Deep learning-based interpretation method for multi-source detection result of hidden dangers in ancient building
Deep learning-based integration and interpretation of multi-source detection data in ancient buildings addresses the limitations of traditional methods, enabling efficient and accurate identification and visualization of hidden dangers for informed decision-making.
Patent Information
- Application Number
- CN202411608685.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-12
- Publication Date
- 2024-12-27
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Traditional detection technologies for ancient buildings cause irreversible damage and have limited detection range and accuracy, while traditional interpretation methods are time-consuming and low in precision for multi-source data integration.
A method using deep learning to integrate and interpret multi-source detection data from geological radar, surface wave method, infrared thermal imaging, and transient electromagnetic technologies, involving data preprocessing, model construction, automatic identification, and multi-dimensional visualization.
Automated and precise identification of hidden dangers with improved efficiency and accuracy, providing scientific decision-making support for protection and maintenance of ancient buildings.
Smart Images
Figure CN119200025A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of ancient building hidden danger detection, and in particular to a method for interpreting multi-source detection results of hidden dangers inside ancient buildings based on deep learning. Background Art
[0002] With the continuous improvement of awareness of cultural heritage protection, the health of ancient building structures has become one of the focuses of protection and repair work. In particular, ancient city wall buildings may have cracks, holes, water seepage and other hidden dangers due to long-term exposure to the natural environment, threatening the stability and safety of ancient buildings. However, due to the complex structure of city wall buildings, traditional detection technologies such as drilling and destructive testing will cause irreversible damage to the cultural relics themselves, and the detection range and accuracy are limited.
[0003] In order to meet these challenges, non-destructive detection technologies are widely used in the field of ancient building detection, including geological radar (GPR), surface wave method, infrared thermal imaging, micro-motion, transient electromagnetic and other technologies. These technologies can obtain imaging data of the internal structure of the building and form high-resolution detection images. However, with the diversification of detection data sources, how to effectively integrate and interpret these multi-source detection data has become a technical bottleneck that needs to be solved urgently. Traditional interpretation methods rely on the experience of technicians and have problems such as time-consuming and low precision. The rapid development of deep learning technology provides a new solution for the efficient interpretation of multi-source data.
[0004] Deep learning can automatically extract features and classify hidden dangers by training a large amount of historical detection data, reducing errors caused by human factors and greatly improving the efficiency and accuracy of hidden danger identification. Therefore, the present invention proposes a multi-source detection data interpretation method based on deep learning, which can effectively solve the problem of interpreting hidden danger detection data of ancient buildings such as city walls, and provide a scientific basis for the protection and repair of ancient buildings. Summary of the invention
[0005] The purpose of the present invention is to provide a method for interpreting the results of multi-source detection of hidden dangers inside ancient buildings based on deep learning. It aims to realize the automatic and precise identification and interpretation of hidden dangers inside ancient buildings by combining multi-source detection data such as geological radar, surface wave method, infrared imaging, micro-motion, transient electromagnetic, etc. with deep learning models, overcoming the limitations of traditional manual analysis, and being able to quickly and comprehensively interpret the location, size, type and distribution characteristics of hidden dangers, providing a basis for the protection and maintenance of ancient city wall-type buildings.
[0006] To achieve the above object, the present invention provides a method for interpreting the multi-source detection results of hidden dangers inside ancient buildings based on deep learning, comprising the following steps: Step S1, collecting multi-source detection data and detection interpretation results, and performing data preprocessing; Step S2: construct a deep learning model based on the multi-source detection data and detection interpretation results preprocessed in step S1, and train the model; Step S3: Based on the deep learning model, automatically identify and interpret hidden dangers to obtain multi-dimensional interpretation results; Step S4: Based on the multi-dimensional interpretation results of step S3, perform multi-dimensional visualization analysis and decision support.
[0007] Preferably, in step S1, multi-source detection data is collected and data preprocessing is performed, and the specific process is as follows: Step S11, collection of multi-source detection data and detection interpretation results, the specific process is as follows: Step S111: First, arrange survey lines at different parts of ancient city wall buildings, and collect various types of detection data result profile files and interpretation result tables; Step S112: further, obtaining spatial position information corresponding to each type of detection result profile file and interpretation result table; Step S113: Finally, collect and obtain multi-source detection data and detection interpretation results for subsequent use; Step S12: preprocessing the multi-source detection data and detection interpretation results obtained in step S11, specifically including: Step S121: standardize and structure the acquired multi-source detection data and detection interpretation results by type; Step S122: In order to ensure the spatial accuracy of the data, the real-time dynamic differential positioning technology is used to calibrate and align the coordinates of the multi-source detection data, and the inconsistent multi-source detection data is interpolated; all spatial coordinates are unified into a coordinate system for the next step to call; Step S123: feature-associate the multi-source detection data after coordinate calibration and registration with the detection interpretation results; wherein the detection interpretation result sample library covers various types of abnormal bodies.
[0008] Preferably, multi-source detection is performed using geological radar, surface wave method, infrared thermal imaging, micro-motion, and transient electromagnetic technology to obtain multi-source detection data; Among them, geological radar is suitable for high-resolution detection in shallow layers to identify hidden dangers such as loose bodies and voids; the surface wave method is used to obtain shear wave velocity information of middle and deep soil layers; and infrared thermal imaging is used to detect temperature differences and reveal material degradation and moisture penetration problems under the surface.
[0009] Preferably, in step S2, based on the multi-source detection data and detection interpretation results preprocessed in step S1, a deep learning model is constructed and trained. The specific process is as follows: Step S21, data labeling and processing: using historical detection data of known hidden danger distribution to label geological radar, surface wave method, infrared thermal imaging, micro-motion, and transient electromagnetic data to determine the location, size, and type of hidden dangers; The data set includes various types of hidden dangers, such as shallow cracks, loose bodies, deep voids, and material degradation, to ensure the diversity of the training set; Step S22, deep learning algorithm selection: using a convolutional neural network, a recurrent neural network or a combination of the two models to perform training for multi-source data interpretation; Among them, convolutional neural networks are good at processing image data and can extract features from two-dimensional or three-dimensional detection images of geological radar; recurrent neural networks are suitable for processing sequence data and can analyze the velocity profiles generated by the surface wave method; Step S23, deep learning model training: by training a large amount of labeled data, the deep learning model automatically identifies the characteristics of hidden dangers; The training process includes the use of data enhancement and regularization techniques to improve the generalization ability of the model and prevent overfitting; Step S24, comprehensive judgment deep learning model training: using the processing results of step S23 and the pre-processed detection interpretation results, the deep learning model automatically trains the comprehensive interpretation and risk assessment method of multi-source detection results.
[0010] Preferably, in step S3, hidden dangers are automatically identified and interpreted based on the deep learning model to obtain multi-dimensional interpretation results, including the following steps: Step S31, the hidden danger automatic identification process is as follows: First, the new multi-source detection data is input into the trained deep learning model; second, the deep learning model automatically identifies and marks the potential danger areas; then, the output includes the location, type, size and distribution characteristics of the potential dangers; Step S32, data fusion and multi-dimensional interpretation process is as follows: First, a deep learning model is used to automatically identify multi-source detection data and mark potential danger areas. Second, the potential danger areas identified by multiple sources are superimposed and comprehensively interpreted through manual or machine weight models. Then, for potential dangers of different depths, the deep learning model distinguishes between shallow cracks and deep voids and gives corresponding risk assessments.
[0011] Preferably, in step S4, multi-dimensional visualization analysis and decision support are performed according to the multi-dimensional interpretation results of step S3, and the specific steps are as follows: Step S41: Based on the multi-dimensional interpretation result of step S3, the spatial distribution of hidden dangers is combined with the three-dimensional model of the city wall through three-dimensional modeling technology; the depth, size and material changes of hidden dangers are intuitively displayed by means of color gradient and transparency adjustment, and multi-dimensional visual analysis is performed; Step S42: Generate a detailed detection report based on the severity of the hidden danger, including the risk level of the hidden danger and repair suggestions; the report is output in Shapefile and GeoJSON standard formats for subsequent analysis and application.
[0012] Therefore, the present invention adopts the above-mentioned method for interpreting the multi-source detection results of hidden dangers inside ancient buildings based on deep learning, and the beneficial effects are as follows: (1) Automated identification and interpretation: Deep learning models are used to automatically identify hidden dangers inside ancient buildings such as city walls, greatly improving interpretation efficiency and solving the problem of time-consuming and low-precision manual analysis of multi-source detection data. (2) High precision and high reliability: The combination of multi-source detection technology and deep learning can effectively identify shallow and deep hidden dangers. By comparing and analyzing multiple data, the accuracy and reliability of the identification results are improved; (3) Multi-dimensional data fusion: The present invention uses spatial registration technology and data fusion technology to unify the data from different detection methods and provide a comprehensive interpretation of hidden dangers. It is suitable for multi-dimensional hidden danger analysis of complex building structures.
[0013] (4) Scientific decision-making support: Through visualization technology and detailed detection reports, relevant protection personnel can obtain intuitive information on the distribution of hidden dangers and make scientific repair decisions based on the interpretation results, ensuring the accuracy and efficiency of cultural relics protection work.
[0014] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS
[0015] Figure 1 This is a flow chart of the interpretation method of multi-source detection results of internal hidden dangers in ancient buildings based on deep learning. DETAILED DESCRIPTION
[0016] The technical solution of the present invention is further described below through the accompanying drawings and embodiments.
[0017] like Figure 1 As shown, the method for interpreting the multi-source detection results of hidden dangers inside ancient buildings based on deep learning of the present invention comprises the following steps: Step S1: Collect multi-source detection data and detection interpretation results, and perform data preprocessing.
[0018] In the present invention, multi-source detection is performed using technologies such as geological radar, surface wave method, infrared thermal imaging, micro-motion, and transient electromagnetic to obtain multi-source detection data. Among them, geological radar is suitable for shallow high-resolution detection to identify hidden dangers such as loose bodies and cavities; the surface wave method is used to obtain shear wave velocity information of medium and deep soil bodies; infrared thermal imaging is used to detect temperature differences and reveal problems such as material degradation and moisture penetration under the surface.
[0019] Step S11, collection of multi-source detection data and detection interpretation results, the specific process is as follows: Step S111, first, arrange survey lines at different parts of the ancient city wall buildings (such as the main body of the city wall, the base, and the side walls), and collect various types of detection data result profile files and interpretation result tables; Step S112: further, obtaining spatial position information corresponding to each type of detection result profile file and interpretation result table; Step S113: Finally, collect and obtain multi-source detection data and detection interpretation results for subsequent use.
[0020] Step S12, preprocessing the multi-source detection data and detection interpretation results obtained in step S11, including the following steps: Step S121: standardize and structure the acquired multi-source detection data and detection interpretation results by type; Step S122: In order to ensure the spatial accuracy of the data, the real-time dynamic differential positioning (RTK) technology is used to calibrate and align the coordinates of the multi-source detection data, and the inconsistent multi-source detection data is interpolated; after all the spatial coordinates are unified into a coordinate system, they are provided for the following steps to call; Step S123: feature-associate the multi-source detection data after coordinate calibration and registration with the detection interpretation results; wherein the detection interpretation result sample library covers various types of abnormal bodies.
[0021] Step S2: Based on the multi-source detection data and detection interpretation results preprocessed in step S1, a deep learning model is constructed and the model is trained.
[0022] Step S21, data labeling and processing: Use historical detection data of known hidden danger distribution to label geological radar, surface wave method, infrared thermal imaging, micro-motion, transient electromagnetic and other data to determine the location, size and type of hidden dangers.
[0023] Among them, the data set includes various types of hidden dangers, such as shallow cracks, loose bodies, deep voids, material degradation, etc., to ensure the diversity of the training set.
[0024] Step S22, deep learning algorithm selection: using convolutional neural network (CNN), recurrent neural network (RNN) or a combined model to perform training for multi-source data interpretation.
[0025] Among them, CNN is good at processing image data and can extract features from two-dimensional or three-dimensional detection images of geological radar; RNN is suitable for processing sequence data and can analyze the velocity profile generated by the surface wave method.
[0026] Step S23, deep learning model training: by training a large amount of labeled data, the deep learning model can automatically identify the characteristics of hidden dangers (such as the shape of cracks, the location of cavities, etc.).
[0027] The training process includes the use of data enhancement and regularization techniques to improve the generalization ability of the model and prevent overfitting.
[0028] Step S24, comprehensive judgment deep learning model training: using the processing results of step S23 and the pre-processed detection interpretation results, the deep learning model automatically trains the comprehensive interpretation and risk assessment method of multi-source detection results.
[0029] Step S3: Based on the deep learning model, automatically identify and interpret hidden dangers to obtain multi-dimensional interpretation results.
[0030] Step S31, the hidden danger automatic identification process is as follows: First, the new multi-source detection data is input into the trained deep learning model; second, the deep learning model automatically identifies and marks the potential danger areas; then, the output includes the location, type, size and distribution characteristics of the potential dangers.
[0031] The above-mentioned automatic hidden danger identification process does not require human intervention, thus reducing human errors.
[0032] Step S32, data fusion and multi-dimensional interpretation process is as follows: First, a deep learning model is used to automatically identify multi-source detection data and mark potential danger areas. Second, the potential danger areas identified by multiple sources are superimposed and comprehensively interpreted through manual or machine weight models. Then, for potential dangers of different depths, the deep learning model distinguishes between shallow cracks and deep voids and gives corresponding risk assessments.
[0033] Step S4: Perform multi-dimensional visualization analysis and decision support based on the multi-dimensional interpretation results of step S3.
[0034] Step S41: Based on the multi-dimensional interpretation results of step S3, the spatial distribution of hidden dangers is combined with the three-dimensional model of the city wall through three-dimensional modeling technology. The depth, size and material changes of hidden dangers are intuitively displayed by means of color gradient and transparency adjustment, and multi-dimensional visual analysis is performed.
[0035] Step S42: Generate a detailed detection report based on the severity of the hidden danger, including the risk level of the hidden danger, repair suggestions, etc. The report can be output in standard formats such as Shapefile and GeoJSON for subsequent analysis and application.
[0036] Example Taking an ancient city wall as an example, based on deep learning, the hidden dangers of this section of the ancient city wall are detected and interpreted. The specific process is as follows: 1. Lay out geological radar and surface wave survey lines in a certain section of the ancient city wall to collect multi-source detection data.
[0037] 2. Input the collected multi-source detection data into the trained deep learning model to automatically identify hidden dangers such as loose bodies and voids in the city wall, and mark their location and size.
[0038] 3. Use 3D modeling tools to combine the spatial information of hidden dangers with the 3D model of the city wall to generate a clear 3D hidden danger distribution map.
[0039] 4. Finally, a detection report is generated, which describes in detail the types, distribution characteristics and repair suggestions of hidden dangers to provide support for subsequent protection and repair work.
[0040] Therefore, the present invention adopts the above-mentioned deep learning-based interpretation method of the multi-source detection results of hidden dangers inside ancient buildings. By combining multi-source detection data such as geological radar, surface wave method, infrared imaging, micro-motion, transient electromagnetic, etc. with the deep learning model, it realizes the automatic and precise identification and interpretation of hidden dangers inside ancient buildings, overcomes the limitations of traditional manual analysis, and can quickly and comprehensively interpret the location, size, type and distribution characteristics of hidden dangers, providing a basis for the protection and maintenance of ancient buildings such as city walls.
[0041] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.
Claims
1. A method for interpreting the results of multi-source detection of hidden dangers inside ancient buildings based on deep learning, characterized in that: The following steps are involved: Step S1, collecting multi-source detection data and detection interpretation results, and performing data preprocessing; Step S2: construct a deep learning model based on the multi-source detection data and detection interpretation results preprocessed in step S1, and train the model; Step S3: Based on the deep learning model, automatically identify and interpret hidden dangers to obtain multi-dimensional interpretation results; Step S4: Based on the multi-dimensional interpretation results of step S3, perform multi-dimensional visualization analysis and decision support.
2. The method for interpreting the multi-source detection results of hidden dangers inside ancient buildings based on deep learning according to claim 1 is characterized in that: In step S1, multi-source detection data is collected and data preprocessing is performed. The specific process is as follows: Step S11, collection of multi-source detection data and detection interpretation results, the specific process is as follows: Step S111: First, arrange survey lines at different parts of ancient city wall buildings, and collect various types of detection data result profile files and interpretation result tables; Step S112: further, obtaining spatial position information corresponding to each type of detection result profile file and interpretation result table; Step S113: Finally, collect and obtain multi-source detection data and detection interpretation results for subsequent use; Step S12: preprocessing the multi-source detection data and detection interpretation results obtained in step S11, specifically including: Step S121: standardize and structure the acquired multi-source detection data and detection interpretation results by type; Step S122: In order to ensure the spatial accuracy of the data, the real-time dynamic differential positioning technology is used to calibrate and align the coordinates of the multi-source detection data, and the inconsistent multi-source detection data is interpolated; all spatial coordinates are unified into a coordinate system for the next step to call; Step S123: feature-associate the multi-source detection data after coordinate calibration and registration with the detection interpretation results; wherein the detection interpretation result sample library covers various types of abnormal bodies.
3. The method for interpreting the multi-source detection results of hidden dangers inside ancient buildings based on deep learning according to claim 2 is characterized in that: Use geological radar, surface wave method, infrared thermal imaging, micro-motion, and transient electromagnetic technology to conduct multi-source detection and obtain multi-source detection data; Among them, geological radar is suitable for high-resolution detection in shallow layers to identify hidden dangers such as loose bodies and voids; the surface wave method is used to obtain shear wave velocity information of middle and deep soil layers; and infrared thermal imaging is used to detect temperature differences and reveal material degradation and moisture penetration problems under the surface.
4. The method for interpreting the multi-source detection results of hidden dangers inside ancient buildings based on deep learning according to claim 2 is characterized in that: In step S2, based on the multi-source detection data and detection interpretation results preprocessed in step S1, a deep learning model is constructed and trained. The specific process is as follows: Step S21, data labeling and processing: using historical detection data of known hidden danger distribution to label geological radar, surface wave method, infrared thermal imaging, micro-motion, and transient electromagnetic data to determine the location, size, and type of hidden dangers; The data set includes various types of hidden dangers, such as shallow cracks, loose bodies, deep voids, and material degradation, to ensure the diversity of the training set; Step S22, deep learning algorithm selection: using a convolutional neural network, a recurrent neural network or a combination of the two models to perform training for multi-source data interpretation; Among them, convolutional neural networks are good at processing image data and can extract features from two-dimensional or three-dimensional detection images of geological radar; recurrent neural networks are suitable for processing sequence data and can analyze the velocity profiles generated by the surface wave method; Step S23, deep learning model training: by training a large amount of labeled data, the deep learning model automatically identifies the characteristics of hidden dangers; The training process includes the use of data enhancement and regularization techniques to improve the generalization ability of the model and prevent overfitting; Step S24, comprehensive judgment deep learning model training: using the processing results of step S23 and the pre-processed detection interpretation results, the deep learning model automatically trains the comprehensive interpretation and risk assessment method of multi-source detection results.
5. The method for interpreting the multi-source detection results of hidden dangers inside ancient buildings based on deep learning according to claim 1 is characterized in that: In step S3, based on the deep learning model, hidden dangers are automatically identified and interpreted to obtain multi-dimensional interpretation results, including the following steps: Step S31, the hidden danger automatic identification process is as follows: First, the new multi-source detection data is input into the trained deep learning model; second, the deep learning model automatically identifies and marks the potential danger areas; then, the output includes the location, type, size and distribution characteristics of the potential dangers; Step S32, data fusion and multi-dimensional interpretation process is as follows: First, a deep learning model is used to automatically identify multi-source detection data and mark potential danger areas. Second, the potential danger areas identified by multiple sources are superimposed and comprehensively interpreted through manual or machine weight models. Then, for potential dangers of different depths, the deep learning model distinguishes between shallow cracks and deep voids and gives corresponding risk assessments.
6. The method for interpreting the multi-source detection results of hidden dangers inside ancient buildings based on deep learning according to claim 5 is characterized in that: In step S4, multi-dimensional visualization analysis and decision support are performed based on the multi-dimensional interpretation results of step S3. The specific steps are as follows: Step S41: Based on the multi-dimensional interpretation result of step S3, the spatial distribution of hidden dangers is combined with the three-dimensional model of the city wall through three-dimensional modeling technology; the depth, size and material changes of hidden dangers are intuitively displayed by means of color gradient and transparency adjustment, and multi-dimensional visual analysis is performed; Step S42: Generate a detailed detection report based on the severity of the hidden danger, including the risk level of the hidden danger and repair suggestions; the report is output in Shapefile and GeoJSON standard formats for subsequent analysis and application.