually intensifies with the accumulation of time, and the influence of the historical deflection on the current deflection is analyzed based on a time decay function ( ) in the form of exponential decay as follows: ； wherein, is a time decay constant, indicating that the historical deflection influence gradually weakens as time increases; By using a time decay function, the long-term deflection evolution of the structure is captured, thus avoiding predictions that rely solely on instantaneous data.

6. The visual two-dimensional deflection detection system based on artificial intelligence according to claim 1, wherein, The deflection prediction model is optimized and tuned based on Bayesian inference and information entropy. The Bayesian inference is used for weight optimization, and the information entropy is used for uncertainty analysis. The Bayesian inference formula is as follows: ； in,( ) is the sensor (given historical data) The weighted distribution probability of ); ) is the likelihood of the data given the weights; ) is the prior probability of the weight, ( ) represents the total probability of the data; The formula for calculating information entropy is as follows: ； Wherein, ( ) represents the distribution of sensor data, ( ) is the probability distribution of sensor state; by calculating the data entropy value of different sensors, the weight is intelligently allocated, and the interference of high entropy data is reduced.

7. The visual two-dimensional deflection detection system based on artificial intelligence according to claim 1, wherein, The dynamic evolution prediction module learns the historical deflection change trend of the structure and predicts the future deflection in advance to prevent the structure from reaching a harmful deformation state. It updates online based on real-time detection data, performs deflection evolution analysis based on the stress-strain relationship of the structure, and visualizes the prediction results as an intuitive trend chart.

8. The visual two-dimensional deflection detection system based on artificial intelligence according to claim 1, wherein, The deflection fault early warning module has a built-in AI-based anomaly detection algorithm. By comparing historical data and current data of structural deflection, it accurately identifies potential anomalies and, in conjunction with the influence of environmental factors, determines whether the structure is in a dangerous state. It then sends early warning signals through various means, including mobile apps and computer terminals.

9. The visual two-dimensional deflection detection system based on artificial intelligence according to claim 1, wherein, The self-learning and optimization module also uploads detection data to the cloud based on edge computing and cloud-based collaborative analysis, compares and learns with other detection cases in the cloud, optimizes model parameters, and automatically identifies repeatability errors during long-term detection and makes optimization adjustments.

Images