Highway goaf space geometric feature recognition method based on subsidence monitoring

By setting up acquisition equipment and management systems in highway goafs, using YOLOX and convolutional neural networks for data processing and feature extraction, the problem of space geometric feature recognition in goafs is solved, automated monitoring and evaluation are realized, and identification is improved The accuracy and stability of the system can promptly detect safety hazards and reduce costs.

CN118982684APending Publication Date: 2024-11-19ZHENGZHOU UNIVERSITY OF LIGHT INDUSTRY +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411057042.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Priority Date
2023-08-21
Filing Date
2024-08-02
Publication Date
2024-11-19

AI Technical Summary

Technical Problem

The existing technology has failed to effectively solve the monitoring and identification of spatial geometric features of highway goaf, making it difficult to assess and manage safety risks.

Method used

Using a subsidence monitoring method, by setting up front-end acquisition equipment and central management system, the YOLOX automatic detection model and convolutional neural network are used to preprocess and feature extraction of historical data, build a space geometric feature recognition model for goaf space, and output real-time acquisition The results of spatial geometric features of empty areas are identified and their impact on highway safety are evaluated.

Benefits of technology

Implement automated monitoring and evaluation, avoid misjudgment caused by human factors, improve the accuracy and stability of identification, promptly detect safety hazards, reduce costs, and understand the spatial characteristics of goaf and the space characteristics of the goaf by combining historical and real-time data. Change trends improve the accuracy and reliability of identification.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118982684A_ABST
    Figure CN118982684A_ABST
Patent Text Reader

Abstract

The invention discloses a road goaf space geometric feature recognition method based on subsidence monitoring, and the method comprises the following steps: S1, setting front-end collection equipment in a monitoring range of a road goaf, and connecting the front-end collection equipment with a central management system through a network transmission system; s2, performing preprocessing and data quality evaluation on the acquired monitoring data, and storing the data in a central management system; s3, collecting historical subsidence monitoring data through the central management system, and performing geometric feature extraction on historical data through a YOLOX automatic detection model; s4, on the basis of a convolutional neural network, taking the extracted features as model parameters, and constructing a goaf space geometric feature recognition model; and S5, connecting the constructed recognition model with a central management system, and outputting a spatial geometric feature recognition result of the goaf in real time. According to the invention, automatic monitoring and evaluation can be realized, and possible potential safety hazards can be found in time.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of highway goaf monitoring, and in particular to a method for identifying spatial geometric features of highway goaf based on subsidence monitoring. Background Art

[0002] Highway goaf refers to the area where, in order to obtain sufficient basic construction conditions during highway construction, different excavation and cleaning measures are taken for the accumulation of sand, stone, soil and other materials formed by mining, earthwork and excavation, and they are regarded as a natural resource. In this process, since some materials cannot be used immediately, many holes and caves are left behind, forming highway goaf.

[0003] Common spatial forms of highway goafs include strips, rings, or geometric forms left in situ. Different development modes need to be adjusted appropriately according to the actual conditions of the goafs. The distribution of goaf elevations mainly refers to the height difference between its bottom and top, and this height difference also has a great impact on the visual effect and development and utilization of the goafs. There are many holes in the goafs, the geological conditions are complex, and the slope and collapse problems are also prominent.

[0004] Therefore, before development and utilization, the site safety risks must be carefully evaluated to ensure the safety of personnel and equipment. By monitoring the subsidence of highway goafs, changes and distortion inside the goafs can be discovered at the first time. According to the monitoring results, the goafs can be evaluated in time to ensure the safety of the goafs and to determine the geometric characteristics of the subsidence depth of the goafs. This information is of great significance for the planning, development, utilization and management of highway goafs. Therefore, it is extremely necessary to find the best development and utilization plan through in-depth understanding of the geometric characteristics of the goafs to improve the economic benefits of the goafs.

[0005] Currently, no effective solution has been proposed for the problems in the related technologies. Summary of the invention

[0006] In view of the shortcomings of the prior art, the present invention provides a method for identifying spatial geometric features of highway goaf areas based on subsidence monitoring. The specific technical solutions adopted by the present invention are as follows:

[0007] A method for identifying spatial geometric features of highway goaf areas based on subsidence monitoring, the identification method comprising the following steps:

[0008] S1. Set up front-end acquisition equipment within the monitoring range of the highway goaf area and connect it to the central management system through the network transmission system;

[0009] S2. Preprocess and evaluate the data quality of the collected monitoring data and store them in the central management system;

[0010] S3. Collect historical subsidence monitoring data through the central management system and extract geometric features of historical data through the YOLOX automatic detection model;

[0011] S4. Based on the convolutional neural network, the extracted features are used as model parameters to build a spatial geometric feature recognition model for goaf areas;

[0012] S5. Connect the constructed recognition model with the central management system to output the spatial geometric feature recognition results of the goaf in real time;

[0013] S6. Based on the identification results, evaluate the impact of the goaf on highway safety and provide feedback to the user.

[0014] Preferably, the preprocessing and data quality assessment of the collected monitoring data and storing them in the central management system include the following steps:

[0015] S21, separating the video data from the text data in the collected monitoring data, and converting the video data into image data;

[0016] S22, processing the text data for data redundancy, data anomaly, and data missing;

[0017] S23. Evaluate the quality of text data based on data integrity, spatial morphology accuracy, goaf height and depth accuracy combined with image data;

[0018] S24, preprocessing the image data;

[0019] S25. Store the text data and image data that have passed the evaluation into the central management system.

[0020] Preferably, the processing of data redundancy, data anomaly and data missing on the text data comprises the following steps:

[0021] S221, taking the monitoring data at the same horizontal position as a processing unit, and extracting the data processing unit to be processed by setting two characteristic attributes;

[0022] S222, arranging the monitoring data in the data processing unit in chronological order, and determining the size of the sliding time window;

[0023] S223, starting from the first time window, taking the spatial form, height and depth of the goaf as feature values, determining whether there is the same data in the same time window, and if there is the same record, selecting the first record as the parameter of the feature value, and deleting the duplicate record;

[0024] S224, after the duplicate value processing is completed, the manifestation type of the abnormal value and the characteristic variable thereof are determined according to the spatial form, height and depth of the goaf;

[0025] S225. Retrieve the record position where the abnormal value is located by traversing the feature variables, delete the entire record by index, and fill in the missing values ​​in the monitoring data by constructing multiple decision trees.

[0026] Preferably, the preprocessing of the image data comprises the following steps:

[0027] S241, extracting color factors of the original image data by a component method, and extracting the maximum value of the color image pixel points;

[0028] S242, extracting the average value of the color image by using the average value method;

[0029] S243, converting the color image into a gray image using a weighted average method;

[0030] S244, reducing noise on the gray image by using a mean algorithm;

[0031] S245, obtaining pixel neighborhood information of the gray image through a Gaussian bilateral filter function, and using a quadratic Taylor series convolution to adaptively enhance the image edge.

[0032] Preferably, the pixel neighborhood information of the gray image is obtained by using the Gaussian bilateral filter function, and the pixel neighborhood information calculation formula in the adaptive enhancement of the image edge by using the quadratic Taylor series convolution is:

[0033]

[0034] In the formula, η represents the image factor, m and t represent the grayscale value of the image, n and a represent the image edge noise queue, θ i With θ j represents the difference of neighborhood information, and λ represents the RGB value of the image in color mode.

[0035] Preferably, the collecting of historical subsidence monitoring data through the central management system and extracting geometric features of the historical data through the YOLOX automatic detection model comprises the following steps:

[0036] S31, in the historical subsidence data, according to the label library and the sample library, expanding the data image in the sample library and saving it in a picture format;

[0037] S32. Construct an improved YOLOX automatic detection model based on the label library and sample library;

[0038] S33. Extract geometric feature parameters based on the YOLOX automatic detection model.

[0039] Preferably, extracting geometric feature parameters according to the YOLOX automatic detection model includes the following steps:

[0040] S331, detecting the geometric center of the circular frame of the goaf depth area according to the YOLOX automatic detection model as the center position of the subsidence area;

[0041] S332, binarizing, skeletonizing and removing interfering depth regions of the detected image to obtain edge position information;

[0042] S333. Calculate the minimum circumscribed circle of the edge position, use the radius of the circumscribed circle as the depth dimension, and obtain the geometric feature parameters in the historical data.

[0043] Preferably, the edge position information obtained by binarizing, skeletonizing and removing interfering depth regions from the detected image comprises the following steps:

[0044] S3321, using an adaptive threshold segmentation method to binarize the detected image, and convert the image into black and white areas, where the black area corresponds to the flat area of ​​the goaf, marked as 1, and the white area corresponds to the subsidence area of ​​the goaf, marked as 2;

[0045] S3322, counting the number of pixels in each area of ​​the image marked as 2, setting a default threshold value of 80, and if it is less than 80, marking the area as 1 and turning it into a black area;

[0046] S3323, performing morphological filling processing on the flat area of ​​the goaf surrounded by the subsidence area of ​​the goaf to change it into the flat area of ​​the goaf, extracting the outline skeleton line of the subsidence area, and performing burr removal processing on the skeleton line;

[0047] S3324, calculating all connected regions in the image, and obtaining coordinate traction of each group of connected regions;

[0048] S3325. Calculate the number of pixels in all connected areas, select the longest arc segment of the main connected area, fit the selected main connected area, and obtain the subsidence edge line.

[0049] Preferably, the method of constructing a goaf spatial geometric feature recognition model based on a convolutional neural network and using the extracted features as model parameters comprises the following steps:

[0050] S41, in a convolutional neural network, obtaining an objective function of the image according to the extracted feature parameters;

[0051] S42, after obtaining the objective function of the image, calculating the optimal solution of the objective function;

[0052] S43. After the calculation is completed, the geometric feature recognition model is constructed.

[0053] Preferably, after obtaining the objective function of the image, the calculation formula for calculating the optimal solution of the objective function is:

[0054]

[0055] In the formula, h represents the number of images, x k Indicates the RGB value of the image in color mode, D s represents the activation function, f i Represents the image output value, g s represents a sparse matrix, Denotes the data enhancement coefficient, D′ s represents the derivative of the activation function.

[0056] Compared with the prior art, the present invention provides a method for identifying spatial geometric features of highway goaf areas based on subsidence monitoring, which has the following beneficial effects:

[0057] (1) The present invention is based on feature extraction of historical subsidence monitoring data, finds the standards of spatial geometric features of highway goaf areas from historical data, connects and identifies real-time goaf area data with geometric features, and verifies areas that meet the characteristics of subsidence areas. The use of computer vision technology for geometric feature recognition can avoid misjudgment caused by human factors, improve accuracy and stability, realize automated monitoring and evaluation, effectively reduce costs, timely discover possible safety hazards, and warn relevant personnel to avoid potential accident risks.

[0058] (2) The present invention extracts historical subsidence areas and connects and identifies them with real-time monitoring data, so that historical subsidence areas can provide an important data basis for real-time monitoring. By extracting historical data and combining them with on-site monitoring data, the spatial characteristics of the goaf can be more comprehensively understood and mastered. The data of the historical subsidence area has a high reference value and can identify the spatial characteristics and change trends of the goaf on a long-term time scale. By combining historical data with real-time monitoring data, the accuracy, stability and reliability of identification can be improved.

[0059] (3) The present invention first processes the data of the historical settlement area and the real-time monitoring data, and performs preprocessing, filtering and denoising operations on the two groups of images, which can effectively improve the image quality, clarity and contrast, facilitate the extraction of required information and eliminate redundant information, so as to more effectively identify the image content and features. By processing the image, it is easier for the identifier to identify the target object or area, thereby improving the recognition rate and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0061] Figure 1 The present invention is a flowchart of a method for identifying spatial geometric features of highway goaf areas based on subsidence monitoring according to an embodiment of the present invention. DETAILED DESCRIPTION

[0062] To further illustrate each embodiment, the present invention provides drawings, which are part of the disclosure of the present invention and are mainly used to illustrate the embodiments. They can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these contents, ordinary technicians in this field should be able to understand other possible implementation methods and advantages of the present invention.

[0063] According to an embodiment of the present invention, a method for identifying spatial geometric features of highway goaf areas based on subsidence monitoring is provided.

[0064] The present invention is further described with reference to the accompanying drawings and specific embodiments. Figure 1 As shown, according to the method for identifying spatial geometric features of highway goaf areas based on subsidence monitoring according to an embodiment of the present invention, the identification method comprises the following steps:

[0065] S1. Set up front-end data collection equipment within the monitoring range of the highway goaf area and connect it to the central management system through a network transmission system.

[0066] In one embodiment, the front-end acquisition equipment is installed and connected within the monitoring range of the highway goaf area, including the following steps:

[0067] According to specific monitoring requirements and environmental conditions, select suitable front-end acquisition equipment, such as sensors, cameras, etc.

[0068] According to the characteristics of the collection equipment and monitoring requirements, the equipment is arranged and installed at the monitoring point. This requires site survey and analysis, and arrangement in combination with actual operations.

[0069] The data collected by the equipment is connected to the central management system through a network transmission system to ensure real-time transmission and processing of monitoring data. The network transmission system can use a wired network or a wireless network, depending on the specific situation.

[0070] After the connection is completed, a certain degree of testing is required to ensure that the network connection between devices is normal, the transmitted data is accurate, the transmission rate is stable, and other aspects meet the requirements.

[0071] Establish a central management system to manage and process the data collected by the front-end acquisition equipment for subsequent analysis and application.

[0072] S2. Preprocess and evaluate the data quality of the collected monitoring data and store them in the central management system.

[0073] In one embodiment, the preprocessing and data quality assessment of the collected monitoring data and storing them in the central management system include the following steps:

[0074] S21, separating the video data from the text data in the collected monitoring data, and converting the video data into image data.

[0075] Specifically, separating the video data from the text data in the collected monitoring data and converting the video data into image data includes the following steps:

[0076] The collected monitoring data needs to be read, including video data and text data, which are stored in the form of files.

[0077] The video data and text data in the read monitoring data can be separated by corresponding programs, tools or manual methods and stored separately.

[0078] Converting video data into images requires video decoding, which restores the encoded video data into video frame data according to certain rules. This step can be processed using a professional video decoder or corresponding decoding library.

[0079] After obtaining the video frame data, it is necessary to further convert the video frame into image data.

[0080] The processed image data is stored and related processing operations are performed.

[0081] S22. Process the text data for data redundancy, data anomaly, and data missing.

[0082] Specifically, the data redundancy, data anomaly and data missing processing of the text data includes the following steps:

[0083] S221, taking the monitoring data at the same horizontal position as a processing unit, and extracting the data processing unit to be processed by setting two characteristic attributes;

[0084] S222, arranging the monitoring data in the data processing unit in chronological order, and determining the size of the sliding time window;

[0085] S223, starting from the first time window, taking the spatial form, height and depth of the goaf as feature values, determining whether there is the same data in the same time window, and if there is the same record, selecting the first record as the parameter of the feature value, and deleting the duplicate record;

[0086] S224, after the duplicate value processing is completed, the manifestation type of the abnormal value and the characteristic variable thereof are determined according to the spatial form, height and depth of the goaf;

[0087] S225. Retrieve the record position where the abnormal value is located by traversing the feature variables, delete the entire record by index, and fill in the missing values ​​in the monitoring data by constructing multiple decision trees.

[0088] S23. Evaluate the quality of text data based on data integrity, spatial morphology accuracy, goaf height and depth accuracy combined with image data;

[0089] Specifically, by evaluating text data with image data, we can consider three aspects: data integrity, spatial morphology accuracy, and goaf height and depth accuracy.

[0090] By comparing the geographic information in the text data with the image data, we can detect missing and incomplete data. If we find that important information in the text data is not covered, we need to consider re-collecting or supplementing relevant information. At the same time, through the panoramic inspection and rapid indexing of image data, we can also better understand the actual effect and existing problems of the sampling area.

[0091] The accuracy of the spatial form of the monitoring data is evaluated by comparing the accuracy of the image data with the spatial data of the collected map. If there is a significant deviation in the image data, it may indicate that there is deformation or noise introduced when the text data is collected.

[0092] By comparing the coordinates and elevation information in the image data with the underground model data, we can confirm whether the height and depth of the goaf are within the specified range. If obvious anomalies are found, reprocessing or resampling may be required.

[0093] S24, pre-processing the image data.

[0094] Specifically, the preprocessing of the image data includes the following steps:

[0095] S241, extracting color factors of the original image data by a component method, and extracting the maximum value of the color image pixel points;

[0096] S242, extracting the average value of the color image by using the average value method;

[0097] S243, converting the color image into a gray image using a weighted average method;

[0098] S244, reducing noise on the gray image by using a mean algorithm;

[0099] S245, obtaining pixel neighborhood information of the gray image through a Gaussian bilateral filter function, and using a quadratic Taylor series convolution to adaptively enhance the image edge.

[0100] The pixel neighborhood information of the gray image is obtained by using the Gaussian bilateral filter function, and the pixel neighborhood information calculation formula in the adaptive enhancement of the image edge by using the quadratic Taylor series convolution is:

[0101]

[0102] In the formula, η represents the image factor, m and t represent the grayscale value of the image, n and a represent the image edge noise queue, θ i With θ j represents the difference of neighborhood information, and λ represents the RGB value of the image in color mode.

[0103] The calculation formula for adaptively enhancing the image edge using quadratic Taylor series convolution is:

[0104]

[0105] In the formula, Q q With Q p They represent the mean brightness enhancement of the edge pixels of the image after the quadratic Taylor series expansion convolution, n and a represent the image edge noise queue, and z represents the image after filtering and denoising.

[0106] S25. Store the text data and image data that have passed the evaluation into the central management system.

[0107] S3. Collect historical subsidence monitoring data through the central management system, and extract geometric features of historical data through the YOLOX automatic detection model.

[0108] In one embodiment, the collecting of historical subsidence monitoring data through the central management system and the extraction of geometric features of the historical data through the YOLOX automatic detection model include the following steps:

[0109] S31. In the historical subsidence data, according to the label library and the sample library, the data images in the sample library are expanded and saved in a picture format.

[0110] Specifically, the historical subsidence data is expanded and saved as PNG images according to the label library and the sample library, including the following steps:

[0111] Read data images in all sample libraries from historical subsidence data.

[0112] Collect labels based on the label library you have and classify them according to the characteristics of the target image. Arrange and sort images of the same category to expand the sample library and finally save them.

[0113] The model is learned using machine learning or computer vision algorithms to train a generative model suitable for use with individual historical subsidence data.

[0114] Generate new image data through the generative model, generate corresponding image samples for each label, and add them to the sample library.

[0115] Save all the expanded image data in the sample library in PNG image format. After completing the above steps, please check again whether all the saved PNG image files meet the expected effects.

[0116] S32. Build an improved YOLOX automatic detection model based on the label library and sample library.

[0117] In one embodiment, the construction of an improved YOLOX automatic detection model based on a tag library and a sample library includes the following steps:

[0118] Input sample image data and extract the subsidence features as the main features;

[0119] Add the Fcous structure to the backbone feature extraction network, replace the activation function with the SiLU function, and perform spatial pyramid pooling;

[0120] The PAFPN structure is used for fusion, and the high-level feature information is first transferred and fused through upsampling, and then the predicted feature map is obtained through downsampling fusion.

[0121] The effective feature layer extracted by the backbone layer is input into the PAFPN enhanced feature extraction network for feature fusion to obtain an enhanced feature layer. The enhanced feature layer uses YOLO-Head to determine whether the feature points are in the sinking area, and uses non-maximum suppression to eliminate duplicate bounding boxes to obtain more accurate detection results.

[0122] Specifically, YOLOX is an automatic detection model based on the target detection algorithm. It performs convolution operations on the input image and uses deep neural networks (DNN) to achieve efficient and accurate object detection.

[0123] S33. Extract geometric feature parameters based on the YOLOX automatic detection model.

[0124] Specifically, the extraction of geometric feature parameters according to the YOLOX automatic detection model includes the following steps:

[0125] S331, detecting the geometric center of the circular frame of the goaf depth area according to the YOLOX automatic detection model as the center position of the subsidence area;

[0126] S332, binarizing, skeletonizing and removing interfering depth regions on the detected image to obtain edge position information.

[0127] The edge position information obtained by binarizing, skeletonizing and removing the interfering depth area of ​​the detected image includes the following steps:

[0128] S3321, using an adaptive threshold segmentation method to binarize the detected image, and convert the image into black and white areas, where the black area corresponds to the flat area of ​​the goaf, marked as 1, and the white area corresponds to the subsidence area of ​​the goaf, marked as 2;

[0129] S3322, counting the number of pixels in each area of ​​the image marked as 2, setting a default threshold value of 80, and if it is less than 80, marking the area as 1 and turning it into a black area;

[0130] S3323, performing morphological filling processing on the flat area of ​​the goaf surrounded by the subsidence area of ​​the goaf to change it into the flat area of ​​the goaf, extracting the outline skeleton line of the subsidence area, and performing burr removal processing on the skeleton line;

[0131] S3324, calculating all connected regions in the image, and obtaining coordinate traction of each group of connected regions;

[0132] S3325. Calculate the number of pixels in all connected areas, select the longest arc segment of the main connected area, fit the selected main connected area, and obtain the subsidence edge line.

[0133] S333. Calculate the minimum circumscribed circle of the edge position, use the radius of the circumscribed circle as the depth dimension, and obtain the geometric feature parameters in the historical data.

[0134] S4. Based on the convolutional neural network, the extracted features are used as model parameters to construct a spatial geometric feature recognition model for goaf areas.

[0135] In one embodiment, the method of constructing a goaf spatial geometric feature recognition model based on a convolutional neural network and using the extracted features as model parameters includes the following steps:

[0136] S41, in a convolutional neural network, obtaining an objective function of the image according to the extracted feature parameters;

[0137] S42: After obtaining the objective function of the image, calculate the optimal solution of the objective function.

[0138] Specifically, after obtaining the objective function of the image, the calculation formula for calculating the optimal solution of the objective function is:

[0139]

[0140] In the formula, h represents the number of images, x k Indicates the RGB value of the image in color mode, D s represents the activation function, f i Represents the image output value, g s represents a sparse matrix, Denotes the data enhancement coefficient, D′ s represents the derivative of the activation function.

[0141] S43. After the calculation is completed, the geometric feature recognition model is constructed.

[0142] Specifically, the constructed recognition model is:

[0143]

[0144] In the formula, h represents the number of images, b i Indicates the RGB value of the image in color mode, D s represents the activation function, g s represents a sparse matrix, n and a represent the image edge noise queue, Represents the data enhancement coefficient, i and s are non-zero natural numbers.

[0145] S5. Connect the constructed recognition model with the central management system to output the spatial geometric feature recognition results of the goaf in real time.

[0146] In one embodiment, the constructed recognition model is connected to the central management system, and the real-time output of the spatial geometric feature recognition results of the goaf includes the following steps:

[0147] First, you need to save the built recognition model into a common deployable format so that it can be easily run on other devices later.

[0148] Depending on the actual scenario, the recognition model is deployed to the cloud or local server, and the relevant service interfaces and authentication permissions are opened so that external systems can call it through the API.

[0149] In order to output the spatial geometric feature recognition results of the goaf in real time, we need to connect the previously built central management system with the server where the model is deployed, and set the corresponding data transmission protocol and parameter specifications.

[0150] Within the central management system, the input data is processed and formatted, and then sent to the server where the recognition model is located through a network request to obtain the recognition results. After further sorting and processing these recognition results, the spatial geometric feature recognition results of the goaf can be output in real time.

[0151] S6. Based on the identification results, evaluate the impact of the goaf on highway safety and provide feedback to the user.

[0152] Based on the recognition results between the real-time collected data and the historical features, the difference between the real-time data and the historical data can be identified, so as to judge the scale and depth of the goaf at that time, thereby realizing the judgment of highway safety. The user end performs corresponding operations on the highway area according to the judgment results to avoid the goaf being too large or too deep, which may lead to ground subsidence, collapse and other problems, thus posing a threat to highway safety.

[0153] In summary, with the help of the above technical scheme of the present invention, the present invention is based on feature extraction of historical subsidence monitoring data, finds the standard of spatial geometric features of highway goaf from historical data, and connects and identifies the real-time goaf data with geometric features, verifies the area that meets the characteristics of the subsidence area, and uses computer vision technology to identify geometric features to avoid misjudgment caused by human factors, improve accuracy and stability, and can realize automatic monitoring and evaluation, effectively reduce costs, timely discover possible safety hazards, and warn relevant personnel to avoid potential accident risks. The present invention extracts historical subsidence areas and connects and identifies with real-time monitoring data, so that the historical subsidence area can provide an important data basis for real-time monitoring. By extracting historical data and combining it with on-site monitoring data, the spatial characteristics of the goaf can be more comprehensively understood and mastered, and the data of the historical subsidence area has a high reference value, and the spatial characteristics and changing trends of the goaf can be identified on a long-term time scale. By combining historical data with real-time monitoring data, the accuracy, stability and reliability of identification can be improved. The present invention first processes the data of historical settlement areas and real-time monitoring data, and performs preprocessing, filtering and denoising operations on the two groups of images, which can effectively improve the image quality, clarity and contrast, facilitate the extraction of required information and eliminate redundant information, so as to more effectively identify the image content and features. By processing the image, it is easier for the identifier to identify the target object or area, thereby improving the recognition rate and accuracy.

[0154] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for identifying spatial geometric features of highway goaf areas based on subsidence monitoring, characterized in that: The identification method comprises the following steps: S1. Set up front-end acquisition equipment within the monitoring range of the highway goaf area and connect it to the central management system through the network transmission system; S2. Preprocess and evaluate the data quality of the collected monitoring data and store them in the central management system; S3. Collect historical subsidence monitoring data through the central management system and extract geometric features of historical data through the YOLOX automatic detection model; S4. Based on the convolutional neural network, the extracted features are used as model parameters to construct a spatial geometric feature recognition model for goaf areas; S5. Connect the constructed recognition model with the central management system to output the spatial geometric feature recognition results of the goaf in real time; S6. Based on the identification results, evaluate the impact of the goaf on highway safety and provide feedback to the user.