Piping dangerous case troubleshooting method and system based on image recognition

Abstract

The invention relates to the technical field of image recognition, and particularly discloses a piping dangerous case troubleshooting method and system based on image recognition, and the method comprises the steps: obtaining and analyzing dam hidden danger data and historical piping data, determining the initial type hidden danger data, the risk distance threshold value and the type potential risk region of each hidden danger type label, and determining the type potential risk region of each hidden danger type label; determining an initial potential risk area of the dam to be checked and optimizing the initial potential risk area; determining isolated piping data, a plurality of troubleshooting type hidden danger data, a plurality of merging type data and a dam potential risk area; constructing a plurality of piping dangerous case models; dam image monitoring data and dam environment monitoring data are collected; and dam dangerous case data are determined. The method can comprehensively cover hidden hidden dangers, special risks and isolated piping, avoids risk omission, accurately adapts to single and composite hidden danger scenes, improves the pertinence of dangerous case identification, strengthens the relevance and reliability of dangerous case judgment, improves the troubleshooting efficiency and precision, and achieves the global dead-corner-free prevention and control of piping dangerous cases.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and in particular to a method and system for investigating piping hazard based on image recognition. Background Technology

[0002] Traditional piping hazard detection relies on manual inspections and single-sensor monitoring. Manual inspections are inefficient, labor-intensive, and prone to omissions and misjudgments due to experience levels and harsh environments. Single sensors can only acquire localized data, making it difficult to cover the entire risk area and insufficient for detecting hidden hazards and early signs of piping. Although current piping hazard detection methods have incorporated image recognition, multiple sensors, and intelligent algorithms, they suffer from low efficiency in multi-source data fusion, models not adapted to complex hazard scenarios, and a lack of dynamic optimization of risk areas based on historical data. Isolated piping and hidden hazards are easily missed, and the false alarm rate is high under complex operating conditions, making it difficult to meet the needs for accurate and efficient piping detection.

[0003] Therefore, this invention proposes a method and system for investigating piping hazard based on image recognition. Summary of the Invention

[0004] This invention provides a method and system for investigating piping hazards based on image recognition. By analyzing acquired dam hazard data and historical piping data of the dam to be investigated, the initial type hazard data, risk distance threshold, and potential risk area for each hazard type label are determined. The initial potential risk area of ​​the dam to be investigated is identified and optimized. Isolated piping data, investigation type hazard data for multiple hazard type labels, and merged type data for multiple combined type labels are identified. The potential risk area of ​​the dam to be investigated is determined. Piping hazard models for each hazard type label and each combined type label are constructed. Dam image monitoring data and dam environmental monitoring data are collected. Hazard investigation is conducted on the dam to be investigated to determine dam hazard data. This method can comprehensively cover conventional hazards, special risks, and isolated piping, avoiding risk omissions. It accurately adapts to single and complex hazard scenarios, improves the targeting of hazard identification, strengthens the correlation and reliability of hazard judgment, improves investigation efficiency and accuracy, and achieves comprehensive and seamless prevention and control of piping hazards.

[0005] This invention provides a method for detecting piping hazard based on image recognition, comprising: S1: Acquire and analyze the dam hazard data and historical piping data of the dam to be investigated, determine the initial type hazard data, risk distance threshold and type potential risk area for each hazard type label, and determine the initial potential risk area of ​​the dam to be investigated. S2: Based on the historical piping data of the embankment to be investigated, the initial potential risk area is optimized, and isolated piping data, investigation type hazard data with multiple hazard type labels, and merged type data with multiple merged type labels are determined, and the potential risk area of ​​the embankment to be investigated is determined. S3: Based on the investigation type hazard data of each hazard type label of the dam to be investigated, the merged type data of each merged type label, and the historical piping data, construct the piping hazard model for each hazard type label, and construct the piping hazard model for each merged type label. S4: Collect image monitoring data of the dam to be investigated based on image sensors, and collect environmental monitoring data of the dam to be investigated based on environmental sensor groups; S5: Based on the piping hazard model of all hidden danger types labeled on the dike to be investigated, the piping hazard model of all combined type labels, the dike image monitoring data, and the dike environmental monitoring data, conduct hazard investigation on the dike to be investigated and determine the dike hazard data.

[0006] Preferably, an image recognition-based method for investigating piping hazards acquires and analyzes dam hazard data and historical piping data for the dam to be investigated, determines the initial hazard data, risk distance threshold, and potential risk area for each hazard type label, and determines the initial potential risk area of ​​the dam to be investigated, including: Acquire data on potential hazards and historical piping data for the dikes to be investigated. The dike hazard data includes multiple locations of hazards on the dike body, as well as hazard type and hazard severity labels for each location. The historical piping data includes sub-data on multiple piping events, which includes piping location, dike body section, dike station number, historical piping time, piping type label, hazard level label, historical environmental vector, and historical image recognition vector. The hazard level labels include Level 1, Level 2, Level 3, and Level 4. Based on the hazard type labels of all hazard locations in the dam hazard data, all hazard locations in the dam body are divided, and the initial type hazard data for each hazard type label is determined. The initial type hazard data includes multiple hazard locations in the dam body and the hazard degree label for each hazard location. Based on the initial type hazard data of each hazard type label of the dam to be investigated, as well as the piping location, historical piping time, and hazard level label in all historical piping sub-data in the historical piping data, calculate the risk distance threshold of each hazard type label of the dam to be investigated. Using the location of each hazard in the initial type hazard data of each hazard type label of the dam to be investigated as the center and the risk distance threshold of each hazard type label of the dam to be investigated as the radius, the potential risk sub-region of each hazard location in the initial type hazard data of each hazard type label of the dam to be investigated is determined. Based on the potential risk sub-regions of all hazard locations in the initial type hazard data of each hazard type label of the dam to be investigated, the type potential risk area of ​​each hazard type label of the dam to be investigated is determined. Based on the type potential risk areas of all hazard type labels of the dam to be investigated, the initial potential risk area of ​​the dam to be investigated is determined.

[0007] Preferably, an image recognition-based method for investigating piping risks optimizes the initial potential risk area based on historical piping data of the dam to be investigated, including: Based on the piping locations of all historical piping sub-data in the historical piping data and the initial potential risk area of ​​the embankment to be investigated, determine whether there are piping locations that do not belong to the initial potential risk area, and determine the initial isolated data based on the historical piping sub-data of all piping locations that do not belong to the initial potential risk area. Determine whether the number of historical piping sub-data in the initial isolated data is below a preset percentage threshold. If so, determine that the initial isolated data is isolated piping data. Otherwise, optimize the risk distance threshold for each hazard type label of the embankment to be investigated. Based on the optimized risk distance thresholds for all hazard type labels and the location of each hazard in the embankment body in the initial hazard data for each hazard type label, determine the optimized initial potential risk area and identify the isolated piping data.

[0008] Preferably, an image recognition-based method for investigating piping hazards identifies isolated piping data, hazard data with multiple hazard type labels, and merged type data with multiple merged type labels, and determines the potential risk area of ​​the dam to be investigated, including: The largest potential risk sub-region in the initial potential risk area is selected as the implicit judgment area for isolated piping data; Based on the piping locations of all historical piping sub-data in the implicit judgment area and isolated piping data, determine whether a implicit judgment area has a preset number or more piping locations. If it does, determine each implicit judgment area with a preset number or more piping locations as an implicit investigation area, and determine the historical piping sub-data of all piping locations that do not belong to any implicit judgment area as the adjusted isolated piping data. If it does not, do not adjust the isolated piping data. For each hidden investigation area, a hazard investigation is carried out. If a hazard is found, the location of the hazard on the embankment, the hazard type label, and the hazard severity label of the hidden investigation area are determined. If no hazard is found, the hidden investigation area is determined to be a special risk sub-area. Determine whether there is any overlap between all potential risk sub-regions and all hidden investigation areas where hazards have been discovered in the initial potential risk area. For each potential risk sub-region and hidden investigation area with regional overlap, merge them to determine each potential merged sub-region. Then, determine the merge type label of each potential merged sub-region based on the hazard type labels corresponding to the two or more potential risk sub-regions and hidden investigation areas with regional overlap. Based on the initial type hazard data of each hazard type label, all potential risk sub-regions without regional overlap and all hidden investigation areas where hazards are found without regional overlap, the location of hazard in the dike body, hazard type label, and hazard severity label are determined, and the investigation type hazard data of each hazard type label of the dike to be investigated is determined. The investigation type hazard data includes multiple hazard locations in the dike body and the hazard severity label of each hazard location. Based on the merging type label of each potential merging sub-region, all potential merging sub-regions are divided, and the merging type data of each merging type label is determined. The merging type data includes multiple potential merging sub-regions, multiple embankment hazard locations in each potential merging sub-region, and hazard degree labels for each embankment hazard location. Based on all potential risk sub-regions without regional overlap in the initial potential risk area, all hidden investigation areas without regional overlap where hidden dangers were found, all special risk sub-regions where no hidden dangers were found, and all potential merged sub-regions, the potential risk areas of the dikes to be investigated are determined.

[0009] Preferably, an image recognition-based method for investigating piping hazards constructs a piping hazard model for each hazard type label, and a piping hazard model for each merged type label, based on the hazard type data for each hazard type label of the dike to be investigated, the merged type data for each merged type label, and historical piping data. The method includes: Based on the potential risk sub-area or hidden investigation area of ​​each hazard location in the hazard data of each hazard type label of the dam to be investigated, and the piping location of all historical piping sub-data in the historical piping data, the historical hazard piping data of each hazard location in the dam to be investigated is determined. The historical environment vector and historical image recognition vector of all historical piping sub-data in the historical hidden danger piping data of all potential risk sub-regions or hidden investigation areas of all hidden danger locations in the investigation type data of each hidden danger type label of the embankment to be investigated are used as the input of the piping hazard model of each hidden danger type label. The piping location, piping type label, and hazard level label of all historical hidden danger piping sub-data in the historical hidden danger piping data of all potential risk sub-regions or hidden investigation areas of all hidden danger locations in the investigation type data of each hidden danger type label of the embankment to be investigated are used as the output of the piping hazard model of each hidden danger type label. Thus, the piping hazard model of each hidden danger type label is constructed. Based on the potential risk sub-regions or hidden investigation areas of all potential hazard locations in the merging type data of each potential merging sub-region of each merging type label of the dam to be investigated, and the piping locations of all historical piping sub-data in the historical piping data, the historical merged piping data of each potential merging sub-region in the merging type data of each merging type label of the dam to be investigated is determined. The piping hazard model for each hazard type label is constructed by taking the locations of all potential hazard locations in the merging sub-regions of the merging type data for each dam to be investigated, the historical environmental vectors of all historical piping sub-data in the historical merging piping data of all potential merging sub-regions, and the historical image recognition vectors as inputs. The piping hazard model for each hazard type label is constructed by taking the piping location, piping type label, and hazard level label of all historical piping sub-data in the historical merging piping data of all potential merging sub-regions of the merging type data for each dam to be investigated as outputs.

[0010] Preferably, a method for investigating piping hazards based on image recognition includes: collecting dam image monitoring data of the dam to be investigated based on image sensors, and collecting dam environmental monitoring data of the dam to be investigated based on an environmental sensor array, comprising: Based on image sensors, regional image monitoring data of each sub-region in the potential risk area of ​​the dam to be investigated is collected. The regional image monitoring data includes multiple monitoring images and the monitoring position and shooting angle of each monitoring image. The sub-regions are: potential risk sub-region, hidden investigation area where hidden dangers are found, special risk sub-region, and potential merged sub-region. Regional environmental monitoring data for each sub-region within the potential risk area of ​​the dam to be investigated is collected based on an environmental sensor array. Isolated image monitoring data for each piping location in the isolated piping data of the embankment to be investigated is collected based on image sensors. The isolated image monitoring data includes multiple monitoring images and the monitoring location and shooting angle of each monitoring image. Isolated environmental monitoring data for each piping location in the isolated piping data of the embankment to be investigated, collected from the environmental sensor array; Based on the regional image monitoring data of all sub-regions in the potential risk area of ​​the dam to be investigated and the isolated image monitoring data of all piping locations in the isolated piping data, the dam image monitoring data of the dam to be investigated is determined. Based on the regional environmental monitoring data of all sub-regions in the potential risk area of ​​the dam to be investigated, and the isolated environmental monitoring data of all piping locations in the isolated piping data, the dam environmental monitoring data of the dam to be investigated is determined.

[0011] Preferably, an image recognition-based method for investigating piping vulnerabilities involves using piping vulnerabilities modeled with all hazard type labels for the dike to be investigated, piping vulnerabilities modeled with all combined type labels, dike image monitoring data, and dike environmental monitoring data to investigate the dike and determine the dike hazard data, including: All monitoring images in the regional image monitoring data of each sub-region of the potential risk area of ​​the dam to be investigated are preprocessed. All preprocessed monitoring images with the same dam chainage in the regional image monitoring data of each sub-region are image aligned to determine the monitoring chainage image set of each dam chainage in each sub-region of the potential risk area of ​​the dam to be investigated. Feature extraction is performed on all monitoring images in the monitoring chainage image set of each chainage of each sub-region in the potential risk area of ​​the dam to be investigated, and the chainage image recognition vector of each chainage of each chainage in each sub-region in the potential risk area of ​​the dam to be investigated is determined. Feature extraction is performed on the environmental monitoring data of each sub-region within the potential risk area of ​​the dam to be investigated, and the regional environmental vector of each sub-region within the potential risk area of ​​the dam to be investigated is determined. If the sub-region is a potential risk sub-region or a hidden investigation area where hidden dangers have been discovered, the regional environmental vector of the sub-region and the set of monitoring station images of all dike station numbers are input into the piping hazard model of the hidden danger type label of the potential risk sub-region or the hidden investigation area where hidden dangers have been discovered. Based on the piping hazard model, the regional hazard data of the sub-region is determined. The regional hazard data includes hazard prediction labels and multiple predicted piping locations when the hazard prediction labels exist, the predicted piping type of each predicted piping location, and the predicted hazard level. If the sub-region is a potential merged sub-region, the regional environment vector of the sub-region and the set of monitoring station images of all dam station numbers are input into the piping hazard model of the merging type label of the potential merged sub-region corresponding to the sub-region, and the regional hazard data of the sub-region is determined based on the piping hazard model. If the sub-region is a special risk sub-region, based on the regional environment vector of the sub-region and the monitoring station image set of all dam station numbers, manual hazard identification is performed to determine the regional hazard data of the sub-region; Based on the dike body part, isolated image monitoring data, isolated environmental monitoring data, and the dike body part, piping type label, hazard level label, and historical environmental vector of all historical piping sub-data corresponding to all piping locations in all isolated piping data, the predicted hazard label and isolated hazard data for each isolated piping are determined. The predicted hazard label includes whether the hazard exists or not. If the predicted hazard label is "hazard exists", the isolated hazard data includes the predicted piping type and the predicted hazard level; otherwise, the isolated hazard data is empty. Based on the regional hazard data of all sub-regions in the potential risk area of ​​the dam to be investigated, as well as the predicted hazard labels and isolated hazard data of all piping locations in the isolated piping data, the dam hazard data of the dam to be investigated is determined.

[0012] This invention provides a method and system for detecting piping hazards based on image recognition, used to execute any one of the image recognition-based piping hazard detection methods in Examples 1 to 7, including: Acquisition Module: Acquires and analyzes the dam hazard data and historical piping data of the dam to be investigated, determines the initial type hazard data, risk distance threshold and type potential risk area for each hazard type label, and determines the initial potential risk area of ​​the dam to be investigated. Determine the module: Based on the historical piping data of the embankment to be investigated, optimize the initial potential risk area, determine isolated piping data, investigation type hazard data with multiple hazard type labels, and merged type data with multiple merged type labels, and determine the potential risk area of ​​the embankment to be investigated. Construction module: Based on the investigation type hazard data of each hazard type label of the dam to be investigated, the combined type data of each combined type label, and the historical piping data, construct the piping hazard model for each hazard type label, and construct the piping hazard model for each combined type label. Acquisition module: Acquires dam image monitoring data based on image sensors, and acquires dam environmental monitoring data based on environmental sensor arrays; Hazard Module: Based on the piping hazard model with all hidden danger type labels of the dike to be investigated, the piping hazard model with all combined type labels, dike image monitoring data, and dike environmental monitoring data, the module conducts hazard investigation on the dike to be investigated and determines the dike hazard data.

[0013] The beneficial effects of this invention compared to existing technologies are as follows: By analyzing the acquired data on dam hazards and historical piping data of the dams to be investigated, the initial type of hazard data, risk distance threshold, and potential risk area for each hazard type label are determined. The initial potential risk area of ​​the dam to be investigated is determined and optimized. Isolated piping data, hazard data for multiple hazard type labels, and merged type data for multiple merged type labels are determined. The potential risk area of ​​the dam to be investigated is determined. A piping hazard model for each hazard type label and a piping hazard model for each merged type label are constructed. Dam image monitoring data and dam environmental monitoring data are collected. Hazard investigation is conducted on the dam to be investigated, and dam hazard data is determined. This invention can comprehensively cover conventional hazards, special risks, and isolated piping, avoiding risk omissions, accurately adapting to single and complex hazard scenarios, improving the targeting of hazard identification, strengthening the correlation and reliability of hazard judgment, improving investigation efficiency and accuracy, and achieving comprehensive and seamless prevention and control of piping hazards.

[0014] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in this application.

[0015] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0016] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a schematic diagram of a piping hazard investigation method based on image recognition in an embodiment of the present invention; Figure 2 This is a flowchart of a piping hazard investigation system based on image recognition, as described in an embodiment of the present invention. Detailed Implementation

[0017] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention. Example 1:

[0018] This invention provides a method for investigating piping hazard based on image recognition, with reference to... Figure 1 ,include: S1: Acquire and analyze the dam hazard data and historical piping data of the dam to be investigated, determine the initial type hazard data, risk distance threshold and type potential risk area for each hazard type label, and determine the initial potential risk area of ​​the dam to be investigated. S2: Based on the historical piping data of the embankment to be investigated, the initial potential risk area is optimized, and isolated piping data, investigation type hazard data with multiple hazard type labels, and merged type data with multiple merged type labels are determined, and the potential risk area of ​​the embankment to be investigated is determined. S3: Based on the investigation type hazard data of each hazard type label of the dam to be investigated, the merged type data of each merged type label, and the historical piping data, construct the piping hazard model for each hazard type label, and construct the piping hazard model for each merged type label. S4: Collect image monitoring data of the dam to be investigated based on image sensors, and collect environmental monitoring data of the dam to be investigated based on environmental sensor groups; S5: Based on the piping hazard model of all hidden danger types labeled on the dike to be investigated, the piping hazard model of all combined type labels, the dike image monitoring data, and the dike environmental monitoring data, conduct hazard investigation on the dike to be investigated and determine the dike hazard data.

[0019] In this embodiment, two types of core data are collected for the dikes to be investigated: dike hazard data and historical piping data. The dike hazard data includes multiple locations of hazards within the dike body. Each hazard location is labeled with a corresponding hazard type tag and hazard severity tag. The hazard type tag distinguishes different types of dike defects, while the hazard severity tag indicates the severity level of each hazard. The historical piping data records detailed information on multiple piping events, including the location of the piping, the corresponding dike section, the dike station number, the time of occurrence, the piping type tag, the hazard level tag, and historical environmental vectors and historical image recognition vectors at the time of the piping. These two types of data are systematically analyzed. All dike hazard locations are classified according to the hazard type tag, forming initial hazard data for each hazard type tag. This dataset contains all dike hazard locations under that type and their hazard severity tags. Combining the spatial correlation between piping locations and hazard locations in the historical piping data, a risk distance threshold corresponding to each hazard type tag is calculated. This threshold is a key basis for delineating potential risk areas. Using the location of the hazard in the embankment body in the initial hazard data of each hazard type label as the center and the corresponding risk distance threshold as the radius, a potential risk sub-region for each hazard location is delineated. The potential risk sub-regions of all hazard locations under the same hazard type label are integrated to form the type potential risk region of that hazard type label. Then, the type potential risk regions of all hazard type labels are summarized to finally determine the initial potential risk region of the embankment to be investigated.

[0020] In this embodiment, the initially delineated potential risk areas are optimized and adjusted based on the collected historical piping data. Each piping location in the historical piping data is checked against the initial potential risk area. Historical piping sub-data corresponding to uncovered piping locations are filtered out to form initial isolated data. The number of historical piping sub-data in the initial isolated data is counted, and its proportion to the total number of historical piping sub-data is calculated. This proportion is compared with a preset proportion threshold. If the proportion is lower than the preset proportion threshold, it indicates that the number of uncovered piping is small, and the initial isolated data is directly identified as isolated piping data. If the proportion is higher than or equal to the preset proportion threshold, it indicates that the original risk distance threshold may be unreasonable, and the risk distance threshold for each hazard type label needs to be adjusted and optimized. Based on the optimized risk distance threshold and the levee hazard location in the initial type hazard data for each hazard type label, potential risk sub-areas and type potential risk areas are redefined to form optimized initial potential risk areas. Then, historical piping sub-data corresponding to piping locations not covered by the optimized initial potential risk areas are filtered out and identified as isolated piping data. During the optimization process, the initial type of hazard data is improved, supplemented with information on hidden hazards discovered during the investigation, forming investigation type hazard data for each hazard type label. Potential risk sub-regions and hidden investigation areas with spatial overlap within the initial potential risk area are merged to form potential merged sub-regions. Based on the hazard type labels corresponding to the merged areas, a merge type label is determined for each potential merged sub-region. All relevant data from all potential merged sub-regions are integrated to form merge type data for each merge type label. By combining the potential risk sub-regions corresponding to all investigation type hazard data, the potential merged sub-regions corresponding to merged type data, the special risk sub-regions, and the locations corresponding to isolated piping data, the potential risk areas of the dam to be investigated are finally determined.

[0021] In this embodiment, based on the hazard data of each hazard type label, the features of all hazard locations on the dike body under that type, as well as relevant information of the corresponding potential risk sub-regions or hidden investigation areas, are extracted. Combined with historical piping data related to that hazard type, including historical environmental vectors, historical image recognition vectors, piping type labels, and hazard level labels, a dedicated piping hazard model is constructed for each hazard type label. This model can adapt to the characteristic patterns of piping caused by a single hazard type. Simultaneously, based on the merged type data of each merged type label, the spatial combination features and related environmental information of all hazard locations on the dike body within the potential merged sub-region are extracted. The historical merged piping data corresponding to that region is summarized, including multi-dimensional historical environmental vectors, historical image recognition vectors, and corresponding piping type labels and hazard level labels, etc., to construct a dedicated piping hazard model for each merged type label. This model is specifically adapted to complex scenarios where hazard synergy leads to piping.

[0022] In this embodiment, image sensors are deployed to comprehensively monitor the potential risk areas of the dike under investigation. These image sensors include high-definition cameras mounted on drones and cameras fixed to the dike body. They can capture images of each sub-area within the potential risk area from different angles and at different times. The sub-areas specifically include potential risk sub-areas, hidden investigation areas where hazards have been discovered, special risk sub-areas, and potential merged sub-areas. The collected regional image monitoring data includes multiple monitoring images and the corresponding monitoring location and shooting angle for each image. Simultaneously, for each piping location in the isolated piping data, image sensors are used for specific image acquisition to obtain isolated image monitoring data. This data also includes multiple monitoring images, monitoring locations, and shooting angles. An environmental sensor group is deployed to collect environmental parameters for each sub-area within the potential risk area of ​​the dike. The environmental sensor group is specifically configured according to the hazard type labels of different sub-areas, collecting core environmental parameters related to piping occurrence to form regional environmental monitoring data for each sub-area. Simultaneously, a corresponding environmental sensor is configured for each piping location in the isolated piping data to collect isolated environmental monitoring data for that location. The regional image monitoring data of all sub-regions are integrated with the isolated image monitoring data of all isolated piping locations to form the dam image monitoring data of the dam to be investigated; the regional environmental monitoring data of all sub-regions are integrated with the isolated environmental monitoring data of all isolated piping locations to form the dam environmental monitoring data of the dam to be investigated.

[0023] In this embodiment, integrated dam image monitoring data and dam environmental monitoring data are used as inputs and applied to corresponding piping hazard models. For potential risk sub-regions and hidden investigation areas with discovered hazards within the potential risk area of ​​the dam, the piping hazard model with the corresponding hazard type label is invoked for hazard analysis; for potentially merged sub-regions, the piping hazard model with the corresponding merged type label is invoked for hazard analysis; for special risk sub-regions, manual hazard identification is performed in conjunction with the collected data; for isolated piping locations, hazard determination is made by referring to historical correlation data and real-time monitoring data. Through model analysis and manual assisted judgment, the hazard situation of each area and each isolated piping location is determined, including whether piping exists, the specific location of the piping, the type of piping, and the hazard level. All this information is summarized and organized to finally form complete dam hazard data for the dam to be investigated.

[0024] The beneficial effects of the above technology are as follows: By analyzing the dam hazard data and historical piping data of the dams to be investigated, the initial type hazard data, risk distance threshold, and potential risk area of ​​each hazard type label are determined. The initial potential risk area of ​​the dams to be investigated is identified and optimized. Isolated piping data, hazard data for multiple hazard type labels, and combined type data for multiple combined type labels are determined. The potential risk area of ​​the dams to be investigated is identified. Piping hazard models for each hazard type label and each combined type label are constructed. Dam image monitoring data and dam environmental monitoring data are collected. Hazard investigation is conducted on the dams to be investigated, and dam hazard data is determined. This technology can comprehensively cover conventional hazards, special risks, and isolated piping, avoiding risk omissions, accurately adapting to single and complex hazard scenarios, improving the targeting of hazard identification, strengthening the correlation and reliability of hazard judgment, improving investigation efficiency and accuracy, and achieving comprehensive and seamless prevention and control of piping hazards. Example 2:

[0025] Based on Example 1, an image recognition-based method for investigating piping hazards acquires and analyzes dam hazard data and historical piping data for the dam to be investigated. It determines the initial type hazard data, risk distance threshold, and potential risk area for each hazard type label, and identifies the initial potential risk area of ​​the dam to be investigated, including: Acquire data on potential hazards and historical piping data for the dikes to be investigated. The dike hazard data includes multiple locations of hazards on the dike body, as well as hazard type and hazard severity labels for each location. The historical piping data includes sub-data on multiple piping events, which includes piping location, dike body section, dike station number, historical piping time, piping type label, hazard level label, historical environmental vector, and historical image recognition vector. The hazard level labels include Level 1, Level 2, Level 3, and Level 4. Based on the hazard type labels of all hazard locations in the dam hazard data, all hazard locations in the dam body are divided, and the initial type hazard data for each hazard type label is determined. The initial type hazard data includes multiple hazard locations in the dam body and the hazard degree label for each hazard location. Based on the initial type hazard data of each hazard type label of the dam to be investigated, as well as the piping location, historical piping time, and hazard level label in all historical piping sub-data in the historical piping data, calculate the risk distance threshold of each hazard type label of the dam to be investigated. Using the location of each hazard in the initial type hazard data of each hazard type label of the dam to be investigated as the center and the risk distance threshold of each hazard type label of the dam to be investigated as the radius, the potential risk sub-region of each hazard location in the initial type hazard data of each hazard type label of the dam to be investigated is determined. Based on the potential risk sub-regions of all hazard locations in the initial type hazard data of each hazard type label of the dam to be investigated, the type potential risk area of ​​each hazard type label of the dam to be investigated is determined. Based on the type potential risk areas of all hazard type labels of the dam to be investigated, the initial potential risk area of ​​the dam to be investigated is determined.

[0026] In this embodiment, two core types of data are comprehensively collected for the dikes to be investigated: dike hazard data and historical piping data. The dike hazard data includes multiple specific locations of hazards within the dike body. Each hazard location is labeled with a hazard type tag and a hazard severity tag. The hazard type tag distinguishes the specific category of the hazard, while the hazard severity tag indicates the severity level of each hazard. The historical piping data consists of sub-data corresponding to multiple piping events. Each sub-data includes the specific location of the piping, the dike section where it occurred, the corresponding dike station number, the historical time of the piping, the piping type tag, and a hazard level tag indicating the severity of the hazard. The hazard level tag is clearly divided into four levels: Level 1, Level 2, Level 3, and Level 4. In addition, it includes historical environmental vectors composed of multiple environmental features affecting the piping, and historical image recognition vectors extracted through analysis of historical piping image sets. These data collectively provide comprehensive support for subsequent analysis.

[0027] In this embodiment, the hazard type labels include embankment cracks, biological cavities, soil structural defects, and shallow damage.

[0028] In this embodiment, the historical environment vector includes the difference in water head inside and outside the dike during historical piping time, the rate of water level rise, the number of days with high water level, the daily cumulative rainfall, the maximum hourly rainfall intensity, the water level of the piezometer on the back side, the permeability coefficient, the surface permeability coefficient, and the thickness of the overburden layer.

[0029] In this embodiment, the piping type label includes erosion, sludge flow, contact scour, and contact sludge flow.

[0030] In this embodiment, after acquiring complete data, the system systematically classifies all collected hazard locations based on the hazard type labels corresponding to all hazard locations within the dam hazard data. Hazard locations belonging to the same hazard type label are grouped into one category, thus forming initial hazard data for each hazard type label. Each initial hazard data set contains multiple hazard locations within that category, and retains the corresponding hazard severity label for each hazard location.

[0031] In this embodiment, based on the initial type hazard data of each hazard type label of the dam to be investigated and the piping location, historical piping time, and hazard level label in all historical piping sub-data of the historical piping data, the risk distance threshold of each hazard type label of the dam to be investigated is calculated. The calculation formula is expressed as follows: ; in, This represents the distance between the initial type hazard data for the i-th hazard type label and the piping location in the j-th historical piping sub-data. This represents the p-th preset distance threshold for the i-th hazard type label. This represents the (p+1)th preset distance threshold for the i-th hazard type label. The location of the piping in the j-th historical piping sub-data is represented by a first indicator function based on the p-th preset distance threshold of the i-th hazard type label. The location of the piping in the j-th historical piping sub-data is represented by a first indicator function based on the (p+1)-th preset distance threshold of the i-th hazard type label. The hazard severity label of the k-th hazard location in the initial hazard data of the i-th hazard type label. This represents the risk distance threshold for the i-th hazard type label. This indicates the location of the k-th embankment hazard in the initial type hazard data for the i-th hazard type label. Let represent the hazard severity label of the k-th hazard location in the initial hazard data for the i-th hazard type label, and let iN1 represent the number of hazard locations in the initial hazard data for the i-th hazard type label. This represents the location of the piping in the j-th historical piping sub-data. Nj represents the risk level label corresponding to the j-th historical piping sub-data point in the historical piping data, and N2 represents the number of historical piping sub-data points in the historical piping data. This represents the historical piping time in the j-th historical piping sub-data, where TC represents the current time. This represents the risk weight of the j-th historical piping sub-data in the historical piping data.

[0032] In this embodiment, This represents the piping risk value corresponding to the hazard level label in the j-th historical piping sub-data. For example, =3, piping risk value is 2. =4.

[0033] In this embodiment, the level value The risk level is determined based on the risk level label in the j-th historical piping sub-data. For example, if the risk level label is Level 1, =1, the danger level label is level two. =2, and so on.

[0034] In this embodiment, after calculating the risk distance threshold for each hazard type label, potential risk sub-regions are delineated according to specific rules. Specifically, each levee hazard location included in the initial hazard data for each hazard type label is used as the geometric center point, i.e., the center of a circle. The risk distance threshold corresponding to that hazard type label is used as the radius to delineate a circular area in space. This circular area is the potential risk sub-region corresponding to the levee hazard location, and it visually reflects the spatial range from that hazard location that may trigger piping hazards.

[0035] In this embodiment, after delineating the potential risk sub-regions for all hazard locations on the dike body, regional integration is performed. First, for each hazard type label, the potential risk sub-regions corresponding to all hazard locations on the dike body in the initial hazard data are summarized and integrated to form a unique type-specific potential risk region for each hazard type label. This region covers the entire spatial range where all hazards under that hazard type may cause piping. Subsequently, the type-specific potential risk regions corresponding to all hazard type labels of the dike to be investigated are integrated again, ultimately forming the initial potential risk region for the dike to be investigated. This region, based on existing data, is a preliminary determination of the entire area requiring focused attention and investigation.

[0036] The beneficial effects of the above technologies are as follows: acquiring and analyzing the dam hazard data and historical piping data of the dam to be investigated, determining the initial type hazard data, risk distance threshold and type potential risk area of ​​each hazard type label, and determining the initial potential risk area of ​​the dam to be investigated, which can improve the accuracy of risk sub-region identification, take into account both local accuracy and overall integrity, and optimize the investigation efficiency and risk identification depth. Example 3:

[0037] Based on Example 2, an image recognition-based method for investigating piping risks optimizes the initial potential risk area based on historical piping data of the embankment to be investigated, including: Based on the piping locations of all historical piping sub-data in the historical piping data and the initial potential risk area of ​​the embankment to be investigated, determine whether there are piping locations that do not belong to the initial potential risk area, and determine the initial isolated data based on the historical piping sub-data of all piping locations that do not belong to the initial potential risk area. Determine whether the number of historical piping sub-data in the initial isolated data is below a preset percentage threshold. If so, determine that the initial isolated data is isolated piping data. Otherwise, optimize the risk distance threshold for each hazard type label of the embankment to be investigated. Based on the optimized risk distance thresholds for all hazard type labels and the location of each hazard in the embankment body in the initial hazard data for each hazard type label, determine the optimized initial potential risk area and identify the isolated piping data.

[0038] In this embodiment, the piping location information contained in all historical piping sub-data is extracted from the historical piping data, and combined with the already delineated initial potential risk areas of the dikes to be investigated. The location of each historical piping is spatially compared with the initial potential risk area one by one, meticulously verifying whether each piping location falls within the spatial range covered by the initial potential risk area. For piping locations confirmed not to be within the initial potential risk area, their corresponding complete historical piping sub-data is extracted, and all such historical piping data are integrated and collected to form initial isolated data.

[0039] In this embodiment, the specific number of historical piping sub-data included in the initial isolated data is counted, and the total number of all historical piping sub-data in the historical piping data is also counted. The proportion of the number of initial isolated data to the total number of historical piping sub-data in the historical piping data is calculated. This calculated proportion is compared with a preset proportion threshold to determine whether the proportion is below the preset proportion threshold. If the proportion is indeed below the preset proportion threshold, it indicates that the number of piping outside the initial potential risk area is small, and the previously collected initial isolated data is directly officially identified as isolated piping data. If the proportion is not below the preset proportion threshold, it indicates that the number of piping outside the initial potential risk area is large, and the risk distance thresholds corresponding to the original hazard type labels may be unreasonable, requiring adjustment and optimization of the risk distance thresholds for each hazard type label. After completing the optimization of the risk distance thresholds for all hazard type labels, the potential risk sub-regions for each hazard location are redefined, using each levee hazard location in the initial hazard data of each hazard type label as the core, combined with the optimized corresponding risk distance thresholds. Subsequently, the potential risk sub-regions of all hazard locations on the levee body under each hazard type label are integrated to form an optimized type potential risk region for each hazard type label. Then, all optimized type potential risk regions for all hazard type labels are comprehensively integrated to determine the optimized initial potential risk region. Finally, the piping locations of all historical piping sub-data in the historical piping data are spatially compared with the optimized initial potential risk region. Historical piping sub-data corresponding to piping locations that are still not within the optimized initial potential risk region are filtered out, and these data are ultimately identified as isolated piping data.

[0040] In this embodiment, the preset percentage threshold can be set to 22%. If the geological conditions of the dam to be investigated are simple and the historical piping data are complete, it can be tightened to 15%. If the geological conditions of the dam to be investigated are complex, it can be relaxed to 25%.

[0041] The beneficial effects of the above technologies are as follows: optimizing the initial potential risk area based on the historical piping data of the dam to be investigated can avoid risk omissions caused by fixed thresholds, accurately locate isolated piping, improve the rationality of risk area delineation, and enhance the dynamic adaptability and accuracy of piping investigation. Example 4:

[0042] Based on Example 3, an image recognition-based method for investigating piping hazards identifies isolated piping data, hazard data with multiple hazard type labels, and merged type data with multiple merged type labels, and determines the potential risk area of ​​the dam to be investigated, including: The largest potential risk sub-region in the initial potential risk area is selected as the implicit judgment area for isolated piping data; Based on the piping locations of all historical piping sub-data in the implicit judgment area and isolated piping data, determine whether a implicit judgment area has a preset number or more piping locations. If it does, determine each implicit judgment area with a preset number or more piping locations as an implicit investigation area, and determine the historical piping sub-data of all piping locations that do not belong to any implicit judgment area as the adjusted isolated piping data. If it does not, do not adjust the isolated piping data. For each hidden investigation area, a hazard investigation is carried out. If a hazard is found, the location of the hazard on the embankment, the hazard type label, and the hazard severity label of the hidden investigation area are determined. If no hazard is found, the hidden investigation area is determined to be a special risk sub-area. Determine whether there is any overlap between all potential risk sub-regions and all hidden investigation areas where hazards have been discovered in the initial potential risk area. For each potential risk sub-region and hidden investigation area with regional overlap, merge them to determine each potential merged sub-region. Then, determine the merge type label of each potential merged sub-region based on the hazard type labels corresponding to the two or more potential risk sub-regions and hidden investigation areas with regional overlap. Based on the initial type hazard data of each hazard type label, all potential risk sub-regions without regional overlap and all hidden investigation areas where hazards are found without regional overlap, the location of hazard in the dike body, hazard type label, and hazard severity label are determined, and the investigation type hazard data of each hazard type label of the dike to be investigated is determined. The investigation type hazard data includes multiple hazard locations in the dike body and the hazard severity label of each hazard location. Based on the merging type label of each potential merging sub-region, all potential merging sub-regions are divided, and the merging type data of each merging type label is determined. The merging type data includes multiple potential merging sub-regions, multiple embankment hazard locations in each potential merging sub-region, and hazard degree labels for each embankment hazard location. Based on all potential risk sub-regions without regional overlap in the initial potential risk area, all hidden investigation areas without regional overlap where hidden dangers were found, all special risk sub-regions where no hidden dangers were found, and all potential merged sub-regions, the potential risk areas of the dikes to be investigated are determined.

[0043] In this embodiment, within the initially defined potential risk area, the areas of all included potential risk sub-areas are statistically analyzed and compared. The sub-area with the largest area is selected and explicitly designated as the implicit judgment area for isolated piping data. This area is chosen based on the wider coverage and higher potential for carrying hidden hazards inherent in the largest area.

[0044] In this embodiment, a defined hidden judgment area is used as the analysis boundary. Simultaneously, the piping location information corresponding to all historical piping sub-data in the isolated piping data is extracted, and these piping locations are spatially correlated with the hidden judgment area. A predetermined quantity standard, or preset quantity, is set to determine whether a single hidden judgment area contains a preset quantity or more piping locations. If a preset quantity or more piping locations are indeed found within a certain hidden judgment area, then this hidden judgment area that meets the quantity condition is designated as a hidden investigation area, meaning that this area is highly likely to contain undiscovered hidden hazards and requires focused investigation. Simultaneously, all historical piping sub-data whose piping locations do not belong to any hidden investigation area are redefined as adjusted isolated piping data; the piping locations corresponding to these data have not yet been found to be associated with hidden hazards. If a preset quantity or more piping locations are not found within any of the hidden judgment areas, then the previously determined isolated piping data is not adjusted in any way and remains in its original state.

[0045] In this embodiment, if three piping points are found within a 20-meter inspection step in the engineering process, it is considered a cluster. The baseline density is equal to 1 / 20 meter, which is approximately equal to 1 / 2000 square meter. Therefore, the preset number of implicit judgment areas should be the area multiplied by 1 / 2000 square meter, which is the area divided by 2000. Thus, the formula for calculating the preset number can be: .

[0046] In this embodiment, for each designated hidden inspection area, professional inspectors are organized to conduct a comprehensive and meticulous hazard inspection using appropriate methods and techniques. During the inspection, key components and indicators such as the embankment structure and soil condition within the area are carefully examined. If a specific hazard is discovered, its location within the hidden inspection area is recorded and confirmed, and the corresponding hazard type and severity labels are clearly defined. This newly discovered hazard information is then added to the embankment hazard data system. If, after a comprehensive inspection, no explicit hazard is found within the hidden inspection area, this area, which has no apparent hazard but contains multiple isolated piping locations, is designated as a special risk sub-area. This area, lacking explicit hazards but with a high risk of piping, needs to be included in the key monitoring scope.

[0047] In this embodiment, the spatial overlap of all potential risk sub-regions within the initial potential risk area, as well as all hidden investigation areas where potential hazards have been identified, is then assessed and verified. The spatial extent of each region is compared one by one to confirm whether two or more regions spatially overlap. For each group of two or more spatially overlapping potential risk sub-regions and hidden investigation areas, they are spatially merged to form a new unified area, i.e., each potential merged sub-region. During the merging process, the original hazard type labels of each participating area are collected and organized. Based on these different hazard type labels, a unique merging type label is determined for each potential merging sub-area. This label can reflect the characteristics of multiple hazard types contained in the merging area. For example, if two potential risk sub-areas and a certain hidden investigation area overlap, the two potential risk sub-areas and the hidden investigation area are merged. The hazard type labels corresponding to the two potential risk sub-areas and the hidden investigation area are levee crack, levee crack, and biological cave, respectively. The merging type label of the merged potential merging sub-area is determined to be levee crack-biological cave. If two or more potential risk sub-areas and hidden investigation areas have the same hazard type label, the merging type label of the merged potential merging sub-area is consistent with the hazard type label of the merged area.

[0048] In this embodiment, two parts of relevant data are collected and integrated based on each hazard type label. The first part consists of the initial hazard data for each hazard type label, including the location of hazard on the embankment, hazard type label, and hazard severity label for all potential risk sub-areas that do not spatially overlap with other areas. The second part consists of all hidden hazard areas that have been identified through investigation and do not spatially overlap with other areas, including the location of hazard on the embankment, hazard type label, and hazard severity label. These two parts of data are summarized and organized to finally determine the investigation type hazard data corresponding to each hazard type label of the embankment to be investigated. Each investigation type hazard data includes multiple hazard locations on the embankment under that hazard type, and the corresponding hazard severity label is fully preserved for each hazard location.

[0049] In this embodiment, all potential merged sub-regions are systematically divided using the identified merge type label for each potential merged sub-region as the classification standard. Potential merged sub-regions with the same merge type label are grouped into one category, thereby determining the merge type data corresponding to each merge type label. Each merge type data contains multiple potential merged sub-regions under that merge type label, as well as multiple levee hazard locations within each potential merged sub-region. Furthermore, a detailed hazard severity label is recorded for each levee hazard location, achieving orderly management of composite hazard area data.

[0050] In this embodiment, regional data from multiple sources is used to determine the final potential risk area of ​​the dam to be investigated. Specifically, this includes all potential risk sub-areas within the initial potential risk area that do not spatially overlap with other areas; all hidden investigation areas where hazards were discovered and do not spatially overlap with other areas; all special risk sub-areas where no hazards were found but were identified as high-risk; and all potential merged sub-areas formed by merging overlapping areas. These areas of different types and origins are comprehensively integrated to form a complete potential risk area for the dam, covering all piping risk points, hazard areas, and special high-risk areas.

[0051] The beneficial effects of the above technologies are as follows: they can identify isolated piping data, investigation type hazard data with multiple hazard type labels, and merged type data with multiple merged type labels, and identify the potential risk areas of the dams to be investigated. This can achieve accurate classification of composite risks, take into account both visible hazards and high-risk areas without hazards, avoid regional redundancy while ensuring comprehensive hazard coverage, and improve the systematicness and accuracy of risk area delineation. Example 5:

[0052] Based on Example 4, an image recognition-based method for investigating piping hazards constructs a piping hazard model for each hazard type label, and a piping hazard model for each combined type label, based on the hazard type data of each hazard type label of the dam to be investigated, the combined type data of each combined type label, and historical piping data. The method includes: Based on the potential risk sub-area or hidden investigation area of ​​each hazard location in the hazard data of each hazard type label of the dam to be investigated, and the piping location of all historical piping sub-data in the historical piping data, the historical hazard piping data of each hazard location in the dam to be investigated is determined. The historical environment vector and historical image recognition vector of all historical piping sub-data in the historical hidden danger piping data of all potential risk sub-regions or hidden investigation areas of all hidden danger locations in the investigation type data of each hidden danger type label of the embankment to be investigated are used as the input of the piping hazard model of each hidden danger type label. The piping location, piping type label, and hazard level label of all historical hidden danger piping sub-data in the historical hidden danger piping data of all potential risk sub-regions or hidden investigation areas of all hidden danger locations in the investigation type data of each hidden danger type label of the embankment to be investigated are used as the output of the piping hazard model of each hidden danger type label. Thus, the piping hazard model of each hidden danger type label is constructed. Based on the potential risk sub-regions or hidden investigation areas of all potential hazard locations in the merging type data of each potential merging sub-region of each merging type label of the dam to be investigated, and the piping locations of all historical piping sub-data in the historical piping data, the historical merged piping data of each potential merging sub-region in the merging type data of each merging type label of the dam to be investigated is determined. The piping hazard model for each hazard type label is constructed by taking the locations of all potential hazard locations in the merging sub-regions of the merging type data for each dam to be investigated, the historical environmental vectors of all historical piping sub-data in the historical merging piping data of all potential merging sub-regions, and the historical image recognition vectors as inputs. The piping hazard model for each hazard type label is constructed by taking the piping location, piping type label, and hazard level label of all historical piping sub-data in the historical merging piping data of all potential merging sub-regions of the merging type data for each dam to be investigated as outputs.

[0053] In this embodiment, the core basis for data association is clearly defined: the hazard data corresponding to each hazard type label of the dike to be investigated, including the potential risk sub-area or hidden investigation area corresponding to each hazard location on the dike body. Simultaneously, the piping location information of all historical piping sub-data records in the historical piping data is extracted. The potential risk sub-area or hidden investigation area of ​​each hazard location on the dike body under each hazard type label is used as the spatial filtering range. All piping locations in the historical piping data are compared one by one to filter out those historical piping sub-data that fall within this spatial range. These filtered historical piping sub-data are then summarized and organized to ultimately determine the historical hazard piping data corresponding to the hazard location on the dike body in the hazard data of the hazard type label, achieving precise binding between hazard location and associated historical piping data.

[0054] In this embodiment, the input data for the piping hazard model of each hazard type label includes three core parts. The first part is the location of each specific hazard in the dam body within the hazard data for that hazard type label. The second and third parts are the historical environmental vectors and historical image recognition vectors of all historical piping sub-data records in the historical hazard piping data of the potential risk sub-region or hidden investigation area corresponding to that hazard location. The model's output data consists of the piping location, piping type label, and hazard level label of all historical piping sub-data records in the historical hazard piping data corresponding to that hazard type label. By establishing a mapping relationship between the input and output data, a piping hazard model specific to each hazard type label is constructed, ensuring that the model can accurately output the corresponding key piping information based on the input hazard, environment, and image features.

[0055] In this embodiment, the same type of hidden danger has a unified piping induction mechanism, and its correlation with the spatial distribution, type characteristics, and hazard level of piping occurrence is stable. By focusing on the spatial attributes of all hazard locations in the levee under this type of hidden danger, and combining the corresponding historical piping data in its potential risk sub-regions or hidden investigation areas, the association information between the spatial characteristics of the hidden danger and historical environmental vectors and historical image recognition vectors is extracted. The visual appearance patterns and risk level evolution logic of piping caused by this type of hidden danger under different environmental conditions are learned, forming a piping judgment logic that is only applicable to this single type of hidden danger.

[0056] In this embodiment, the location of potential hazards in the dike serves as a spatial benchmark, directly influencing the distribution range of piping locations. The model learns the spatial relative relationship between piping locations and potential hazard locations from historical data to determine the high-incidence area of ​​piping around such potential hazards. Parameter combinations such as the difference in water head between the inside and outside of the dike and the rate of water level rise in the historical environmental vector map piping type labels. Different combinations of environmental parameters correspond to specific piping occurrence mechanisms. For example, a high water head difference combined with a thin overburden layer tends to map a siltation type, while a medium water head difference combined with crack development characteristics tends to map a burial type. At the same time, the extreme degree of environmental parameters is positively correlated with the hazard level label; the closer the parameter is to or exceeds the critical value, the higher the hazard level mapped. Visual features such as sand ring contours and water seepage area in the historical image recognition vector are mapped to piping type labels. The exclusive visual features of different piping types directly correspond to specific type labels, forming a multi-dimensional and precise mapping from potential hazard spatial features + environmental inducing features + visual appearance features to piping location + piping type + hazard level.

[0057] In this embodiment, for each merge type label, historical merged piping data is determined. Taking each potential merge sub-region under each merge type label as a unit, potential risk sub-regions or hidden investigation areas corresponding to all levee hazard locations within that potential merge sub-region are extracted. Simultaneously, piping location information for all historical piping sub-data in the historical piping data is retrieved. Using all potential risk sub-regions or hidden investigation areas within that potential merge sub-region as the overall spatial filtering range, all piping locations in the historical piping data are compared one by one, filtering out all historical piping sub-data whose piping locations fall within this spatial range. These filtered historical piping sub-data are integrated and aggregated to determine the historical merged piping data corresponding to the potential merge sub-region in the merge type data of that merge type label, providing data support for the model construction of the merge type label.

[0058] In this embodiment, the input data for the piping hazard model of each merge type label includes three key parts. The first part is the location of all levee hazard locations within all potential merged sub-regions in the merge type data of that merge type label. The second and third parts are the historical environmental vectors and historical image recognition vectors of all historical piping sub-data records in the historical merged piping data of all potential merged sub-regions, respectively. The output data of the model is the piping location, piping type label, and hazard level label of all historical piping sub-data records in the historical merged piping data corresponding to that merge type label. By establishing the correspondence between the above inputs and outputs, a piping hazard model specific to each merge type label is constructed, enabling the model to adapt to complex hazard scenarios and accurately identify piping hazards that may be caused by the combined effect of one or more hazards.

[0059] In this embodiment, the construction of the piping hazard model with merged type labels focuses on the combination characteristics of hazard types within the merged area and the synergistic pattern of piping induction. The potential merged sub-areas may contain concentrated distributions of the same type of hazard or cover the superposition and coexistence of different types of hazard. For merged areas with the same hazard type label, the core is to strengthen the concentration effect of this type of hazard. The dense distribution of the same type of hazard will amplify the probability of piping induction. The model needs to learn the strengthening correlation pattern between the concentrated area of ​​this type of hazard and environmental vectors and image recognition vectors. For merged areas with different hazard type labels, such as levee cracks-biological caves, the core is to capture the synergistic mechanism of multiple types of hazard. The structural defects of different hazard can be interconnected and complementary, forming channels or conditions that are more likely to cause piping. The model needs to explore the specific correlation between the spatial combination patterns of multiple types of hazard and the characteristics of piping occurrence, while avoiding the limitations of single hazard type models. It should be adapted to the composite risk scenario of concentrated strengthening of the same type and synergistic induction of different types, ensuring that the model can accurately identify the occurrence pattern of piping within the merged area.

[0060] In this embodiment, the spatial distribution characteristics of all hazard locations on the dike directly map to piping locations. Hazards of the same type, such as the core overlapping area of ​​two concentrated areas of dike cracks, tend to map as the core location of high-incidence piping. Hazards of different types, such as the connecting nodes or adjacent areas of dike cracks and biological caves, tend to map as piping-prone locations. The model learns the correspondence between hazard combination layouts and piping locations in historical data to lock the key distribution range of piping within the merged area. The combined parameters of all historical environmental vectors map to piping type labels. When the same hazard type is merged, the extreme degree of environmental parameters is strongly correlated with the piping type that this type of hazard is suitable for. For example, in the merged area of ​​two dike cracks, the high water level rise rate + the difference in water head inside and outside the dike tends to map to the subsurface erosion type. When different hazard types are merged, the combination of environmental parameters and multiple hidden... The synergistic mechanism of risk assessment is adapted, such as the merging area of ​​dike cracks and biological caves, where a moderate head difference plus a sustained high water level tends to map the contact scour type; the overall strength of historical environmental vectors and the feature saliency of historical image recognition vectors jointly map the risk level label. The closer the environmental parameters are to the critical value, and the more obvious the visual features of piping in the image, such as the diameter of the sand ring and the area of ​​water inflow, the higher the risk level is mapped; the combined features of historical image recognition vectors help to confirm the piping type. Piping caused by different combinations of risks often presents a unique visual superposition effect. For example, piping caused by dike cracks and biological caves may simultaneously show the characteristics of eroded sand rings and concentrated water inflow from contact scour, forming a precise synergistic mapping from the spatial features of risk combination + the combined features of environmental parameters + the visual superposition features to the location of piping + the type of piping + the risk level.

[0061] The beneficial effects of the above technology are as follows: Based on the investigation type hazard data of each hazard type label of the dam to be investigated, the combined type data of each combined type label, and historical piping data, a piping hazard model for each hazard type label and a piping hazard model for each combined type label can be constructed, which can achieve accurate adaptation to different risk scenarios and improve the pertinence and reliability of piping hazard identification. Example 6:

[0062] Based on Example 1, a method for investigating piping hazards based on image recognition is provided, which involves collecting image monitoring data of the dam to be investigated based on image sensors and collecting environmental monitoring data of the dam to be investigated based on an environmental sensor array, including: Based on image sensors, regional image monitoring data of each sub-region in the potential risk area of ​​the dam to be investigated is collected. The regional image monitoring data includes multiple monitoring images and the monitoring position and shooting angle of each monitoring image. The sub-regions are: potential risk sub-region, hidden investigation area where hidden dangers are found, special risk sub-region, and potential merged sub-region. Regional environmental monitoring data for each sub-region within the potential risk area of ​​the dam to be investigated is collected based on an environmental sensor array. Isolated image monitoring data for each piping location in the isolated piping data of the embankment to be investigated is collected based on image sensors. The isolated image monitoring data includes multiple monitoring images and the monitoring location and shooting angle of each monitoring image. Isolated environmental monitoring data for each piping location in the isolated piping data of the dam to be investigated, collected from the environmental sensor array; Based on the regional image monitoring data of all sub-regions in the potential risk area of ​​the dam to be investigated and the isolated image monitoring data of all piping locations in the isolated piping data, the dam image monitoring data of the dam to be investigated is determined. Based on the regional environmental monitoring data of all sub-regions in the potential risk area of ​​the dam to be investigated, and the isolated environmental monitoring data of all piping locations in the isolated piping data, the dam environmental monitoring data of the dam to be investigated is determined.

[0063] In this embodiment, the core of defining the data collection scope is the potential risk area of ​​the dike to be inspected. This area includes four types of sub-areas: potential risk sub-areas, hidden inspection areas where hidden dangers have been discovered, special risk sub-areas, and potential merged sub-areas. For these sub-areas, image sensors are used to collect regional image monitoring data. The image sensors can be flexibly selected according to the data collection scenario. For example, drones equipped with high-definition visible light cameras are used for large-area patrols, which can capture wide-area monitoring images from multiple angles in the air; high-definition cameras fixed to the dike body are used for continuous monitoring at fixed points, which can capture surface changes in specific sub-areas in real time; and infrared thermal imagers are suitable for nighttime or low-visibility environments, which can identify hidden seepage or water inrush signs through temperature differences. During the data acquisition process, the image sensor captures multiple monitoring images to ensure coverage of every key corner of the sub-area. It also accurately records the monitoring location of each image, which is usually precisely marked by the dam station number combined with the offset distance. The shooting angle of each monitoring image is also recorded. Commonly, overhead shots are used to present the overall layout of the area, side shots are used to capture the details of the dam slope facade, and level shots are used to observe the near-field features of the ground surface. These monitoring images, monitoring locations, and shooting angles together constitute the regional image monitoring data of each sub-area.

[0064] In this embodiment, an environmental sensor array is used to collect regional environmental monitoring data for each sub-region within the potential risk area of ​​the dam. For different sub-regions within the potential risk area, the environmental sensor array is precisely configured around the key influencing factors of piping corresponding to the hazard type label. For example, in the potential risk sub-region labeled as dam body cracks, because cracks easily form seepage channels, the difference in water head inside and outside the dam drives concentrated seepage, and excessive seepage pressure easily triggers piping, the environmental sensor array is configured with a piezometer to monitor changes in seepage pressure around the cracks, dual water level gauges inside and outside the dam to calculate the water head difference in real time, a crack displacement sensor to track the crack opening and closing degree, a soil moisture sensor to monitor soil saturation, and a rain gauge to collect precipitation data, comprehensively covering the inducing factors such as channel formation, seepage driving force, and soil condition. In the special risk sub-region of sandy soil dam foundations, where there are no clear hazards but a history of high piping incidence, the loose sand particles and sudden rise in water level easily lead to the inability to dissipate pore water pressure, and excessive hydraulic gradient... The system is designed to detect seepage caused by sand boiling. A rapid-response water level gauge captures water level changes, a high-frequency pore water pressure sensor records pressure hysteresis, a hydraulic gradient sensor monitors seepage gradients, a cover layer pressure sensor verifies surface protection effectiveness, and a sand migration monitoring sensor detects particle movement. For potential merged sub-regions tagged as "dike crack-biological cave," where cracks and caves connect to form composite seepage channels, the difference in water head between the inside and outside of the dike drives concentrated seepage, and soil and defects collaborate in destruction, a multi-point piezometer is used to construct the pressure field distribution, a water level difference gauge provides driving force data, a three-dimensional displacement sensor monitors the continuity of cracks, a soil settlement sensor tracks structural stability, and a comprehensive seepage flow monitor infers the channel conductivity. Each sensor group precisely matches the seepage induction logic of its corresponding sub-region, achieving comprehensive monitoring of key influencing factors. The environmental sensor group continuously collects environmental parameters closely related to seepage occurrence, forming exclusive regional environmental monitoring data for each sub-region, comprehensively reflecting the hydrological, soil, and other environmental conditions of the sub-region.

[0065] In this embodiment, each piping location recorded in the isolated piping data of the embankment to be investigated needs to be specifically monitored, and image sensors are used to collect isolated image monitoring data. Portable high-definition cameras can be used here, allowing staff to easily capture details of the piping location from close range and multiple angles. Small drones equipped with miniature cameras can also be used to flexibly photograph isolated piping locations that are difficult to access. High-speed cameras can also be configured to capture rapidly changing dynamic features such as water inflow and sand flow. During data collection, multiple monitoring images are taken, focusing on the surface condition of the piping location, such as whether there are signs of water inflow, sand rings, or soil heave. Simultaneously, the monitoring location of each image is accurately recorded to ensure complete correspondence with the actual coordinates of the isolated piping, as well as the shooting angle of each image, so as to reconstruct the scene from different perspectives. These multiple monitoring images, monitoring locations, and shooting angles together constitute the isolated image monitoring data for that isolated piping location.

[0066] In this embodiment, while acquiring image data of the isolated piping location, an environmental sensor array is simultaneously used to collect isolated environmental monitoring data. This environmental sensor array includes a handheld portable piezometer, which can be quickly deployed around the isolated piping location to monitor local seepage pressure; a miniature water level gauge that can be inserted into underground or nearby water bodies to record water level changes; a portable soil moisture meter that can measure the soil moisture content around the piping location in real time; and a small rain gauge to record rainfall in the local area. This environmental sensor array continuously collects key environmental parameters of the isolated piping location, promptly capturing subtle changes in environmental conditions, forming isolated environmental monitoring data for that location, and providing real-time environmental evidence for analyzing whether piping will recur at that location.

[0067] In this embodiment, once all regional image monitoring data for all sub-regions within the potential risk area of ​​the dam has been collected, and isolated image monitoring data for all piping locations in the isolated piping data has also been collected, the integration of the dam image monitoring data is initiated. During the integration process, firstly, the regional image monitoring data for all sub-regions is summarized and categorized by sub-region type and monitoring location to ensure clear data organization; then, isolated image monitoring data for all isolated piping locations is included in the summary scope and arranged in coordinate order of the isolated piping locations; next, all data undergoes format standardization processing, unifying image resolution, storage format, and standardizing the identification method of monitoring locations and the recording unit of shooting angles; simultaneously, duplicate image data and invalid data, such as blurry or unidentifiable images, are removed, and missing monitoring location or shooting angle information is supplemented, ultimately forming a complete image dataset covering all risk-related areas and isolated risk points of the dam to be investigated. This dataset is the dam image monitoring data for the dam to be investigated.

[0068] In this embodiment, environmental monitoring data is integrated to form dam environmental monitoring data. First, the regional environmental monitoring data of all sub-regions within the potential risk area of ​​the dam are systematically summarized and statistically classified according to sub-region category and monitoring parameters. Then, all isolated environmental monitoring data for all piping locations in the isolated piping data are included and grouped according to isolated piping location. Next, data standardization processing is performed, unifying the units of all environmental parameters, such as water level in meters and rainfall in millimeters. Outliers are removed from the collected data, such as data exceeding the sensor range or clearly inconsistent with actual environmental conditions. Methods such as moving averages are used to smooth data with large fluctuations to ensure data stability and accuracy. Finally, all processed environmental data are integrated and compiled to form a complete environmental dataset that comprehensively and accurately reflects the environmental conditions of each risk area and isolated risk point of the dam under investigation. This dataset is the dam environmental monitoring data for the dam under investigation.

[0069] The beneficial effects of the above technologies are as follows: by collecting image monitoring data of the dikes to be investigated based on image sensors and collecting environmental monitoring data of the dikes to be investigated based on environmental sensor groups, it is possible to achieve deep matching between data collection and risk characteristics, avoid invalid data collection in non-risk areas, improve data quality and utilization efficiency, and provide comprehensive and highly targeted data support for accurate judgment of piping risks. Example 7:

[0070] Based on Example 6, an image recognition-based method for investigating piping hazards is proposed. This method utilizes piping hazard models with all hazard type labels for the dam to be investigated, piping hazard models with all combined type labels, dam image monitoring data, and dam environmental monitoring data to investigate hazards in the dam and determine dam hazard data, including: All monitoring images in the regional image monitoring data of each sub-region of the potential risk area of ​​the dam to be investigated are preprocessed. All preprocessed monitoring images with the same dam chainage in the regional image monitoring data of each sub-region are image aligned to determine the monitoring chainage image set of each dam chainage in each sub-region of the potential risk area of ​​the dam to be investigated. Feature extraction is performed on all monitoring images in the monitoring chainage image set of each chainage of each sub-region in the potential risk area of ​​the dam to be investigated, and the chainage image recognition vector of each chainage of each chainage in each sub-region in the potential risk area of ​​the dam to be investigated is determined. Feature extraction is performed on the environmental monitoring data of each sub-region within the potential risk area of ​​the dam to be investigated, and the regional environmental vector of each sub-region within the potential risk area of ​​the dam to be investigated is determined. If the sub-region is a potential risk sub-region or a hidden investigation area where hidden dangers have been discovered, the regional environmental vector of the sub-region and the set of monitoring station images of all dike station numbers are input into the piping hazard model of the hidden danger type label of the potential risk sub-region or the hidden investigation area where hidden dangers have been discovered. Based on the piping hazard model, the regional hazard data of the sub-region is determined. The regional hazard data includes hazard prediction labels and multiple predicted piping locations when the hazard prediction labels exist, the predicted piping type of each predicted piping location, and the predicted hazard level. If the sub-region is a potential merged sub-region, the regional environment vector of the sub-region and the set of monitoring station images of all dam station numbers are input into the piping hazard model of the merging type label of the potential merged sub-region corresponding to the sub-region, and the regional hazard data of the sub-region is determined based on the piping hazard model. If the sub-region is a special risk sub-region, based on the regional environment vector of the sub-region and the monitoring station image set of all dam station numbers, manual hazard identification is performed to determine the regional hazard data of the sub-region; Based on the dike body part, isolated image monitoring data, isolated environmental monitoring data, and the dike body part, piping type label, hazard level label, and historical environmental vector of all historical piping sub-data corresponding to all piping locations in all isolated piping data, the predicted hazard label and isolated hazard data for each isolated piping are determined. The predicted hazard label includes whether the hazard exists or not. If the predicted hazard label is "hazard exists", the isolated hazard data includes the predicted piping type and the predicted hazard level; otherwise, the isolated hazard data is empty. Based on the regional hazard data of all sub-regions in the potential risk area of ​​the dam to be investigated, as well as the predicted hazard labels and isolated hazard data of all piping locations in the isolated piping data, the dam hazard data of the dam to be investigated is determined.

[0071] In this embodiment, preprocessing is performed on all monitoring images contained in the regional image monitoring data of each sub-region within the potential risk area of ​​the dam to be investigated. The preprocessing addresses potential issues such as noise interference, uneven lighting, and image blurring. For example, denoising algorithms are used to eliminate random noise generated during shooting, brightness equalization is used to adjust the consistency of image brightness under different lighting conditions, and image sharpening enhances detail features, ensuring that each monitoring image clearly presents the true condition of the dam surface. Then, all preprocessed monitoring images with the same dam station number in the regional image monitoring data of each sub-region are selected. Spatial alignment is performed on these images by matching feature points in the images to ensure spatial consistency of images at the same station number, eliminating the influence of shooting angle or positional deviations. Finally, a complete monitoring station number image set is formed for each dam station number in each sub-region.

[0072] In this embodiment, for each monitoring station image set of each sub-region within the potential risk area of ​​the dike to be investigated, feature extraction is performed on all monitoring images contained therein. The extracted features are mainly visual features related to piping hazards, such as the presence of typical piping phenomena such as sand rings, water seepage, soil heave, and bubble trajectories in the images. Simultaneously, the specific morphological parameters of these features are quantified, such as the diameter of the sand ring, the area of ​​the water seepage zone, and the height of the soil heave. These extracted visual features are then structurally integrated to form a unique station image recognition vector for each dike station in each sub-region. This vector comprehensively reflects the visual characteristic state of the station location, providing a visual basis for subsequent hazard assessment.

[0073] In this embodiment, environmental features are extracted from the regional environmental monitoring data of each sub-region within the potential risk area of ​​the dike to be investigated. The extracted features cover various environmental parameters closely related to piping occurrence, such as the head difference between inside and outside the dike, the rate of water level rise, seepage pressure, soil moisture content, rainfall, and hydraulic gradient. These environmental parameters are standardized and processed, abnormal fluctuation data are removed, and core parameters that can truly reflect the environmental state of the sub-region and are related to piping induction are retained to form a regional environmental vector for each sub-region. This vector provides the core environmental basis for subsequent risk assessment.

[0074] In this embodiment, if the currently processed sub-region is a potential risk sub-region or a hidden investigation area where potential hazards have been discovered, then the extracted regional environmental vector of the sub-region, along with the set of monitoring station images corresponding to all dam station numbers in the sub-region, are input into a piping hazard model that matches the hazard type label corresponding to the sub-region. This piping hazard model was specifically built for this hazard type label and can accurately identify the characteristic patterns of piping caused by this type of hazard. The model will comprehensively analyze the input environmental vector and image set, and through the learned historical correlation patterns, determine whether there is a piping hazard in the sub-region. Finally, it outputs the regional hazard data for the sub-region. This data specifically includes a hazard prediction label, multiple predicted piping locations when the hazard prediction label is present, and the predicted piping type and predicted hazard level corresponding to each predicted piping location.

[0075] In this embodiment, if the currently processed sub-region is a potential merging sub-region, then the regional environment vector of the sub-region and the set of monitoring station images of all dike stations in the sub-region are input into the piping hazard model corresponding to the merging type label of the potential merging sub-region. This model is specifically constructed for the synergistic effect of multiple locations of the same type of hazard or the synergistic effect of multiple types of hazards under the merging type label. It can adapt to the complex laws of piping caused by compound hazards. Through comprehensive analysis of the input data, the model identifies the piping risk in scenarios with multiple superimposed hazards, and then determines the regional hazard data of the potential merging sub-region. The data content also includes hazard prediction labels and multiple predicted piping locations when the hazard prediction labels exist, as well as the predicted piping type and predicted hazard level corresponding to each predicted piping location.

[0076] In this embodiment, if the currently processed sub-region is a special risk sub-region, since such regions lack clear hazard type labels and sufficient historical data to support model training, a manual hazard identification method is adopted. Relevant personnel will combine the environmental parameters reflected by the regional environmental vector of the sub-region with the visual characteristics presented by the monitoring station image set of all embankment stations to conduct a comprehensive and detailed analysis and judgment. Through professional knowledge and practical experience, they will identify whether there are signs of piping, and ultimately determine the regional hazard data for this special risk sub-region.

[0077] In this embodiment, for each piping location recorded in the isolated piping data, information about the levee body at that location, along with previously collected isolated image monitoring data and isolated environmental monitoring data, are collected. Simultaneously, from the historical piping sub-data corresponding to all isolated piping data, all historical piping sub-data that are completely consistent with the levee body location of the current isolated piping to be determined are precisely extracted. This is because the geological structure, soil characteristics, stress state, and seepage path of the same levee body location, such as the outer side of the levee toe, the middle section of the water-facing slope, and the area around structures penetrating the levee, are highly consistent, making the inducing conditions, evolution patterns, and performance characteristics of piping more comparable and providing the most suitable historical reference for determining the current isolated piping. Subsequently, these extracted historical piping sub-data are clustered using historical environmental vectors, piping type labels, and hazard level labels as clustering dimensions. Historical data with similar environmental conditions and piping characteristics are clustered into several typical categories. Each category represents the corresponding pattern of environmental cause-piping type-hazard level for that levee body location, and the cluster center of each cluster category is determined. Next, a similarity analysis is conducted, comparing the isolated environmental monitoring data of the current isolated piping with the core environmental features of each cluster category to calculate the similarity at the parameter level. Simultaneously, visual features are extracted from the isolated image monitoring data and matched with the piping appearance features associated with the historical piping sub-data of the corresponding cluster category. This comprehensive similarity score between the current isolated piping and each cluster category is then calculated. If the maximum similarity score exceeds a preset threshold, it indicates that the environmental conditions and visual features of the current isolated piping highly match a certain type of historical typical piping pattern, and the predicted hazard label indicates the presence of a hazard. The isolated hazard data directly adopts the piping type label and hazard level label of that cluster category. If none of the maximum similarity scores reach the preset threshold, it indicates that the characteristics of the current isolated piping differ significantly from all historical typical patterns under this levee section, and there is insufficient evidence to determine the presence of a hazard. The predicted hazard label indicates no hazard, and the isolated hazard data is empty.

[0078] In this embodiment, all data is integrated and summarized, collecting regional hazard data for all sub-regions within the potential risk area of ​​the dike to be investigated. This includes hazard data for potential risk sub-regions, hidden investigation areas where hidden dangers have been discovered, special risk sub-regions, and potential merged sub-regions. Simultaneously, predicted hazard labels and corresponding isolated hazard data for all piping locations in isolated piping data are collected. These data are then integrated to determine the dike hazard data for all piping risk conditions of the dike to be investigated.

[0079] The beneficial effects of the above technologies are as follows: acquiring and analyzing data on potential hazards and historical piping data of the dikes to be investigated, determining the initial type of hazard data, risk distance threshold, and potential risk area for each hazard type label, and determining the initial potential risk area of ​​the dikes to be investigated. This allows for differentiated adaptation to hazard investigation, balancing accuracy with adaptability to special scenarios, strengthening risk traceability, achieving comprehensive risk coverage, and improving the pertinence, comprehensiveness, and reliability of dike hazard assessment. Example 8:

[0080] This invention provides a method and system for detecting piping hazards based on image recognition, used to execute any one of the image recognition-based piping hazard detection methods in Examples 1 to 7, with reference to... Figure 2 ,include: Acquisition Module: Acquires and analyzes the dam hazard data and historical piping data of the dam to be investigated, determines the initial type hazard data, risk distance threshold and type potential risk area for each hazard type label, and determines the initial potential risk area of ​​the dam to be investigated. Determine the module: Based on the historical piping data of the embankment to be investigated, optimize the initial potential risk area, determine isolated piping data, investigation type hazard data with multiple hazard type labels, and merged type data with multiple merged type labels, and determine the potential risk area of ​​the embankment to be investigated. Construction module: Based on the investigation type hazard data of each hazard type label of the dam to be investigated, the combined type data of each combined type label, and the historical piping data, construct the piping hazard model for each hazard type label, and construct the piping hazard model for each combined type label. Acquisition module: Acquires dam image monitoring data based on image sensors, and acquires dam environmental monitoring data based on environmental sensor arrays; Hazard Module: Based on the piping hazard model with all hidden danger type labels of the dike to be investigated, the piping hazard model with all combined type labels, dike image monitoring data, and dike environmental monitoring data, the module conducts hazard investigation on the dike to be investigated and determines the dike hazard data.

[0081] The beneficial effects of the above technology are as follows: By analyzing the dam hazard data and historical piping data of the dams to be investigated, the initial type hazard data, risk distance threshold, and potential risk area of ​​each hazard type label are determined. The initial potential risk area of ​​the dams to be investigated is identified and optimized. Isolated piping data, hazard data for multiple hazard type labels, and combined type data for multiple combined type labels are determined. The potential risk area of ​​the dams to be investigated is identified. Piping hazard models for each hazard type label and each combined type label are constructed. Dam image monitoring data and dam environmental monitoring data are collected. Hazard investigation is conducted on the dams to be investigated, and dam hazard data is determined. This technology can comprehensively cover conventional hazards, special risks, and isolated piping, avoiding risk omissions, accurately adapting to single and complex hazard scenarios, improving the targeting of hazard identification, strengthening the correlation and reliability of hazard judgment, improving investigation efficiency and accuracy, and achieving comprehensive and seamless prevention and control of piping hazards.

[0082] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for investigating piping hazard based on image recognition, characterized in that, include: S1: Acquire and analyze the dam hazard data and historical piping data of the dam to be investigated, determine the initial type hazard data, risk distance threshold and type potential risk area for each hazard type label, and determine the initial potential risk area of ​​the dam to be investigated. S2: Based on the historical piping data of the embankment to be investigated, the initial potential risk area is optimized, and isolated piping data, investigation type hazard data with multiple hazard type labels, and merged type data with multiple merged type labels are determined, and the potential risk area of ​​the embankment to be investigated is determined. S3: Based on the investigation type hazard data of each hazard type label of the dam to be investigated, the merged type data of each merged type label, and the historical piping data, construct the piping hazard model for each hazard type label, and construct the piping hazard model for each merged type label. S4: Collect image monitoring data of the dam to be investigated based on image sensors, and collect environmental monitoring data of the dam to be investigated based on environmental sensor groups; S5: Based on the piping hazard model of all hidden danger types labeled on the dike to be investigated, the piping hazard model of all combined type labels, the dike image monitoring data, and the dike environmental monitoring data, conduct hazard investigation on the dike to be investigated and determine the dike hazard data.

2. The method for investigating piping hazard based on image recognition according to claim 1, characterized in that, Acquire and analyze dam hazard data and historical piping data for the dams to be investigated, determine the initial hazard data, risk distance threshold, and potential risk area for each hazard type label, and identify the initial potential risk area for the dams to be investigated, including: Acquire data on potential hazards and historical piping data for the dikes to be investigated. The dike hazard data includes multiple locations of hazards on the dike body, as well as hazard type and hazard severity labels for each location. The historical piping data includes sub-data on multiple piping events, which includes piping location, dike body section, dike station number, historical piping time, piping type label, hazard level label, historical environmental vector, and historical image recognition vector. The hazard level labels include Level 1, Level 2, Level 3, and Level 4. Based on the hazard type labels of all hazard locations in the dam hazard data, all hazard locations in the dam body are divided, and the initial type hazard data for each hazard type label is determined. The initial type hazard data includes multiple hazard locations in the dam body and the hazard degree label for each hazard location. Based on the initial type hazard data of each hazard type label of the dam to be investigated, as well as the piping location, historical piping time, and hazard level label in all historical piping sub-data in the historical piping data, calculate the risk distance threshold of each hazard type label of the dam to be investigated. Using the location of each hazard in the initial type hazard data of each hazard type label of the dam to be investigated as the center and the risk distance threshold of each hazard type label of the dam to be investigated as the radius, the potential risk sub-region of each hazard location in the initial type hazard data of each hazard type label of the dam to be investigated is determined. Based on the potential risk sub-regions of all hazard locations in the initial type hazard data of each hazard type label of the dam to be investigated, the type potential risk area of ​​each hazard type label of the dam to be investigated is determined. Based on the type potential risk areas of all hazard type labels of the dam to be investigated, the initial potential risk area of ​​the dam to be investigated is determined.

3. The method for investigating piping hazard based on image recognition according to claim 2, characterized in that, The initial potential risk area is optimized based on historical piping data of the embankment to be investigated, including: Based on the piping locations of all historical piping sub-data in the historical piping data and the initial potential risk area of ​​the embankment to be investigated, determine whether there are piping locations that do not belong to the initial potential risk area, and determine the initial isolated data based on the historical piping sub-data of all piping locations that do not belong to the initial potential risk area. Determine whether the number of historical piping sub-data in the initial isolated data is below a preset percentage threshold. If so, determine that the initial isolated data is isolated piping data. Otherwise, optimize the risk distance threshold for each hazard type label of the embankment to be investigated. Based on the optimized risk distance thresholds for all hazard type labels and the location of each hazard in the embankment body in the initial hazard data for each hazard type label, determine the optimized initial potential risk area and identify the isolated piping data.

4. The method for investigating piping hazard based on image recognition according to claim 3, characterized in that, Identify isolated piping data, hazard data with multiple hazard type labels, and merged data with multiple combined type labels, and determine the potential risk areas of the dams to be investigated, including: The largest potential risk sub-region in the initial potential risk area is selected as the implicit judgment area for isolated piping data; Based on the piping locations of all historical piping sub-data in the implicit judgment area and isolated piping data, determine whether a implicit judgment area has a preset number or more piping locations. If it does, determine each implicit judgment area with a preset number or more piping locations as an implicit investigation area, and determine the historical piping sub-data of all piping locations that do not belong to any implicit judgment area as the adjusted isolated piping data. If it does not, do not adjust the isolated piping data. For each hidden investigation area, a hazard investigation is carried out. If a hazard is found, the location of the hazard on the embankment, the hazard type label, and the hazard severity label of the hidden investigation area are determined. If no hazard is found, the hidden investigation area is determined to be a special risk sub-area. Determine whether there is any overlap between all potential risk sub-regions and all hidden investigation areas where hazards have been discovered in the initial potential risk area. For each potential risk sub-region and hidden investigation area with regional overlap, merge them to determine each potential merged sub-region. Then, determine the merge type label of each potential merged sub-region based on the hazard type labels corresponding to the two or more potential risk sub-regions and hidden investigation areas with regional overlap. Based on the initial type hazard data of each hazard type label, all potential risk sub-regions without regional overlap and all hidden investigation areas where hazards are found without regional overlap, the location of hazard in the dike body, hazard type label, and hazard severity label are determined, and the investigation type hazard data of each hazard type label of the dike to be investigated is determined. The investigation type hazard data includes multiple hazard locations in the dike body and the hazard severity label of each hazard location. Based on the merging type label of each potential merging sub-region, all potential merging sub-regions are divided, and the merging type data of each merging type label is determined. The merging type data includes multiple potential merging sub-regions, multiple embankment hazard locations in each potential merging sub-region, and hazard degree labels for each embankment hazard location. Based on all potential risk sub-regions without regional overlap in the initial potential risk area, all hidden investigation areas without regional overlap where hidden dangers were found, all special risk sub-regions where no hidden dangers were found, and all potential merged sub-regions, the potential risk areas of the dikes to be investigated are determined.

5. The method for investigating piping hazard based on image recognition according to claim 4, characterized in that, Based on the hazard type data for each hazard type label of the dike to be investigated, the combined type data for each combined type label, and historical piping data, a piping hazard model is constructed for each hazard type label, and a piping hazard model is also constructed for each combined type label, including: Based on the potential risk sub-area or hidden investigation area of ​​each hazard location in the hazard data of each hazard type label of the dam to be investigated, and the piping location of all historical piping sub-data in the historical piping data, the historical hazard piping data of each hazard location in the dam to be investigated is determined. The historical environment vector and historical image recognition vector of all historical piping sub-data in the historical hidden danger piping data of all potential risk sub-regions or hidden investigation areas of all hidden danger locations in the investigation type data of each hidden danger type label of the embankment to be investigated are used as the input of the piping hazard model of each hidden danger type label. The piping location, piping type label, and hazard level label of all historical hidden danger piping sub-data in the historical hidden danger piping data of all potential risk sub-regions or hidden investigation areas of all hidden danger locations in the investigation type data of each hidden danger type label of the embankment to be investigated are used as the output of the piping hazard model of each hidden danger type label. Thus, the piping hazard model of each hidden danger type label is constructed. Based on the potential risk sub-regions or hidden investigation areas of all potential hazard locations in the merging type data of each potential merging sub-region of each merging type label of the dam to be investigated, and the piping locations of all historical piping sub-data in the historical piping data, the historical merged piping data of each potential merging sub-region in the merging type data of each merging type label of the dam to be investigated is determined. The piping hazard model for each hazard type label is constructed by taking the locations of all potential hazard locations in the merging sub-regions of the merging type data for each dam to be investigated, the historical environmental vectors of all historical piping sub-data in the historical merging piping data of all potential merging sub-regions, and the historical image recognition vectors as inputs. The piping hazard model for each hazard type label is constructed by taking the piping location, piping type label, and hazard level label of all historical piping sub-data in the historical merging piping data of all potential merging sub-regions of the merging type data for each dam to be investigated as outputs.

6. The method for investigating piping hazard based on image recognition according to claim 1, characterized in that, Image monitoring data of the dam to be inspected was collected using image sensors, and environmental monitoring data of the dam to be inspected was collected using an environmental sensor array, including: Based on image sensors, regional image monitoring data of each sub-region in the potential risk area of ​​the dam to be investigated is collected. The regional image monitoring data includes multiple monitoring images and the monitoring position and shooting angle of each monitoring image. The sub-regions are: potential risk sub-region, hidden investigation area where hidden dangers are found, special risk sub-region, and potential merged sub-region. Regional environmental monitoring data for each sub-region within the potential risk area of ​​the dam to be investigated is collected based on an environmental sensor array. Isolated image monitoring data for each piping location in the isolated piping data of the embankment to be investigated is collected based on image sensors. The isolated image monitoring data includes multiple monitoring images and the monitoring location and shooting angle of each monitoring image. Isolated environmental monitoring data for each piping location in the isolated piping data of the dam to be investigated, collected from the environmental sensor array; Based on the regional image monitoring data of all sub-regions in the potential risk area of ​​the dam to be investigated and the isolated image monitoring data of all piping locations in the isolated piping data, the dam image monitoring data of the dam to be investigated is determined. Based on the regional environmental monitoring data of all sub-regions in the potential risk area of ​​the dam to be investigated, and the isolated environmental monitoring data of all piping locations in the isolated piping data, the dam environmental monitoring data of the dam to be investigated is determined.

7. The method for investigating piping hazard based on image recognition according to claim 6, characterized in that, Based on the piping hazard model with all hazard type labels for the dike to be investigated, the piping hazard model with all combined type labels, dike image monitoring data, and dike environmental monitoring data, a hazard investigation is conducted on the dike to be investigated to determine the dike hazard data, including: All monitoring images in the regional image monitoring data of each sub-region of the potential risk area of ​​the dam to be investigated are preprocessed. All preprocessed monitoring images with the same dam chainage in the regional image monitoring data of each sub-region are image aligned to determine the monitoring chainage image set of each dam chainage in each sub-region of the potential risk area of ​​the dam to be investigated. Feature extraction is performed on all monitoring images in the monitoring chainage image set of each chainage of each sub-region in the potential risk area of ​​the dam to be investigated, and the chainage image recognition vector of each chainage of each chainage in each sub-region in the potential risk area of ​​the dam to be investigated is determined. Feature extraction is performed on the environmental monitoring data of each sub-region within the potential risk area of ​​the dam to be investigated, and the regional environmental vector of each sub-region within the potential risk area of ​​the dam to be investigated is determined. If the sub-region is a potential risk sub-region or a hidden investigation area where hidden dangers have been discovered, the regional environmental vector of the sub-region and the set of monitoring station images of all dike station numbers are input into the piping hazard model of the hidden danger type label of the potential risk sub-region or the hidden investigation area where hidden dangers have been discovered. Based on the piping hazard model, the regional hazard data of the sub-region is determined. The regional hazard data includes hazard prediction labels and multiple predicted piping locations when the hazard prediction labels exist, the predicted piping type of each predicted piping location, and the predicted hazard level. If the sub-region is a potential merged sub-region, the regional environment vector of the sub-region and the set of monitoring station images of all dam station numbers are input into the piping hazard model of the merging type label of the potential merged sub-region corresponding to the sub-region, and the regional hazard data of the sub-region is determined based on the piping hazard model. If the sub-region is a special risk sub-region, based on the regional environment vector of the sub-region and the monitoring station image set of all dam station numbers, manual hazard identification is performed to determine the regional hazard data of the sub-region; Based on the dike body part, isolated image monitoring data, isolated environmental monitoring data, and the dike body part, piping type label, hazard level label, and historical environmental vector of all historical piping sub-data corresponding to all piping locations in all isolated piping data, the predicted hazard label and isolated hazard data for each isolated piping are determined. The predicted hazard label includes whether the hazard exists or not. If the predicted hazard label is "hazard exists", the isolated hazard data includes the predicted piping type and the predicted hazard level; otherwise, the isolated hazard data is empty. Based on the regional hazard data of all sub-regions in the potential risk area of ​​the dam to be investigated, as well as the predicted hazard labels and isolated hazard data of all piping locations in the isolated piping data, the dam hazard data of the dam to be investigated is determined.

8. A method and system for investigating piping hazard based on image recognition, characterized in that, A method for investigating piping hazard based on image recognition as described in any one of claims 1 to 7, comprising: Acquisition Module: Acquires and analyzes the dam hazard data and historical piping data of the dam to be investigated, determines the initial type hazard data, risk distance threshold and type potential risk area for each hazard type label, and determines the initial potential risk area of ​​the dam to be investigated. Determine the module: Based on the historical piping data of the embankment to be investigated, optimize the initial potential risk area, determine isolated piping data, investigation type hazard data with multiple hazard type labels, and merged type data with multiple merged type labels, and determine the potential risk area of ​​the embankment to be investigated. Construction module: Based on the investigation type hazard data of each hazard type label of the dam to be investigated, the combined type data of each combined type label, and the historical piping data, construct the piping hazard model for each hazard type label, and construct the piping hazard model for each combined type label. Acquisition module: Acquires dam image monitoring data based on image sensors, and acquires dam environmental monitoring data based on environmental sensor arrays; Hazard Module: Based on the piping hazard model with all hidden danger type labels of the dike to be investigated, the piping hazard model with all combined type labels, dike image monitoring data, and dike environmental monitoring data, the module conducts hazard investigation on the dike to be investigated and determines the dike hazard data.

Images