A method, medium, and electronic equipment for object-oriented collection of inspection hazards
By dividing the space into multi-level spatial units and performing multi-source analysis, and by calculating the spatial mapping confidence level using location, address, image, and historical evidence, the problem of inaccurate hazard identification in the existing inspection system is solved, and the accurate attribution and management of hazards are achieved.
Patent Information
- Application Number
- CN202611436622.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-16
- Publication Date
- 2026-11-17
AI Technical Summary
Existing inspection systems struggle to accurately identify the spatial location of the same hazard, leading to redundant statistics and low management efficiency. They are unable to map and aggregate data across different dimensions, lack conflict resolution for evidence from multiple sources, and cannot support spatial hotspot analysis and facility vulnerability identification.
By dividing the space into multi-level spatial units, obtaining evidence vectors through multi-source analysis, calculating spatial mapping confidence, and comprehensively utilizing location, address, image, facility, and historical evidence, the inspection records are automatically aggregated into the same potential hazard object.
It enables precise spatial attribution management of potential hazards, avoids duplicate statistics and attribution errors, supports spatial hotspot analysis and facility vulnerability identification, and improves the traceability of inspection data and management efficiency.
Smart Images

Figure CN123115291A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of big data processing, specifically relating to a method, medium, and electronic equipment for object-oriented collection of inspection hazards. Background Technology
[0002] Existing inspection systems often manage single inspection records, single reported events, or single rectification tasks. When inspectors report the same hazard at different times, from different entrances, or from different map locations, it is often difficult to determine whether it points to the same spatial location, the same facility component, or the same persistent problem. This results in the same hazard being repeatedly counted, repeatedly assigned, or unable to be transformed into a stable object over a long period of time.
[0003] Existing technologies for spatial location identification often rely on single coordinates or manually selected points. In actual inspections, GPS is prone to drift within buildings, underground spaces, densely populated streets, and between buildings in industrial parks; existing textual addresses contain aliases, abbreviations, misspellings, and inconsistencies in floor descriptions; while photographs can depict facilities and components, they are difficult to directly pinpoint to standard spatial units. Therefore, existing methods for attributing hazards based solely on single location information are prone to errors.
[0004] Existing hazard management technologies lack object-oriented archives for spatial units and facility components. Current solutions can only record a single discovery, rectification, or closure, failing to describe the changing status of a hazard from initial discovery, persistence, rectification, recurrence, expansion, to eventual closure. Furthermore, current statistical methods can only summarize quantities, unable to accurately answer the true persistence status of hazards in a specific building, passageway, facility, or area of responsibility.
[0005] Existing technologies are inadequate in handling conflicts between evidence from multiple sources. For example, coordinates may fall on building A, textual descriptions may point to building B, photographs may identify a fire hydrant in area C, and historical rectification records may correspond to point D. In the absence of a unified mapping, scoring, and conflict resolution scheme, existing technologies either directly adopt one source or rely entirely on manual judgment, which affects both efficiency and data collection accuracy.
[0006] Current hazard statistics rarely reflect the hierarchical relationships between spatial units. A hazard may belong to a specific facility component, or it may belong to a room, floor, building, courtyard, grid, or area of responsibility. Because current technologies cannot map and aggregate across different dimensions, they struggle to support spatial heat zone analysis, responsibility scope verification, facility vulnerability identification, and tracking of recurring hazards. Summary of the Invention
[0007] In view of this, in order to overcome the shortcomings of the prior art, the present invention provides a method, medium and electronic equipment for object-oriented collection of inspection hazards.
[0008] In a first aspect, the present invention provides a method for object-oriented collection of inspection hazards, the method of the present invention comprising:
[0009] A method for object-oriented collection of potential hazards during inspections, characterized by the following steps:
[0010] Based on spatial boundary and hierarchical relationship data, multi-level spatial units are divided to obtain a spatial unit hierarchy library;
[0011] Receive patrol records, perform multi-source analysis on the patrol records, and obtain evidence vectors including location evidence, address evidence, image evidence, facility evidence, historical evidence, and time evidence;
[0012] Based on the location evidence and address evidence in the evidence vector, a set of candidate spatial units is recalled from the spatial unit hierarchy library. The spatial mapping confidence of the patrol record and each candidate spatial unit is calculated respectively. The candidate spatial unit with the highest spatial mapping confidence score is determined as the target spatial unit.
[0013] Based on the target space unit, determine whether the inspection record corresponds to an existing hidden danger object; if it corresponds, the inspection record is assigned to the corresponding hidden danger object; if it does not correspond, a new hidden danger object is created.
[0014] Furthermore, the step of dividing spatial units into multiple levels based on spatial boundary and hierarchical relationship data to obtain a spatial unit hierarchy library includes:
[0015] Based on spatial boundary and floor structure data, spatial units are divided according to the granularity of region, building, floor, room and facility point, and a parent-child hierarchical relationship is established between spatial units;
[0016] For each spatial unit, extract hierarchical paths, normalized address fragments, geometric location buckets, facility types, component orientations, and historical high-frequency keywords to obtain the spatial fingerprint of each spatial unit;
[0017] Integrate all spatial units and their corresponding spatial fingerprints to obtain a spatial unit hierarchy library.
[0018] Furthermore, the multi-source analysis of the patrol records to obtain an evidence vector including location evidence, address evidence, image evidence, facility evidence, historical evidence, and time evidence includes:
[0019] Address fragments and facility names are extracted from the textual descriptions in the inspection records to obtain address evidence;
[0020] Based on the location data in the patrol records, location evidence is obtained;
[0021] Scene category and sign information are extracted from the images in the patrol records to obtain image evidence;
[0022] Component information is obtained from the facility codes in the inspection records to obtain facility evidence;
[0023] Historical evidence is obtained based on the historical data in the aforementioned patrol records;
[0024] The address evidence, location evidence, image evidence, facility evidence, and historical evidence are combined to obtain an evidence vector.
[0025] Further, the step of recalling a set of candidate spatial units from the spatial unit hierarchy based on the location evidence and address evidence in the evidence vector, calculating the spatial mapping confidence score between the inspection record and each candidate spatial unit, and determining the candidate spatial unit with the highest spatial mapping confidence score as the target spatial unit includes:
[0026] Based on the location evidence and address evidence in the evidence vector, a set of candidate spatial units is recalled from the spatial unit hierarchical library;
[0027] For each candidate spatial unit, the geometric relationship, address matching relationship, image recognition relationship, facility correspondence relationship, historical association relationship, time matching relationship and evidence conflict relationship between the inspection record and the candidate spatial unit are converted into mapping features to obtain a set of mapping features;
[0028] The mapping feature set is scored according to configurable rules or ranking models to obtain the spatial mapping confidence of each candidate spatial unit.
[0029] The candidate spatial unit with the highest confidence score in spatial mapping is determined as the target spatial unit.
[0030] Further, the step of recalling a set of candidate spatial units from the spatial unit hierarchy library based on the location evidence and address evidence in the evidence vector includes:
[0031] Based on the location evidence, determine the geographical coordinate range of the patrol record;
[0032] Based on the address evidence, address keywords are obtained from the patrol records;
[0033] In the spatial unit hierarchical library, spatial units whose geographical range covers the geographical coordinate range are recalled to obtain a preliminary set of spatial units;
[0034] Based on the address keywords, the preliminary spatial unit set is matched and filtered to obtain a candidate spatial unit set.
[0035] Furthermore, for each candidate spatial unit, the geometric relationship, address matching relationship, image recognition relationship, facility correspondence relationship, historical association relationship, time matching relationship, and evidence conflict relationship between the inspection record and the candidate spatial unit are converted into mapping features to obtain a set of mapping features, including:
[0036] Based on the location evidence, calculate the geometric matching degree between the location of the patrol record and the candidate spatial unit;
[0037] Based on the address evidence, calculate the address matching degree between the address text in the inspection record and the standard address and alias of the candidate spatial unit;
[0038] Based on the image evidence, calculate the image matching degree between the image recognition results in the inspection record and the known facility features of the candidate spatial unit;
[0039] Based on the facility evidence, calculate the correspondence between the facility codes in the inspection records and the facility records of the candidate space units;
[0040] Based on the historical evidence, calculate the historical correlation between the inspection records and the historical hazard records of the candidate spatial units;
[0041] Based on the time evidence, calculate the time matching degree between the patrol time of the patrol record and the time of the candidate spatial unit;
[0042] The degree of evidence conflict is obtained by detecting whether the spatial units pointed to by the location evidence, address evidence, image evidence, facility evidence, and historical evidence are consistent in different dimensions.
[0043] By integrating the geometric matching degree, address matching degree, image matching degree, facility correspondence degree, historical correlation degree, time matching degree, and evidence conflict degree, a mapping feature set is obtained.
[0044] Further, the step of scoring the mapping feature set according to configurable rules or a ranking model to obtain the spatial mapping confidence of each candidate spatial unit includes:
[0045] Based on the spatial type, patrol task type, or equipment acquisition accuracy, weights are assigned to each mapping feature in the mapping feature set to obtain a weighted scoring rule.
[0046] According to the weighted scoring rule, the mapping feature set is weighted and calculated to obtain the spatial mapping confidence score of each candidate spatial unit; or...
[0047] The mapping feature set is input into a pre-trained ranking model to obtain the spatial mapping confidence of each candidate spatial unit; the ranking model is a machine learning model trained with the mapping features corresponding to historical inspection records as training samples and the spatial attribution results confirmed by manual review as training labels.
[0048] Further, determining the target spatial unit from the candidate spatial unit set based on the highest spatial mapping confidence includes:
[0049] When the highest spatial mapping confidence score is greater than the first threshold and the difference between the highest and second highest spatial mapping confidence scores is greater than the distinction threshold, the candidate spatial unit corresponding to the highest spatial mapping confidence score is automatically identified as the target spatial unit.
[0050] When the highest spatial mapping confidence is between the second threshold and the first threshold, or when the spatial mapping confidence of multiple candidate spatial units is close, the inspection record is marked as pending review.
[0051] When the spatial mapping confidence of all candidate spatial units is lower than the second threshold, a spatial unit missing indication is generated.
[0052] In a second aspect, the present invention provides a storage medium comprising: computer program instructions for storing computer program instructions which, when executed by a processor, implement the method described in any of the preceding claims.
[0053] Thirdly, the present invention provides an electronic device including a memory and a processor, characterized in that the memory is used to store one or more computer program instructions, wherein the one or more computer program instructions are executed by the processor to describe the method described in any of the preceding embodiments.
[0054] The technical solution of this invention elevates the spatial attribution of potential hazards from a single coordinate point to a computable multi-level spatial structure by dividing the space into multi-level spatial units and obtaining a spatial unit hierarchy library. The spatial unit hierarchy library covers multiple dimensions such as region, building, floor, room, and facility point, enabling hazards to be located and aggregated according to different management needs in corresponding dimensions. This solves the problem in existing technologies where hazard statistics can only be summarized by quantity and cannot accurately answer the true existence status of hazards within a specific building, floor, or facility point. The technical solution of this invention obtains evidence vectors including location evidence, address evidence, image evidence, facility evidence, historical evidence, and time evidence by performing multi-source analysis on inspection records. This transforms inspection records from raw, multi-format mixed data into a structured, computable, multi-dimensional evidence representation. Compared to existing technologies that rely solely on a single coordinate or manually selected point for spatial positioning, this invention, through the joint representation of multi-source evidence, allows subsequent spatial mapping to comprehensively utilize various information such as coordinates, address text, image recognition, facility ledgers, and historical records, reducing the impact of single-source bias on positioning accuracy. The technical solution of this invention recalls a set of candidate spatial units from a spatial unit hierarchy library, calculates the spatial mapping confidence level between inspection records and each candidate spatial unit, and obtains the target spatial unit based on the highest spatial mapping confidence level, thus achieving more accurate spatial attribution management. In the recall phase, this invention can cover all potentially matching spatial units, avoiding the problem of missing potential hazards due to bias in single pieces of evidence. In the mapping scoring phase, this invention determines the target spatial unit by comprehensively calculating scores from multiple pieces of evidence, solving the problem in existing technologies where GPS drifts easily inside buildings, underground spaces, and densely populated areas, leading to incorrect hazard attribution. The technical solution of this invention determines whether an inspection record corresponds to an existing hazard object based on the target spatial unit. If they correspond, they are assigned to an existing hazard object; otherwise, a new hazard object is created, transforming the inspection management object from a record to the management of hazard objects. In this invention, records pointing to the same spatial location, the same facility component, or the same problem type in multiple inspections are automatically grouped into the same hazard object, rather than being managed separately as isolated records. This solves the problem of repeated statistics and assignment of the same hazard in existing technologies, enabling inspection data to be continuously transformed into traceable and analyzable management data. Attached Figure Description
[0055] The above and other objects, features and advantages of the present invention will become clearer from the following description of embodiments of the invention with reference to the accompanying drawings, in which:
[0056] Figure 1 This is a flowchart of a method for objectifying and aggregating potential hazards according to an embodiment of the present invention;
[0057] Figure 2This is a schematic diagram of an object-oriented data collection system for potential inspection hazards according to an embodiment of the present invention;
[0058] Figure 3 This is a schematic diagram of an electronic device implementing the present invention. Detailed Implementation
[0059] The present invention is described below based on embodiments, but the invention is not limited to these embodiments. In the detailed description of the invention below, certain specific details are described in detail. Those skilled in the art will fully understand the invention even without these details. To avoid obscuring the essence of the invention, well-known methods, processes, flows, elements, and circuits are not described in detail.
[0060] Furthermore, those skilled in the art should understand that the accompanying drawings provided herein are for illustrative purposes only and are not necessarily drawn to scale.
[0061] Unless the context explicitly requires it, words such as "including" or "contains" in the instruction manual should be interpreted as including rather than exclusive or exhaustive; that is, meaning "including but not limited to".
[0062] In the description of this invention, it should be understood that the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance. Furthermore, in the description of this invention, unless otherwise stated, "a plurality of" means two or more.
[0063] Example 1
[0064] like Figure 1 As shown, the method of the present invention includes steps 100-500.
[0065] Step 100: Based on the spatial boundary and hierarchical relationship data, divide the space into multiple levels of spatial units to obtain a spatial unit hierarchy library;
[0066] Step 200: Receive patrol records, perform multi-source analysis on the patrol records, and obtain evidence vectors including location evidence, address evidence, image evidence, facility evidence, historical evidence, and time evidence;
[0067] Step 300: Based on the location evidence and address evidence in the evidence vector, recall the set of candidate spatial units from the spatial unit hierarchy library, calculate the spatial mapping confidence of the patrol record and each candidate spatial unit, and determine the candidate spatial unit with the highest spatial mapping confidence score as the target spatial unit.
[0068] Step 400: Based on the target space unit, determine whether the inspection record corresponds to an existing hidden danger object; if it corresponds, then assign the inspection record to the corresponding hidden danger object; if it does not correspond, then create a new hidden danger object.
[0069] The steps in this invention will be described in detail below.
[0070] Step 100: Based on the spatial boundary and hierarchical relationship data, divide the space into multiple levels of spatial units to obtain a spatial unit hierarchy library.
[0071] This invention pre-constructs or dynamically builds a spatial unit hierarchy library. Based on spatial boundaries and floor structure data, this invention divides spatial units into granularities at the level of region, building, floor, room, and facility point, establishing parent-child hierarchical relationships between spatial units. Each spatial unit includes at least a unit identifier, unit type, parent unit identifier, spatial boundary, center point, floor or height information, standard address, address alias, associated facility set, scope of responsibility, and a summary of historical hazards.
[0072] For locations that cannot be represented by precise boundaries, this invention constructs virtual spatial units through spatial positioning descriptions and identifiable descriptions, such as "the west stairwell on the first floor of a certain building," "the fire escape on the north side of basement B2," and "the storm drain grate on the east side of a certain road." Virtual spatial units can inherit the responsibility scope of their parent spatial units while retaining their own aliases, reference images, and adjacent objects. This method of the present invention enables standardized spatial attribution for inspection records lacking precise coordinates in indoor, underground, and semi-open spaces, solving the problem of GPS drift leading to incorrect hazard attribution in existing technologies.
[0073] To support cross-granularity aggregation, this invention extracts hierarchical paths, normalized address fragments, geometric location buckets, facility types, component orientations, and historical high-frequency keywords for each spatial unit, serving as a spatial fingerprint for each unit. The spatial fingerprint of this invention transforms each spatial unit into a computational object that can be quickly retrieved and compared, enabling subsequent inspection records to find potentially matching targets within a large number of spatial units.
[0074] The spatial unit hierarchical library of this invention is not a storage space list, but a structured index library that supports multi-way recall. In this invention, the spatial fingerprint of each spatial unit provides multiple searchable index channels: geometric location buckets enable the invention to quickly filter geographically overlapping spatial units based on coordinate range; normalized address fragments and address aliases enable the invention to perform text matching retrieval based on address keywords; facility type enables the invention to filter based on facility category; and historical high-frequency keywords enable the invention to perform associated retrieval based on location names that repeatedly appear in historical inspections. These index channels of this invention can be used individually or in combination. Recall in this invention refers to: when location evidence and address evidence from an inspection record arrive, this invention retrieves potentially matching spatial units from the spatial unit hierarchical library through the aforementioned index channels. Recall in this invention does not require a single hit on a spatial unit, but rather aims to cover as many potentially matching spatial units as possible. Then, in subsequent steps 320, a mapping feature set is calculated, and in step 330, the mapping feature set is scored to obtain spatial mapping confidence, resulting in precise ranking. Therefore, this broad-first-refinement approach of the present invention can avoid the exclusion of correct candidate spatial units during the recall stage due to a single piece of evidence bias, thereby ensuring the accuracy of subsequent mapping scores.
[0075] This invention comprehensively considers all spatial units and their corresponding spatial fingerprints to obtain a spatial unit hierarchical library. Through this multi-level structure, this invention can locate and aggregate potential hazards at different granularities, thereby solving the problem that existing technologies cannot perform mapping and aggregation between different spatial granularities.
[0076] Step 200: Receive the patrol record, perform multi-source analysis on the patrol record, and obtain an evidence vector including location evidence, address evidence, image evidence, facility evidence, historical evidence, and time evidence.
[0077] After receiving patrol records, this invention performs multi-source analysis to extract various types of evidence required for spatial attribution and object aggregation from multiple information channels. The multi-source analysis of this invention includes: address evidence extraction, location evidence extraction, image evidence extraction, facility evidence extraction, and historical evidence extraction.
[0078] The address evidence extraction method of this invention involves extracting address fragments and facility names from the textual descriptions in inspection records to obtain address evidence. The textual descriptions include notes and speech-to-text transcription. The purpose of address evidence extraction in this invention is to transform unstructured textual information into computable spatial location information.
[0079] The location evidence extraction method of this invention involves obtaining location evidence based on location data from patrol records. Location data includes latitude and longitude, map-selected locations, patrol routes, and check-in points. The location evidence extraction method of this invention considers not only the coordinates themselves but also the accessibility and rationality between the coordinates and spatial units.
[0080] The image evidence extraction method of this invention involves extracting scene category and signage information from images recorded during inspections to obtain image evidence. Extractable information from the images of this invention includes facility category, door number, floor number, QR code, equipment nameplate, and obvious hazard features. The value of the image evidence of this invention lies in its ability to reflect spatial information that coordinates and text cannot accurately express, such as building numbers and facility nameplates appearing in photographs.
[0081] The facility evidence extraction method of this invention involves obtaining component information from the facility codes in inspection records to obtain facility evidence. The facility codes of this invention include QR codes, NFC tags, or ledger numbers. The facility evidence of this invention can provide precise spatial positioning because each facility code has a unique corresponding spatial location in the ledger.
[0082] The historical evidence extraction method of this invention involves obtaining historical evidence based on historical data from patrol records. This historical evidence is generated based on the same patrol person, similar time periods, adjacent locations, similar descriptions, and historical potential hazards, and is used to determine whether the record might be a continuation or recurrence of an existing hazard. This historical evidence allows the invention to associate current records with historical hazard files, avoiding the misjudgment of multiple reports of the same hazard as multiple independent events.
[0083] This invention combines address evidence, location evidence, image evidence, facility evidence, and historical evidence obtained after investigation to obtain an evidence vector. This evidence vector transforms patrol records from raw, multi-formatted data into structured, computable, multi-dimensional evidence, providing a unified data foundation for subsequent spatial mapping and object aggregation.
[0084] Step 300: Based on the location evidence and address evidence in the evidence vector, recall the set of candidate spatial units from the spatial unit hierarchy library, calculate the spatial mapping confidence of the patrol record and each candidate spatial unit, and determine the candidate spatial unit with the highest spatial mapping confidence score as the target spatial unit.
[0085] Step 300 of the present invention includes four sub-steps, namely steps 310-340.
[0086] Step 310: Based on the location evidence and address evidence in the evidence vector, recall the set of candidate spatial units from the spatial unit hierarchy library;
[0087] Step 320: For each candidate spatial unit, convert the geometric relationship, address matching relationship, image recognition relationship, facility correspondence relationship, historical association relationship, time matching relationship and evidence conflict relationship between the inspection record and the candidate spatial unit into mapping features to obtain a set of mapping features;
[0088] Step 330: Based on configurable rules or a ranking model, score the mapping feature set to obtain the spatial mapping confidence of each candidate spatial unit;
[0089] Step 340: The candidate spatial unit with the highest spatial mapping confidence score is determined as the target spatial unit.
[0090] The sub-steps of the present invention will be described in detail below.
[0091] Step 310: Based on the location evidence and address evidence in the evidence vector, recall the set of candidate spatial units from the spatial unit hierarchy library.
[0092] Step 310 of the present invention includes four sub-steps, namely steps 311-314.
[0093] Step 311: Determine the geographical coordinate range of the patrol record based on the location evidence;
[0094] Step 312: Obtain address keywords from the patrol record based on the address evidence;
[0095] Step 313: In the spatial unit hierarchy library, recall spatial units whose geographical range covers the geographical coordinate range to obtain a preliminary set of spatial units;
[0096] Step 314: Based on the address keywords, match and filter the preliminary spatial unit set to obtain a candidate spatial unit set.
[0097] The following is a detailed explanation of each sub-step.
[0098] Step 311: Determine the geographical coordinate range of the patrol record based on the location evidence.
[0099] This invention determines the geographic coordinate range of patrol records based on location evidence. Specifically, the geographic coordinate range can be a circular area formed by adding an error radius to the coordinate points, or it can be the envelope of the patrol trajectory. When the patrol record includes only a single coordinate point, this invention determines the error radius based on the accuracy of the data acquisition equipment and historical drift statistics for that coordinate point; when the patrol record includes a patrol trajectory, this invention takes the convex hull of all points on the trajectory as the geographic coordinate range. By determining the geographic coordinate range, this invention provides coarse-grained spatial filtering conditions for subsequent spatial recall, avoiding the exclusion of correct spatial units from the recall range due to coordinate drift.
[0100] Step 312: Obtain address keywords from the patrol record based on the address evidence.
[0101] This invention obtains address keywords from inspection records based on address evidence. Specifically, address keywords include building names, floor descriptions, room numbers, facility names, and directional terms. This invention standardizes address keywords, for example, unifying "Building 3," "Building #3," and "Building 3" into standardized expressions to improve the accuracy of subsequent address matching. The purpose of extracting address keywords in this invention is to provide text retrieval basis for subsequent address matching and filtering, ensuring that even with significant coordinate deviations, correct candidate spatial units can still be retained through address text.
[0102] Step 313: In the spatial unit hierarchy library, recall spatial units whose geographical range covers the geographical coordinate range to obtain a preliminary spatial unit set.
[0103] This invention, within a spatial unit hierarchical library, uses geometric location bucket indexes from spatial fingerprints to recall spatial units whose geographic range covers the aforementioned geographic coordinate range, thus obtaining a preliminary set of spatial units. In this invention, this step uses coarse-grained spatial filtering to narrow the search scope from all spatial units to a reasonable candidate set. In this step, recall does not require a unique unit hit in one attempt, but rather aims to cover as many possible spatial ranges as possible. The purpose of this approach is to reduce the number of spatial units requiring subsequent mapping feature calculations, thereby improving processing efficiency, while simultaneously avoiding omissions due to coordinate drift through relaxed spatial filtering conditions.
[0104] Step 314: Based on the address keywords, match and filter the preliminary spatial unit set to obtain a candidate spatial unit set.
[0105] This invention uses address keywords to match and filter a preliminary set of spatial units, obtaining a candidate set of spatial units. The matching and filtering process involves comparing address keywords with the standard addresses, address aliases, and historical high-frequency keywords of the candidate spatial units, retaining those with a high degree of matching. The purpose of this invention is to further narrow the broad candidate set retrieved solely based on geographical scope through secondary filtering of address text, ensuring that the candidate spatial units entering the mapping and scoring stage are relevant to the inspection records in both spatial and address aspects, thus improving the accuracy and efficiency of subsequent scoring. This coarse-to-fine recall strategy avoids omissions caused by biases from single pieces of evidence.
[0106] Step 320: For each candidate spatial unit, convert the geometric relationship, address matching relationship, image recognition relationship, facility correspondence relationship, historical association relationship, time matching relationship and evidence conflict relationship between the inspection record and the candidate spatial unit into mapping features to obtain a set of mapping features.
[0107] Step 320 includes 8 sub-steps, namely steps 321-328.
[0108] Step 321: Based on the location evidence, calculate the geometric matching degree between the location of the patrol record and the candidate spatial unit;
[0109] Step 322: Based on the address evidence, calculate the address matching degree between the address text in the inspection record and the standard address and alias of the candidate spatial unit;
[0110] Step 323: Based on the image evidence, calculate the image matching degree between the image recognition results in the inspection record and the known facility features of the candidate spatial unit;
[0111] Step 324: Based on the facility evidence, calculate the correspondence between the facility code in the inspection record and the facility in the component ledger of the candidate space unit;
[0112] Step 325: Based on the historical evidence, calculate the historical correlation between the inspection records and the historical hazard records of the candidate spatial units;
[0113] Step 326: Based on the time evidence, calculate the time matching degree between the patrol time of the patrol record and the candidate spatial unit;
[0114] Step 327: Detect whether the location evidence, address evidence, image evidence, facility evidence, and historical evidence are consistent in different directions to obtain the degree of evidence conflict;
[0115] Step 328: Integrate the geometric matching degree, address matching degree, image matching degree, facility correspondence degree, historical correlation degree, time matching degree, and evidence conflict degree to obtain a mapping feature set.
[0116] The sub-steps of the present invention will be described in detail below.
[0117] Step 321: Calculate the geometric matching degree between the location of the patrol record and the candidate spatial unit based on the location evidence.
[0118] The geometric matching degree of this invention considers factors such as whether the coordinates fall within the boundary of the candidate spatial cell, the distance from the center point, and the degree of deviation from the reachable path. When the coordinates fall within the boundary of the candidate spatial cell, the geometric matching degree is high; when the coordinates are outside the boundary but close to the center point, the geometric matching degree decreases with distance.
[0119] This invention determines the spatial unit to which each piece of evidence in the evidence vector points, and then performs pairwise comparisons to detect any discrepancies in the points of reference between different pieces of evidence. Specifically, location evidence determines the candidate spatial unit to which it points based on its geographical coordinate range; address evidence determines the candidate spatial unit to which it points based on address keyword matching; image evidence determines the candidate spatial unit to which it points based on image recognition results; facility evidence determines the candidate spatial unit to which it points based on the spatial location of the facility code in the ledger; and historical evidence determines the candidate spatial unit to which it points based on the spatial unit number of the historical potential hazard object. When two pieces of evidence point to the same candidate spatial unit, the conflict value of the evidence pair is zero; when two pieces of evidence point to different candidate spatial units, the conflict value of the evidence pair is determined based on the hierarchical distance between the spatial units to which the two pieces of evidence point: if the two spatial units belong to sibling nodes under the same parent unit, the conflict value is low; if the two spatial units belong to different parent branches, the conflict value is high. After calculating the conflict value for each pair of evidence pairs, this invention takes the weighted average of the conflict values of each evidence pair as the degree of evidence conflict. Among them, the conflict between facility evidence and location evidence, and the conflict between image evidence and address evidence are assigned higher weights, because the conflict between these two sets of evidence usually means that there is a substantial contradiction in the location information rather than a difference in expression.
[0120] The geometric matching degree of this invention is used to calculate the geometric proximity between the spatial location of the inspection record and the candidate spatial unit, which is the most direct positioning basis in subsequent scoring.
[0121] Step 322: Based on the address evidence, calculate the address matching degree between the address text in the inspection record and the standard address and alias of the candidate spatial unit.
[0122] The address matching degree of this invention considers whether the building name mentioned in the text description is consistent with the standard address or historical alias of the candidate spatial unit. During matching, this invention uses normalized address keywords to compare the text with the standard address and address alias of the spatial unit. A complete match results in the highest address matching degree, while partial matches decrease according to the proportion of the matched fragments. The purpose of the address matching degree of this invention is to compensate for positioning errors caused by coordinate drift; when coordinates are inaccurate, the address text can still provide effective spatial pointing information.
[0123] Step 323: Based on the image evidence, calculate the image matching degree between the image recognition result in the inspection record and the known facility features of the candidate spatial unit.
[0124] The image matching degree of this invention considers whether the facility category identified in the photograph matches the facility type associated with the candidate spatial unit. In an optional embodiment, image recognition can be performed by computer vision models, OCR, or manual annotation. A high image matching degree is achieved when the building number, facility nameplate, or scene features identified in the photograph match known information about the candidate spatial unit. The image matching degree of this invention provides a third layer of spatial verification beyond coordinates and text, especially in indoor environments where visual information in photographs is often more reliable than coordinates.
[0125] Step 324: Based on the facility evidence, calculate the correspondence between the facility code in the inspection record and the facility in the component ledger of the candidate space unit.
[0126] In this invention, if a facility code directly points to a component within a candidate spatial unit, the facility correspondence is high. The facility codes in this invention include QR codes, NFC identifiers, or ledger numbers, and each facility code has a unique corresponding spatial location in the ledger. The purpose of the facility correspondence in this invention is to provide the most accurate spatial positioning basis; when a facility code explicitly points to a specific spatial unit, its positioning accuracy is far higher than that of coordinates and address text.
[0127] Step 325: Based on the historical evidence, calculate the historical correlation between the inspection records and the historical hazard records of the candidate spatial units.
[0128] In this invention, a high degree of historical correlation is considered when a candidate spatial unit has previously exhibited similar hazards. The historical correlation of this invention also considers factors such as the same inspector, similar time periods, and adjacent locations. The purpose of the historical correlation in this invention is to link the current inspection record with the historical hazard archive, enabling the invention to identify multiple reports of the same hazard and avoid misjudging the same problem as multiple independent events.
[0129] Step 326: Based on the time evidence, calculate the time matching degree between the patrol time of the patrol record and the time of the candidate spatial unit.
[0130] The time matching degree term in this invention considers whether the inspection time falls within the space's open hours and whether it aligns with the inspection plan. In this invention, a higher time matching degree is considered when the inspection time matches the space unit's open hours or the planned inspection period. The purpose of the time matching degree in this invention is to provide supplementary verification of reasonableness; for example, hazards reported outside of open hours may require stronger supporting evidence to confirm attribution.
[0131] Step 327: Detect whether the location evidence, address evidence, image evidence, facility evidence, and historical evidence are consistent in different directions to obtain the degree of evidence conflict.
[0132] This invention determines the spatial unit to which each piece of evidence in the evidence vector points, and then compares them pairwise to detect whether there is a discrepancy in the direction between different pieces of evidence.
[0133] Specifically, location evidence determines the target spatial unit based on its geographic coordinate range; address evidence determines the target spatial unit based on address keyword matching; image evidence determines the target spatial unit based on image recognition results; facility evidence determines the target spatial unit based on the spatial location of the facility code in the ledger; and historical evidence determines the target spatial unit based on the spatial unit number of the historical potential hazard object. When two pieces of evidence point to the same candidate spatial unit, the conflict value of the evidence pair is zero. When two pieces of evidence point to different candidate spatial units, the conflict value of the evidence pair is determined based on the hierarchical distance between the spatial units pointed to by the two pieces of evidence: if the two spatial units belong to sibling nodes under the same parent unit, the conflict value is low; if the two spatial units belong to different parent branches, the conflict value is high. This invention calculates the conflict value for each pair of evidence pairs and takes the weighted average of the conflict values of each evidence pair as the degree of evidence conflict. Among them, the conflict between facility evidence and location evidence, and the conflict between image evidence and address evidence are assigned higher weights, because the conflict between these two sets of evidence usually means that there is a substantial contradiction in the location information rather than a difference in expression.
[0134] In this invention, a higher degree of evidence conflict indicates a greater contradiction between the various pieces of evidence, and the confidence level of the corresponding candidate spatial unit should be reduced accordingly. The purpose of the evidence conflict degree in this invention is to measure the consistency among multiple sources of evidence, so that subsequent scoring can automatically reduce the confidence level in cases of evidence conflict, rather than blindly accepting a particular source.
[0135] Step 328: Integrate the geometric matching degree, address matching degree, image matching degree, facility correspondence degree, historical correlation degree, time matching degree, and evidence conflict degree to obtain a mapping feature set.
[0136] Specifically, the mapping feature set of this invention transforms various relationships between inspection records and candidate spatial units into computable feature representations, providing a data foundation for subsequent spatial mapping scoring. In this invention, each feature in the mapping feature set is an independent scoring criterion, and subsequent steps can flexibly configure the weights of each feature according to different inspection environments and accuracy requirements, rather than being limited to a fixed set of weights. This flexible feature integration method enables this invention to adapt to different inspection environments.
[0137] Step 330: Score the mapping feature set according to configurable rules or ranking models to obtain the spatial mapping confidence of each candidate spatial unit.
[0138] Step 330 of the present invention includes three sub-steps, namely steps 331 to 333.
[0139] Step 331: Based on the space type, patrol task type, or equipment acquisition accuracy, assign weights to each mapping feature in the mapping feature set to obtain weighted scoring rules;
[0140] Step 332: According to the weighted scoring rules, perform weighted calculation on the mapping feature set to obtain the spatial mapping confidence of each candidate spatial unit;
[0141] Alternatively, in step 333, the set of mapping features is input into a pre-trained ranking model to obtain the spatial mapping confidence of each candidate spatial unit; the ranking model is a machine learning model trained with the mapping features corresponding to historical inspection records as training samples and the spatial attribution results confirmed by manual review as training labels.
[0142] The following is a detailed description of each sub-step of the present invention.
[0143] Step 331: Based on the space type, patrol task type, or equipment acquisition accuracy, assign weights to each mapping feature in the mapping feature set to obtain weighted scoring rules.
[0144] Specifically, different spatial types can be configured with different feature priorities. For example, the weights of address matching and image matching can be increased in indoor environments, while the weight of geometric matching can be increased in open areas. Furthermore, the type of inspection task also affects the weight configuration. For instance, the weight of facility correspondence can be increased in fire safety facility inspections, and the weight of historical correlation can be increased in environmental sanitation inspections. In this invention, the accuracy of equipment data acquisition also affects the weights; when the coordinate acquisition accuracy is low, the weight of geometric matching should be reduced accordingly.
[0145] Therefore, the weighted scoring rule enables the present invention to flexibly adjust the scoring strategy according to different inspection environments and equipment conditions, rather than using a fixed weight combination, thereby improving the scoring accuracy of the method of the present invention in various environments.
[0146] Step 332: Perform weighted calculation on the mapping feature set according to the weighted scoring rule to obtain the spatial mapping confidence of each candidate spatial unit; or, Step 333: Input the mapping feature set into a pre-trained ranking model to obtain the spatial mapping confidence of each candidate spatial unit.
[0147] This invention provides two scoring methods. In one optional embodiment, the aforementioned weighted scoring rule is applicable to situations where the patrol environment is relatively fixed and feature weights can be determined manually based on experience. In another optional embodiment, a ranking model is used. Specifically, the ranking model of this invention is a machine learning model trained using the mapping features corresponding to historical patrol records as training samples and the spatial attribution results confirmed by manual review as training labels. The ranking model of this invention can automatically learn the nonlinear relationships between features and can achieve higher scoring accuracy when there is sufficient data. This configurable scoring method allows this invention to adapt to different patrol environments and accuracy requirements, solving the problem that single coordinate positioning in complex environments is prone to errors in existing technologies.
[0148] Step 340: The candidate spatial unit with the highest spatial mapping confidence score is determined as the target spatial unit.
[0149] Step 340 of the present invention includes three sub-steps, namely steps 341-343.
[0150] Step 341: When the highest spatial mapping confidence is greater than the first threshold and the difference between the highest and second highest spatial mapping confidence is greater than the distinction threshold, the candidate spatial unit corresponding to the highest spatial mapping confidence is automatically confirmed as the target spatial unit.
[0151] Step 342: When the highest spatial mapping confidence is between the second threshold and the first threshold, or when the spatial mapping confidence of multiple candidate spatial units is close, the inspection record is marked as pending review.
[0152] Step 343: When the spatial mapping confidence of all candidate spatial units is lower than the second threshold, a spatial unit missing prompt is generated.
[0153] The sub-steps of the present invention will be described in detail below.
[0154] Step 341: When the highest spatial mapping confidence is greater than the first threshold and the difference between the highest and second highest spatial mapping confidence is greater than the distinction threshold, the candidate spatial unit corresponding to the highest spatial mapping confidence is automatically confirmed as the target spatial unit.
[0155] When multiple pieces of evidence consistently point to the same spatial unit, this invention can automatically determine the attribution. The first threshold ensures a sufficiently high absolute level of confidence, while the distinction threshold ensures a sufficient gap between the highest and second-highest scores, meaning there are no other equally likely candidates. The advantage of setting two thresholds simultaneously is that it avoids false positives when the absolute confidence value is low, and it also avoids ambiguous judgments when multiple candidate scores are close, thus maintaining a high accuracy rate during automatic confirmation.
[0156] Step 342: When the highest spatial mapping confidence is between the second threshold and the first threshold, or when the spatial mapping confidence of multiple candidate spatial units is close, the inspection record is marked as pending review.
[0157] In other words, at this point, there are certain conflicts or insufficiencies in the evidence, requiring manual verification to ensure accuracy. Inspection records awaiting review will be added to a review queue, where management personnel will confirm them based on the on-site situation or other supplementary information. This invention avoids the potential for attribution errors caused by blind automatic confirmation by introducing timely human intervention when automatic judgment is insufficient, while also avoiding the inefficiency of relying entirely on manual judgment.
[0158] Step 343: When the spatial mapping confidence of all candidate spatial units is lower than the second threshold, a spatial unit missing prompt is generated.
[0159] At this point, the spatial unit hierarchy library may be missing corresponding spatial units, or the address descriptions may be abnormal, requiring the administrator to supplement or correct the spatial unit data. The spatial unit missing information includes location information from the inspection records, address information, and information on several candidate spatial units with the highest scores during the recall process, for the administrator's reference. Through step 343, this invention can promptly identify coverage gaps in the spatial unit hierarchy library, promote the continuous improvement of spatial unit data, and thus improve the collection coverage rate of this invention.
[0160] Through the above determination, the present invention can achieve a balance between automatic confirmation and manual review, ensuring both processing efficiency under high confidence conditions and judgment accuracy under low confidence conditions.
[0161] Step 400: Based on the target space unit, determine whether the inspection record corresponds to an existing hidden danger object; if it corresponds, then assign the inspection record to the corresponding hidden danger object; if it does not correspond, then create a new hidden danger object.
[0162] After determining or tentatively defining the target spatial unit, the present invention also needs to determine whether the inspection record corresponds to an existing potential hazard.
[0163] This invention constructs object fingerprints for existing potential hazards. The object fingerprint of this invention includes spatial unit number, facility number, component number, hazard category, location bucket, key image features, and effective time window. The function of the object fingerprint in this invention is to condense the core features of potential hazards into a computationally comparable representation. For records lacking facility numbers, this invention obtains a temporary object fingerprint based on spatial unit, hazard category, and location bucket, ensuring object matching can be performed even with incomplete information.
[0164] Subsequently, this invention calculates the object similarity between the inspection record and the existing potential hazard object. The calculation of object similarity comprehensively considers the following factors: spatial unit consistency, judging whether the target spatial unit of the inspection record and the spatial unit number of the potential hazard object are the same. If they are the same, the spatial unit consistency takes the highest value; if they are different but belong to sibling nodes under the same parent spatial unit, the medium value is taken; if they belong to different parent branches, the low value is taken.
[0165] For facility or component consistency, determine whether the facility code in the inspection record matches the facility number of the object with potential hazards. If the facility codes are completely consistent, take the highest value for facility or component consistency. If the facility codes belong to different components of the same facility, take the medium value. If the facility codes are completely different, take the low value.
[0166] The consistency of hazard categories determines whether the hazard category in the inspection record and the hazard category of the hazard object belong to the same category. If they are completely consistent, the highest value is taken; if they belong to different subcategories under the same superior category, the medium value is taken; and if they belong to completely different categories, the lowest value is taken.
[0167] Image feature similarity compares the scene category, facility category, and sign information extracted from the image evidence in the patrol record with the existing image features of the potential hazard, and calculates the cosine similarity between the feature vectors.
[0168] Problem description similarity is calculated by segmenting the textual descriptions in the inspection records with the problem descriptions of the potential hazards and calculating the keyword overlap. The higher the keyword overlap ratio, the higher the problem description similarity.
[0169] Time window proximity is calculated by comparing the time difference between the inspection time recorded in the inspection record and the most recent discovery time of the potential hazard. When the time difference is within the effective time window, the time window proximity decreases as the time difference increases. When the time difference exceeds the effective time window, the time window proximity drops to zero.
[0170] The above factors are weighted and summed according to configurable weights to obtain the object similarity.
[0171] After calculating object similarity, this invention determines whether the exclusion condition is met. The exclusion condition is used to prevent substantially different potential problems from being mistakenly merged into the same object.
[0172] The specific criteria for determining exclusion include: first, determining whether the spatial units of the inspection record and the hazard object are in mutually exclusive areas; second, determining whether the facility numbers of the inspection record and the hazard object are clearly different; third, determining whether the hazard categories of the inspection record and the hazard object belong to non-mergeable categories; and fourth, determining whether the historical confirmation results of the inspection record and the hazard object are marked as different objects. If any of the above criteria are met, the exclusion condition is deemed valid. The purpose of setting exclusion conditions is to ensure that even if two inspection records are similar in textual description, they will not be incorrectly merged as long as they have clear differences in spatial location, facility affiliation, or hazard type, thus guaranteeing the accuracy of data collection.
[0173] If the exclusion condition is met, an object to be reviewed is generated, awaiting manual confirmation.
[0174] If the exclusion condition is not met, the similarity of the objects is used to determine whether the aggregation confirmation condition is met. If it is met, the inspection record is assigned to the corresponding hidden danger object, and the evidence chain and version record of that hidden danger object are updated. If the aggregation confirmation condition is not met and there is no hidden danger object that meets the confirmation condition, a new hidden danger object is created.
[0175] Through the aforementioned aggregation and judgment process, this invention improves the inspection management object from a single record to a potential hazard object. In this invention, records pointing to the same location, the same facility, or the same problem type from multiple inspections are automatically aggregated into the same potential hazard object, rather than being managed separately as isolated records. This improvement solves the problem of repeated statistics and assignment of the same potential hazard in existing technologies, enabling inspection data to be continuously accumulated into valuable management assets.
[0176] The present invention also includes step 500.
[0177] Step 500: Based on the new inspection records, rectification records, review results, item cancellation materials, and time intervals, trigger the status change of the hidden danger object and obtain the status transition result.
[0178] This invention maintains a complete lifecycle status for each potential hazard. Status changes are triggered by new inspection records, rectification records, review results, item closure materials, and time intervals. This invention transforms potential hazards from static records into dynamically trackable objects through status migration, enabling managers to clearly understand the current stage and changes of each hazard.
[0179] The lifecycle state transitions of this invention include the following types:
[0180] Added: When no similar potential hazard object exists and the spatial mapping confidence level meets the confirmation criteria, a new potential hazard object is created, and the state transition result is obtained. The newly created potential hazard object is assigned a unique object number and an initial chain of evidence.
[0181] Ongoing: When a new inspection record is highly similar to an unresolved hazard and the hazard still exists, an additional version record is added, updating the most recent discovery time and evidence chain to obtain the state transition result. Updating the ongoing status continuously enriches the hazard's file, allowing subsequent analysis to utilize the complete evidence chain to understand the hazard's duration and changing trends.
[0182] "Under Rectification": When rectification measures are implemented but lack confirmation of completion, the system is marked as "Under Rectification" and a status transition result is obtained. The "Under Rectification" status indicates that the potential hazard has been identified and is being addressed, but its elimination has not yet been confirmed.
[0183] Completed Items: When the rectification evidence meets the closure rules or the hazard is manually confirmed to be eliminated, the closure version is recorded, and the status transition result is obtained. Complete historical files are retained for closured hazard objects for subsequent query and analysis.
[0184] Recurrence: When a previously resolved hidden danger reappears at the same location and of the same type after the recurrence time window, a recurrence version is created, associated with the original object, and the number of recurrences is calculated to obtain the state transition result. Recurrence identification is one of the key capabilities of this invention, enabling managers to discover recurring hidden dangers that have not been completely resolved, thereby allowing for more thorough remediation measures.
[0185] Expanding the scope of impact: When the same potential hazard is associated with a larger spatial area or more facility components, the scope of impact is expanded to obtain the state transition results. Identifying the expanded state helps managers understand whether the impact of the hazard is spreading and adjust the remediation plan in a timely manner.
[0186] Transfer: When a hazard migrates from its original spatial unit to an adjacent spatial unit, a transfer association is created, and the state transfer result is obtained. Identifying the transfer state enables managers to track the spatial movement trajectory of the hazard.
[0187] Pending Review: When conflicts exist in spatial mapping or object aggregation, a review item is generated awaiting manual confirmation to obtain the state transition result. The pending review status ensures that human intervention can be introduced in a timely manner when automatic judgment is insufficient to guarantee data quality.
[0188] Through the aforementioned state transitions, this invention can generate versioned archives and state transition trajectories for potential hazards, supporting continuous analysis of recurring hazards, long-term unresolved hazards, clusters of similar hazards, and abnormal areas of responsibility.
[0189] Example 2
[0190] like Figure 2 As shown, the present invention also provides a system for object-oriented collection of inspection hazards, including a spatial unit hierarchical library construction module, a multi-source evidence analysis module, a spatial mapping module, and an object collection module.
[0191] The spatial unit hierarchy library construction module is used to divide spatial units into multiple levels based on spatial boundary and hierarchical relationship data, thereby obtaining a spatial unit hierarchy library. Specifically, the module divides spatial units according to the granularity of region, building, floor, room, and facility point, based on spatial boundary and floor structure data, and establishes parent-child hierarchical relationships between spatial units. It extracts hierarchical paths, normalized address fragments, geometric location buckets, facility types, component orientations, and historical high-frequency keywords for each spatial unit to obtain its spatial fingerprint. All spatial units and their corresponding spatial fingerprints are integrated to obtain the spatial unit hierarchy library. For points that cannot be expressed with precise boundaries, the module constructs virtual spatial units through spatial positioning descriptions and identifiable descriptions. These virtual spatial units inherit the responsibility scope of their parent spatial units while retaining their own aliases, reference images, and adjacent objects. The operations performed by the spatial unit hierarchy library construction module are the same as step 100 in the method embodiment, and will not be repeated here.
[0192] The multi-source evidence analysis module receives patrol records and performs multi-source analysis on them to obtain an evidence vector including location evidence, address evidence, image evidence, facility evidence, historical evidence, and time evidence. Specifically, the multi-source evidence analysis module extracts address fragments and facility names from the text description of the patrol records to obtain address evidence; obtains location evidence based on the location data of the patrol records; extracts scene categories and sign information from the images of the patrol records to obtain image evidence; obtains component information from the facility codes of the patrol records to obtain facility evidence; obtains historical evidence based on the historical data of the patrol records; and combines the address evidence, location evidence, image evidence, facility evidence, and historical evidence to obtain an evidence vector. The operations performed by the multi-source evidence analysis module are the same as step 200 in the method embodiment, and will not be repeated here.
[0193] The spatial mapping module is used to recall a set of candidate spatial units from the spatial unit hierarchy library based on the location and address evidence in the evidence vector, calculate the spatial mapping confidence score between the patrol record and each candidate spatial unit, and determine the candidate spatial unit with the highest spatial mapping confidence score as the target spatial unit. Specifically, the spatial mapping module recalls a set of candidate spatial units from the spatial unit hierarchy library based on the location and address evidence in the evidence vector; for each candidate spatial unit, it converts the geometric relationship, address matching relationship, image recognition relationship, facility correspondence relationship, historical association relationship, time matching relationship, and evidence conflict relationship between the patrol record and the candidate spatial unit into mapping features to obtain a set of mapping features; it scores the set of mapping features according to configurable rules or a ranking model to obtain the spatial mapping confidence score of each candidate spatial unit; and it determines the candidate spatial unit with the highest spatial mapping confidence score as the target spatial unit. The operations performed by the spatial mapping module are the same as step 300 in the method embodiment, and will not be repeated here.
[0194] The object aggregation module is used to determine whether the inspection record corresponds to an existing hidden danger object based on the target spatial unit. If they correspond, the inspection record is aggregated into the corresponding hidden danger object; if they do not correspond, a new hidden danger object is created. Specifically, the object aggregation module extracts the spatial unit number, facility number, component number, hidden danger category, location bucket, key image features, and effective time window from the existing hidden danger object to obtain the object fingerprint; calculates the object similarity between the inspection record and the existing hidden danger object; determines whether the exclusion condition is met based on the object similarity; if the exclusion condition is not met and the object similarity meets the aggregation confirmation condition, the inspection record is aggregated into the corresponding hidden danger object; if the exclusion condition is met, an object to be reviewed is generated; if the aggregation confirmation condition is not met and there is no hidden danger object that meets the confirmation condition, a new hidden danger object is created. The operations performed by the object aggregation module are the same as step 400 in the method embodiment, and will not be repeated here.
[0195] Preferably, the system further includes a state transition module, used to trigger state changes of the potential hazard object based on new inspection records, rectification records, review results, clearance materials, and time intervals, and obtain state transition results. The state transition module maintains the complete lifecycle state of the potential hazard object, including state types such as newly added, existing, under rectification, cleared, recurring, expanded, transferred, and pending review. The operations performed by the state transition module are the same as step 500 in the method embodiment, and will not be repeated here.
[0196] Preferably, the system further includes a review and write-back module, used to generate review tasks for inspection records where the spatial mapping confidence level is lower than a preset threshold or where there are conflicting evidence, and to write back the manual confirmation results to the spatial unit alias, facility correspondence, object fingerprint, and scoring parameters. The beneficial effect of the review and write-back module is that the present invention can gradually correct the spatial alias, facility correspondence, and aggregation threshold in actual use, thereby continuously improving the accuracy of subsequent automatic aggregation.
[0197] Preferably, the system further includes a data access module for accessing inspection records, images and videos, coordinate addresses, facility ledgers, historical hazards, rectification feedback, and manual review results, and uniformly numbering and timestamping data from different sources to provide a unified format of data input for the multi-source evidence analysis module, spatial mapping module, object collection module, and state transition module.
[0198] Preferably, the system further includes an output service module, which is used to output object-oriented aggregation results, spatial hot zones, recurrence lists, pending review lists, and responsibility scope prompts to the front-end map, inspection board, hidden danger ledger, and statistical analysis interface.
[0199] Example 3:
[0200] Taking the inspection of public areas in a certain park as an example, the specific implementation process of the present invention is explained.
[0201] Patrolman A reported "damaged fire hydrant door" at "East side passage on the first floor of Building 3", uploaded photos and selected map locations. Due to indoor positioning drift, the coordinates fell near the boundary of Building 2, but the photos identified "Building 3", "fire hydrant" and "east side passage" signs, and the text description also included "east side of the first floor of Building 3".
[0202] In step 310, the present invention performs a candidate spatial unit recall. The geographical coordinate range is determined based on location evidence, and address keywords such as "Building 3," "first floor," "east side," and "fire hydrant" are obtained based on address evidence. In the spatial unit hierarchy library, spatial units covering the coordinate area are recalled to obtain a preliminary set, and then filtered based on address keywords to obtain candidate spatial units such as Building 2, Building 3, the public passageway on the first floor of Building 3, and the fire-fighting facility point on the east side of Building 3.
[0203] Subsequently, in step 320, the present invention extracts mapping features for each candidate spatial unit. Taking "Building 3 - First Floor - East Passage - Fire Hydrant 01" as an example: the geometric matching degree is low because the coordinates are biased towards Building 2; however, the address matching degree is high because the text description includes "East side of the first floor of Building 3"; the image matching degree is high because "Building 3" and "fire hydrant" are identified in the photo; the facility correspondence degree is high because the facility code in the inspection record points to the fire hydrant; the historical correlation degree is high because similar hazards have occurred at this location before; and the evidence conflict degree is high because the coordinates and address, and the image pointing are inconsistent.
[0204] In step 330, the present invention performs a weighted calculation on the above-mentioned mapping features using configurable scoring rules. Since the reliability of address and image evidence is higher than that of coordinate evidence during indoor inspections of the park, the scoring rules assign higher weights to address matching and image matching. After comprehensive scoring, in step 340, the present invention identifies "Building 3 - First Floor - East Passage - Fire Hydrant 01" as the target spatial unit and records the reason for the coordinate conflict.
[0205] In step 400, the present invention queries existing potential hazards and finds that the spatial unit, facility component, and hazard category of a closed hazard object H202605-003 are highly consistent with the current record, and the closure time is more than the set recurrence window from the current inspection time. Based on this, the present invention identifies the current record as a recurring hazard and generates a recurring version of H202605-003, rather than creating a completely independent new hazard.
[0206] If the judgment is confirmed by manual review, the present invention will write back the confirmation result, making it easier for similar photos, address aliases and facility locations to be automatically mapped in the future.
[0207] Through the above process, this invention achieves the following technical effects: First, by using multi-evidence mapping scoring, coordinates, addresses, images, facility ledgers, and historical records are all used for location judgment, reducing misattribution caused by single GPS drift. In this specific implementation, although the coordinates were biased towards Building 2, the hazard was correctly located to Building 3 through comprehensive judgment of multiple evidences such as address, image, and facility ledger. Second, by using object fingerprint and object similarity matching, the current inspection record is automatically aggregated into existing recurring hazard objects, rather than creating independent new hazards, reducing duplicate statistics and allowing managers to intuitively understand the recurrence history of the fire hydrant hazard. Third, through manual review and write-back of results, this invention continuously optimizes spatial aliases and scoring parameters, thereby continuously improving the accuracy of subsequent automatic aggregation.
[0208] Example 4
[0209] Figure 3 This is a schematic diagram of an electronic device according to an embodiment of the present invention.
[0210] like Figure 3 The illustrated electronic device 3 includes a general hardware architecture, comprising at least a processor 31 and a memory 32. The processor 31 and memory 32 are connected via a bus 33. The memory 32 is adapted to store instructions or programs executable by the processor 31. The processor 31 can be a standalone microprocessor or a collection of one or more microprocessors. Thus, the processor 31 executes the instructions stored in the memory 32 to perform the method flow of the embodiments of the present invention as described above, thereby realizing data processing and control of other devices. The bus 33 connects the aforementioned components together, and also connects these components to a display controller 34, a display device, and an input / output (I / O) device 35. The input / output (I / O) device 35 can be a mouse, keyboard, modem, network interface, touch input device, motion-sensing input device, printer, and other devices known in the art. Typically, the input / output device 35 is connected to the system via an input / output (I / O) controller 36.
[0211] Those skilled in the art will understand that embodiments of this application can provide methods, apparatus (devices), or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-readable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0212] This application is described with reference to flowchart illustrations of methods, apparatus (devices), and computer program products according to embodiments of this application. It should be understood that each step in the flowchart can be implemented by computer program instructions.
[0213] These computer program instructions may be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including an instruction means, the implementation process of which is described in the instruction means. Figure 1 The function specified in one or more processes.
[0214] These computer program instructions may also be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing device, produce instructions for implementing processes. Figure 1 A device for a function specified in one or more processes.
[0215] Another embodiment of the present invention relates to a non-volatile readable storage medium for storing a computer-readable program for use by a computer to perform embodiments of some or all of the above methods.
[0216] That is, those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program specifying the relevant hardware. This program is stored in a readable storage medium and includes several instructions to cause a device (which may be a microcontroller, chip, etc.) or processor to execute all or part of the steps of the methods described in the embodiments of this application. The aforementioned readable storage medium includes various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.
[0217] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. For those skilled in the art, the present invention can be modified and varied in various ways. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of the present invention should be included within the scope of protection of the present invention.
Claims
1. A method for patrolling a hazard object, characterized by, Includes the following steps: Based on spatial boundary and hierarchical relationship data, multi-level spatial units are divided to obtain a spatial unit hierarchy library; Receive patrol records, perform multi-source analysis on the patrol records, and obtain evidence vectors including location evidence, address evidence, image evidence, facility evidence, historical evidence, and time evidence; Based on the location evidence and address evidence in the evidence vector, a set of candidate spatial units is recalled from the spatial unit hierarchy library. The spatial mapping confidence of the patrol record and each candidate spatial unit is calculated respectively. The candidate spatial unit with the highest spatial mapping confidence score is determined as the target spatial unit. Based on the target space unit, determine whether the inspection record corresponds to an existing hidden danger object; if it corresponds, the inspection record is assigned to the corresponding hidden danger object; if it does not correspond, a new hidden danger object is created.
2. The method of claim 1, wherein, The process of dividing spatial units into multiple levels based on spatial boundary and hierarchical relationship data to obtain a spatial unit hierarchy library includes: Based on spatial boundary and floor structure data, spatial units are divided according to the granularity of region, building, floor, room and facility point, and a parent-child hierarchical relationship is established between spatial units; For each spatial unit, extract hierarchical paths, normalized address fragments, geometric location buckets, facility types, component orientations, and historical high-frequency keywords to obtain the spatial fingerprint of each spatial unit; Integrate all spatial units and their corresponding spatial fingerprints to obtain a spatial unit hierarchy library.
3. The method of claim 1, wherein, The process of performing multi-source analysis on the patrol records to obtain an evidence vector that includes location evidence, address evidence, image evidence, facility evidence, historical evidence, and time evidence includes: Address fragments and facility names are extracted from the textual descriptions in the inspection records to obtain address evidence; Based on the location data in the patrol records, location evidence is obtained; Scene category and sign information are extracted from the images in the patrol records to obtain image evidence; Component information is obtained from the facility codes in the inspection records to obtain facility evidence; Historical evidence is obtained based on the historical data in the aforementioned patrol records; The address evidence, location evidence, image evidence, facility evidence, and historical evidence are combined to obtain an evidence vector.
4. The method of claim 1, wherein, The step of recalling a set of candidate spatial units from the spatial unit hierarchy library based on the location and address evidence in the evidence vector, calculating the spatial mapping confidence score between the patrol record and each candidate spatial unit, and determining the candidate spatial unit with the highest spatial mapping confidence score as the target spatial unit includes: Based on the location evidence and address evidence in the evidence vector, a set of candidate spatial units is recalled from the spatial unit hierarchical library; For each candidate spatial unit, the geometric relationship, address matching relationship, image recognition relationship, facility correspondence relationship, historical association relationship, time matching relationship and evidence conflict relationship between the inspection record and the candidate spatial unit are converted into mapping features to obtain a set of mapping features; The mapping feature set is scored according to configurable rules or ranking models to obtain the spatial mapping confidence of each candidate spatial unit. The candidate spatial unit with the highest confidence score in spatial mapping is determined as the target spatial unit.
5. The method of claim 4, wherein, The step of recalling a set of candidate spatial units from the spatial unit hierarchy library based on the location evidence and address evidence in the evidence vector includes: Based on the location evidence, determine the geographical coordinate range of the patrol record; Based on the address evidence, address keywords are obtained from the patrol records; In the spatial unit hierarchical library, spatial units whose geographical range covers the geographical coordinate range are recalled to obtain a preliminary set of spatial units; Based on the address keywords, the preliminary spatial unit set is matched and filtered to obtain a candidate spatial unit set.
6. The method of claim 4, wherein, For each candidate spatial unit, the geometric relationship, address matching relationship, image recognition relationship, facility correspondence relationship, historical association relationship, time matching relationship, and evidence conflict relationship between the inspection record and the candidate spatial unit are converted into mapping features, and the resulting mapping feature set includes: Based on the location evidence, calculate the geometric matching degree between the location of the patrol record and the candidate spatial unit; Based on the address evidence, calculate the address matching degree between the address text in the inspection record and the standard address and alias of the candidate spatial unit; Based on the image evidence, calculate the image matching degree between the image recognition results in the inspection record and the known facility features of the candidate spatial unit; Based on the facility evidence, calculate the correspondence between the facility codes in the inspection records and the facility records of the candidate space units; Based on the historical evidence, calculate the historical correlation between the inspection records and the historical hazard records of the candidate spatial units; Based on the time evidence, calculate the time matching degree between the patrol time of the patrol record and the time of the candidate spatial unit; The degree of evidence conflict is obtained by detecting whether the spatial units pointed to by the location evidence, address evidence, image evidence, facility evidence, and historical evidence are consistent in different dimensions. By integrating the geometric matching degree, address matching degree, image matching degree, facility correspondence degree, historical correlation degree, time matching degree, and evidence conflict degree, a mapping feature set is obtained.
7. The method of claim 4, wherein, The step of scoring the mapping feature set according to configurable rules or a ranking model to obtain the spatial mapping confidence of each candidate spatial unit includes: Based on the spatial type, patrol task type, or equipment acquisition accuracy, weights are assigned to each mapping feature in the mapping feature set to obtain a weighted scoring rule. According to the weighted scoring rule, the mapping feature set is weighted and calculated to obtain the spatial mapping confidence score of each candidate spatial unit; or... The mapping feature set is input into a pre-trained ranking model to obtain the spatial mapping confidence of each candidate spatial unit; the ranking model is a machine learning model trained with the mapping features corresponding to historical inspection records as training samples and the spatial attribution results confirmed by manual review as training labels.
8. The method of claim 4, wherein, The step of determining the target spatial unit from the candidate spatial unit set based on the highest spatial mapping confidence includes: When the highest spatial mapping confidence score is greater than the first threshold and the difference between the highest and second highest spatial mapping confidence scores is greater than the distinction threshold, the candidate spatial unit corresponding to the highest spatial mapping confidence score is automatically identified as the target spatial unit. When the highest spatial mapping confidence is between the second threshold and the first threshold, or when the spatial mapping confidence of multiple candidate spatial units is close, the inspection record is marked as pending review. When the spatial mapping confidence of all candidate spatial units is lower than the second threshold, a spatial unit missing indication is generated.
9. A computer-readable storage medium for storing computer program instructions, characterized in that, The computer program instructions, when executed by a processor, implement the method as described in any one of claims 1-8.
10. An electronic device comprising a memory and a processor, characterized in that, The memory is used to store one or more computer program instructions, wherein the one or more computer program instructions are executed by the processor to implement the method as described in any one of claims 1-8.
