Digitized methods, systems, devices, and media for on-site safety inspections and consultations

Abstract

A digital method, apparatus, equipment, and medium for on-site safety inspection and consultation are disclosed, relating to the technical field of on-site supervision. The method includes: receiving an inspection request carrying information to be inspected; generating an inspection instruction based on the inspection request; obtaining the current status of the inspector based on the inspection instruction; identifying inspectors currently idle as target inspectors; and sending the information to be inspected to the inspection terminal corresponding to the target inspector; receiving target images collected by the target inspector based on the information to be inspected; inputting the target images into a pre-trained image analysis model to obtain a set of target inspection reports; inputting the set of target inspection reports into a pre-trained data statistical model to obtain a visual analysis report generated based on preset statistical rules; and sending the set of target inspection reports and the visual analysis report to the user terminal that issued the inspection request. Implementing the technical solution provided in this application can digitize on-site inspection and consultation information, improving information transmission efficiency.

Description

Technical Field

[0001] This application relates to the technical field of on-site supervision, specifically to a digital method, system, electronic device, and storage medium for on-site safety inspection and consultation. Background Technology

[0002] With the rapid development of information technology, digital management has gradually permeated various industries, bringing about tremendous changes in improving management efficiency and quality. In the field of on-site safety management, digital transformation has become an inevitable trend. By introducing advanced information technology, real-time monitoring, rapid response, and precise decision-making regarding on-site safety conditions can be achieved, effectively reducing the probability of safety accidents and protecting the lives and property of personnel. This not only helps improve the safety production level of enterprises but also meets society's demand for a safe and efficient production environment.

[0003] Traditional on-site safety inspections and consultations primarily rely on manual paper-based records and communication. When a safety inspection is required, staff typically communicate the inspection task by phone or in person. Inspectors manually record the inspection findings, creating a paper report, which is then passed down and summarized at each level. Issues discovered during the inspection are also reported upwards verbally or through written documents. During the consultation phase, users often need to visit the consultation point in person to communicate face-to-face with professionals and obtain relevant safety information and advice.

[0004] Manual communication and coordination are inefficient, easily leading to untimely and inaccurate information transmission, affecting the allocation and execution of inspection tasks. Paper records are not only cumbersome and error-prone, but also unfavorable for data storage, retrieval, and analysis. When faced with large amounts of inspection data and complex security issues, it is difficult to quickly generate effective analysis reports, failing to provide strong support for decision-making. Furthermore, traditional consultation methods are limited by time and space, failing to meet users' needs for accessing security information anytime, anywhere. Summary of the Invention

[0005] This application provides a digital method for on-site safety inspections and consultations, which can digitize on-site inspection and consultation information to improve information transmission efficiency.

[0006] In a first aspect, this application provides a digital method for on-site safety inspection and consultation, including a digital management platform, a user terminal, and an inspection terminal, applied to the digital management platform, the method comprising: Receive a check request sent by a user terminal carrying information to be checked, and generate a check instruction based on the check request. The information to be checked includes, but is not limited to, the object to be checked, the scope of the check, and the check time. Based on the inspection instruction, the current status and current location of at least one inspector are obtained. The inspector whose current status is idle and whose current location is closest to the inspection object is identified as the target inspector. The inspection information to be inspected is sent to the inspection terminal bound to the target inspector so that the inspection terminal can remind the target inspector when it receives the inspection information to be inspected. Receive the target image corresponding to the information to be inspected, which is collected and uploaded by the inspection personnel using the inspection terminal, and obtain on-site environmental data by connecting to the smart construction site sensor; The target region of the target image is determined using target detection technology, and a reference object within the target region is determined. The reference object is a base object with known size or properties. Based on the reference objects and preset inspection standards, abnormal objects within the target area are identified; The risk diffusion coefficient of the abnormal object is calculated based on the on-site environmental data, and the risk level and rectification plan are determined based on the risk diffusion coefficient. Based on the aforementioned abnormal objects, risk diffusion coefficients, risk levels, and rectification plans, a set of target inspection reports corresponding to each abnormal object is generated according to a preset template. The target inspection report set is input into a pre-trained data statistical model to obtain a visualization analysis report corresponding to the target inspection report set generated based on preset statistical rules. The target inspection report set and the visualization analysis report are sent to the user terminal that issued the inspection request, so that the user terminal can display the visualization analysis report when it receives it.

[0007] By adopting the above technical solution, when the digital management platform receives an inspection request carrying information such as the inspection object, inspection scope, and inspection time, it generates an inspection task. Then, it obtains the current status and location of the inspectors, identifying those currently idle and closest to the inspection object as the target inspectors. The inspection task and pending information are then sent to the user terminal, which notifies the target inspectors upon receipt. This efficiently allocates inspection tasks and improves the response speed of inspectors. Subsequently, the platform receives target images captured by the target inspectors at the inspection object's location using the user terminal, and also receives on-site environmental data detected by smart construction site sensors connected to the user terminal. Target detection technology is used to identify target areas within the target images, reducing subsequent data processing. Reference objects within the target areas are identified, and then analyzed based on these reference objects and preset inspection standards to identify abnormal objects. A target inspection report is generated according to a preset template based on these abnormal objects. This achieves efficient conversion from target images to detailed inspection reports, improving the accuracy and automation of on-site safety inspection report generation and providing a reliable basis for subsequent data analysis and decision-making. After identifying abnormal objects in images using target detection technology, the system automatically determines the degree of violation based on reference object dimensions and preset standards. It also dynamically adjusts the risk level according to on-site environmental data such as temperature, humidity, dust concentration, and noise levels, ensuring the remediation plan is targeted and timely. Furthermore, the visual analysis report presents inspection results intuitively through trend charts and risk heat maps, helping managers quickly grasp the on-site safety situation, formulate differentiated regulatory strategies, and effectively reduce the accident rate.

[0008] Optionally, after generating the target inspection report set based on the abnormal object according to the preset template, the process further includes: Retrieve the entry identifier of each case library in the preset case library set; Match each target inspection report in the target inspection report with each of the inbound identifiers to determine the target inbound identifier corresponding to each target inspection report; Receive the rectification results corresponding to each target inspection report, including but not limited to rectification completion time, rectification success rate, and hazard recurrence rate; Each target inspection report, the corresponding on-site environmental data, and the rectification results are stored in a case library with a corresponding target entry identifier, so that the corresponding target inspection report can be obtained using the target entry identifier.

[0009] By adopting the above technical solutions, closed-loop management and knowledge accumulation of inspection data can be achieved. By linking and storing inspection reports, on-site environmental data, and rectification results with a case library, a standardized knowledge base for handling safety hazard is constructed. Subsequent similar inspection tasks can directly access rectification plans and handling experiences from historical cases, significantly improving problem-solving efficiency. Simultaneously, the image analysis model, through reference object comparison and anomaly recognition algorithms, ensures the objectivity and accuracy of inspection results, avoiding subjective biases from human judgment. The dynamic update mechanism of the case library can also continuously optimize the risk assessment model, shifting safety inspections from passive response to proactive prevention. This provides data support for the long-term safety management of smart construction sites, enabling rapid retrieval of similar cases during actual safety inspections, providing historical references and experience for safety inspections and consultations, improving the efficiency and accuracy of safety inspections, and reducing potential safety risks.

[0010] Optionally, the method further includes: Receive a case retrieval request carrying request information, including but not limited to inspection type, inspection address, and current site environment data; The first similarity between the request information and the entry identifier of each case library is calculated using the cosine similarity algorithm. At least one case library with the highest first similarity and greater than or equal to a preset similarity threshold is determined as the target case library. Calculate the second similarity between the request information and each target inspection report in the target case library, and calculate the scenario fit between the request information and each target inspection report in the target case library; The second similarity and scene adaptability corresponding to each target inspection report are weighted and summed according to a preset weight ratio to obtain the comprehensive matching degree corresponding to each target inspection report. At least one target inspection report with the highest overall matching degree and greater than or equal to the preset matching threshold is identified as a target case, and the target case is sent to the user terminal corresponding to the case acquisition request.

[0011] By adopting the above technical solutions, accurate retrieval and efficient reuse of case knowledge are achieved. Through cosine similarity algorithm and scenario adaptability weighted calculation, the system can quickly locate historical cases that highly match the current inspection scenario, ensuring the relevance and practicality of the target cases. This helps inspectors learn from the experience of handling similar problems and shorten the decision-making cycle. Simultaneously, the dynamic update mechanism of the case database associates and stores new inspection reports, environmental data, and rectification results with the database entry identifier, forming a closed-loop knowledge management system. This not only provides data support for subsequent inspection tasks but also improves the accuracy of case recommendations by continuously accumulating case data to optimize the matching algorithm model. This promotes the transformation of safety inspections from experience-driven to data-driven, further enhancing the intelligence and standardization of on-site supervision.

[0012] Optionally, before sending the target case to the user terminal corresponding to the case acquisition request, the method further includes: The user role that issued the case retrieval request is determined based on the request information; The data access permission level is determined based on the relevance of the user role to the target case library; The target case is anonymized using the preset processing rules corresponding to the data acquisition permission level.

[0013] By adopting the above technical solutions, refined access control and data security are achieved during case retrieval. By dynamically matching the relevance of user roles to the case library, the system can implement differentiated information display strategies based on data access permission levels. For example, cost data in rectification plans can be hidden from ordinary inspectors, while complete case details are only available to management personnel. Information anonymization employs a combination of field-level encryption and content filtering to automatically mask sensitive information such as personnel names and specific location coordinates in cases, while retaining core handling logic and environmental parameters, ensuring a balance between data sharing and privacy protection. Furthermore, the collaborative optimization of permission levels and similarity algorithms avoids interference from irrelevant cases with inspectors and promotes the orderly flow of case knowledge within the organization through a tiered authorization mechanism, ultimately building a secure, controllable, accurate, and efficient intelligent regulatory knowledge ecosystem.

[0014] Optionally, the method further includes: Receive target questions and current environment data sent by user terminals; The target question is input into a pre-trained intelligent question-answering model to obtain an initial set of answers corresponding to the target question; Calculate the third similarity between the on-site environmental data and the current environmental data for each answer in the initial answer set, and determine the feasibility of each answer based on the third similarity. Calculate the fourth similarity between the target question and each answer in the answer set, and determine at least one answer whose fourth similarity is greater than or equal to a preset similarity threshold and is feasible as the target answer set; The target answer set is sent to the user terminal corresponding to the target question.

[0015] By adopting the above technical solutions, a deep integration of intelligent question answering based on multi-dimensional data matching and refined access control is achieved. By performing dual verification of the fourth similarity (semantic similarity) and the third similarity (environmental data suitability) between the target question and the answer set, the system can accurately select solutions that both conform to the essence of the problem and are suitable for the on-site working conditions, avoiding the problem of theoretical answers being out of touch with actual scenarios. Data access permissions are dynamically allocated for different user roles. For example, only operational guidance answers are available to frontline inspectors, while managers are shown complete cases containing in-depth data such as rectification costs and historical recurrence rates, satisfying differentiated needs while reducing the risk of information overload. Information anonymization processing employs field-level access control and dynamic masking technology to automatically hide sensitive fields such as specific coordinates and personnel names in the cases while retaining core processing logic, ensuring secure and compliant data sharing. In addition, the system continuously optimizes the question-and-answer model through user feedback data. When the answer is applied on-site and generates new rectification results, the environmental parameters and similarity weights in the case library are automatically updated, forming a knowledge loop of "problem-answer-feedback-iteration". This promotes the upgrading of consulting services from passive response to proactive prediction, further improving the efficiency of on-site problem solving and the level of intelligent safety supervision.

[0016] Optionally, the method further includes: Receive follow-up questions and current inspection task information sent by the user terminal; Based on the follow-up questions and current environmental data, the target answer is dynamically adjusted to obtain the final answer; Obtain the target image corresponding to the current inspection task, and determine the target tool based on the size data of the reference object in the target image; The final answer and target tools will be sent to the user's terminal.

[0017] By adopting the above technical solutions, dynamic and scenario-based adaptation of inspection and consultation services has been achieved. By receiving follow-up questions from user terminals and current inspection task information, the system can optimize the target answer in real time based on on-site environmental data, ensuring a high degree of match between the solution and actual working conditions. It intelligently recommends suitable tools based on the size data of reference objects in the target image, avoiding inefficiencies or safety risks caused by incompatible tools. Simultaneously, the intelligent question-answering model accurately locates the core of the problem through semantic similarity calculation, and the dynamic adjustment mechanism can deepen or correct the answer based on follow-up questions, forming an interactive closed loop of "question-answer-feedback-optimization." Furthermore, the integrated processing of tool recommendation and answer push reduces the switching operations of inspectors between multiple systems, significantly improves on-site response speed, and promotes the upgrade of consultation services from static knowledge query to dynamic decision support, providing full-process technical support for efficient supervision of smart construction sites.

[0018] Optionally, sending the information to be checked to the inspection terminal bound to the target inspector includes: Obtain historical inspection data within a first preset time period. The historical inspection data includes, but is not limited to, inspection items, inspection order, hazard discovery rate, corresponding hazard rectification success rate, on-site environmental data, and risk diffusion coefficient. The effective weights of each inspection item are calculated using a linear regression algorithm based on historical inspection data, and the inspection strategy corresponding to the information to be inspected is determined based on the effective weights. The information to be checked and the inspection strategy are sent to the inspection terminal bound to the target inspector.

[0019] By adopting the above technical solutions, the dynamic and precise generation of inspection strategies is achieved, improving the scientific nature of task allocation and execution efficiency. By integrating historical inspection data with real-time environmental parameters, the system uses a linear regression algorithm to quantify the weight of each inspection item, ensuring that strategy formulation is both based on historical experience and adapted to current working conditions. The pre-trained analysis model automatically optimizes the inspection sequence and key focus items by learning from massive amounts of cases. For example, it prioritizes the allocation of inspection resources to areas with high-risk diffusion coefficients and increases the frequency of re-inspections for hazard types with low rectification success rates. This dual strategy generation mechanism retains the objectivity of data-driven approaches while avoiding the limitations of human experience through algorithm iteration. Furthermore, the integrated distribution of inspection information and strategies enables target inspectors to quickly grasp the core of the task and key execution points, reducing on-site decision-making time. Simultaneously, the system continuously collects strategy execution feedback data and dynamically adjusts algorithm weights and model parameters, forming a closed-loop management system of "data-strategy-execution-optimization," driving the upgrade of inspection tasks from standardization to intelligence.

[0020] Optionally, the method further includes: Obtain the number of times the same type of hidden danger occurs at the same inspection address within the second preset time period and the risk diffusion coefficient at each occurrence; If the number of occurrences of the same type of hidden danger is greater than or equal to the preset number, and there is at least one risk diffusion coefficient greater than or equal to the preset diffusion threshold, then an upgraded inspection strategy shall be adopted for the hidden danger. The upgraded inspection strategy includes, but is not limited to, increasing the requirements for steering detection equipment, expanding the inspection scope, and adjusting the frequency of re-inspection. The information to be checked and the upgraded inspection strategy will be sent to the inspection terminal bound to the target inspector.

[0021] By adopting the above technical solutions, dynamic upgrading and precise prevention and control of hidden dangers have been achieved. Through trend analysis and risk assessment of historical hidden danger data, the system can proactively identify high-frequency recurring hidden danger types with the risk of spread, triggering intelligent upgrades to inspection strategies. For example, for high-risk hidden dangers that recur frequently, increasing the configuration of dedicated detection equipment can improve data collection accuracy, expanding the inspection scope can cover potentially related areas, and adjusting the frequency of re-inspections can strengthen the tracking of rectification effects, forming a closed-loop management system covering the entire lifecycle from "discovery-rectification-prevention". In addition, the strategy upgrade mechanism dynamically adjusts the intensity of prevention and control in conjunction with the real-time risk diffusion coefficient, avoiding resource waste while ensuring that key hidden dangers are prioritized for handling, effectively reducing the probability of similar accidents recurring, and promoting the transformation of safety supervision from passive response to proactive prevention.

[0022] Optionally, the method further includes: The evolution records of a preset number of similar hazards are obtained from the case library. Evolution path data of the same type is generated based on the evolution records. The evolution records include the on-site environmental data and the change of risk diffusion coefficient at each evolution. The evolution path data includes the hazard progression chain, the corresponding evolution cycle, and the risk diffusion coefficient change curve under different environmental parameters. Acquire real-time inspection data and current environmental data at the inspection site; The real-time inspection data, real-time on-site environmental data, and evolution path data are matched to determine the target evolution path, remaining evolution cycle, and real-time trend of risk diffusion coefficient corresponding to the current hidden danger. Based on the remaining evolution cycle and the changing trend of the risk diffusion coefficient, an early warning time window and an early warning level are generated. Based on the target evolution path and real-time on-site environmental data, preventive inspection recommendations are generated; The warning time window, warning level, preventive inspection recommendations, and risk diffusion coefficient trend chart are sent to user terminals and inspection terminals.

[0023] By adopting the above technical solutions, the system achieves forward-looking prediction and precise prevention and control of potential hazards. By mining historical evolution patterns of similar hazards in the case database and combining real-time environmental data and risk diffusion trends, the system can dynamically generate hazard progression paths and remaining evolution cycles, providing managers with a visualized time-based early warning window. Preventive inspection suggestions generated based on the target evolution path allow for targeted allocation of inspection resources and adjustment of inspection frequency and scope, shifting from "post-event rectification" to "pre-event prevention." The linked display of early warning levels and risk diffusion curves helps supervisors intuitively grasp the development trend of hazards and scientifically prioritize responses; while the dynamic early warning information received simultaneously by inspection terminals ensures that frontline personnel can take intervention measures before risks spread, effectively curbing the escalation of accidents. Furthermore, the continuous accumulation of evolution path data can optimize prediction model parameters, improve the accuracy of trend prediction under different environmental conditions, and form an intelligent prevention and control closed loop of "data-prediction-response-feedback," significantly enhancing the initiative and foresight of on-site safety management.

[0024] A second aspect of this application provides a digital system for on-site safety inspection and consultation, the system comprising a digital management platform, a user terminal, and an inspection terminal; The digital management platform includes a receiving module, a processing module, a data acquisition module, a target determination module, an anomaly determination module, a risk calculation module, a report generation module, a statistics module, and a sending module. The receiving module is used to receive an inspection request sent by a user terminal, which carries inspection information, including but not limited to the inspection object, inspection scope, and inspection time. The processing module is used to generate an inspection instruction based on the inspection request, and obtain the current status and current location of at least one inspector based on the inspection instruction. The inspector whose current status is idle and whose current location is closest to the inspection object is identified as the target inspector. The inspection information to be inspected is sent to the inspection terminal bound to the target inspector so that the inspection terminal will remind the target inspector when it receives the inspection information to be inspected. The acquisition module is used to receive the target image corresponding to the information to be inspected, which is collected and uploaded by the target inspector using the inspection terminal, and to obtain on-site environmental data by connecting to the smart construction site sensor; The target determination module is used to determine the target region of the target image using target detection technology, and to determine the reference object within the target region, wherein the reference object is a reference object with known size or properties; An anomaly determination module is used to determine abnormal objects within the target area based on the reference object and preset inspection standards; The risk calculation module is used to calculate the risk diffusion coefficient of the abnormal object based on the on-site environmental data, and to determine the risk level and rectification plan based on the risk diffusion coefficient. The report generation module is used to combine the abnormal object, risk diffusion coefficient, risk level and rectification plan to generate a set of target inspection reports corresponding to each abnormal object according to a preset template. The statistics module is used to input the target inspection report set into a pre-trained data statistics model to obtain a visualization analysis report corresponding to the target inspection report set generated based on preset statistical rules. The sending module is used to send the target inspection report set and the visualization analysis report to the user terminal that issued the inspection request, so that the user terminal can display the visualization analysis report when it receives it.

[0025] In summary, one or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages: 1. Receive inspection requests and assign tasks based on the status and location of inspectors. Combined with smart construction site sensors to acquire on-site environmental data, the system achieves automated task allocation and real-time acquisition of environmental data, thereby improving the efficiency of on-site safety inspections. 2. By using target detection technology and reference objects to identify abnormal objects, and combining this with on-site environmental data to calculate the risk expansion coefficient and risk level, a rectification plan is generated, which reduces subjective judgment and improves the accuracy of abnormal object identification and risk assessment. 3. The system generates target inspection reports according to preset templates and produces visual analysis reports, which improves the efficiency of inspection report generation and analysis and meets the requirements of modern enterprises for high efficiency and accuracy in safety management. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of the overall process of a digital method for on-site safety inspection and consultation provided in an embodiment of this application; Figure 2 This is a schematic diagram of the structure of a digital system for on-site safety inspection and consultation disclosed in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of an electronic device disclosed in an embodiment of this application.

[0027] Explanation of reference numerals in the attached figures: 500, electronic device; 501, processor; 502, communication bus; 503, user interface; 504, network interface; 505, memory. Detailed Implementation

[0028] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0029] In the description of the embodiments of this application, the words "for example" or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design that is described as "for example" or "for instance" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design options. Rather, the use of the words "for example" or "for instance" is intended to present the relevant concepts in a specific manner.

[0030] In the description of the embodiments of this application, the term "multiple" means two or more. For example, multiple systems means two or more systems, and multiple screen terminals means two or more screen terminals. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the indicated technical features. Thus, a feature defined with "first" or "second" may explicitly or implicitly include one or more of that feature. The terms "comprising," "including," "having," and variations thereof all mean "including but not limited to," unless otherwise specifically emphasized.

[0031] In this application embodiment, the digital management platform, user terminal, and inspection terminal can all be one of the following: smart interactive flat panel, mobile phone, tablet computer, laptop computer, desktop computer, all-in-one computer, vehicle multimedia, server, or workstation.

[0032] Reference Figure 1 This is a flowchart illustrating a digital method for on-site safety inspection and consultation disclosed in an embodiment of this application. Figure 1 As shown, the system includes a digital management platform, a user terminal, and an inspection terminal. The method, applied to the digital management platform, includes at least the following steps: S1 receives an inspection request from a user terminal carrying information to be inspected, and generates an inspection task based on the inspection request. The information to be inspected includes, but is not limited to, the inspection object, the inspection scope, and the inspection time.

[0033] Specifically, during and after the construction process, the construction site needs to be inspected to ensure construction safety and quality. At this time, the on-site construction personnel or on-site supervisors issue an inspection request on their corresponding user terminals and provide the relevant information to be inspected. The information to be inspected can be information pre-stored on the user terminal and directly called up, or it can be entered by the on-site construction personnel or on-site supervisors on the human-machine interface of the user terminal or on-site control terminal when issuing the inspection request. After receiving the inspection request, the digital management platform generates an inspection task based on the inspection request so that the inspectors can carry out the inspection work according to the inspection task and the information to be inspected.

[0034] In this embodiment, the digital management platform first performs structured parsing on the information to be checked in the inspection request sent by the user terminal, extracting the unique identifier of the inspection object, the geographical boundary coordinates of the inspection range, and the timestamp format of the inspection time; secondly, it automatically fills in standard inspection items by matching with a preset inspection task template library; finally, it performs integrity verification on the parsed information, and if there are missing key parameters such as inspection time, it automatically returns supplementary prompts to the user terminal until a standardized inspection instruction containing task priority, execution time limit, and associated resource list is generated.

[0035] The inspection targets include, but are not limited to, special equipment, construction areas, safety protection facilities, hazardous chemical storage points, temporary power facilities, fire lanes and emergency evacuation routes, operating radius of large machinery and equipment, high-altitude work platforms, and confined space work sites.

[0036] The unique identifier for an inspection target can be a unique coded information such as equipment number or area number. Examples include the factory serial number of special equipment, the GIS coordinate code of the construction area, the electronic tag ID of a hazardous chemical storage cabinet, the circuit number of temporary electrical equipment, and the registration code of a high-altitude work platform. These identifiers must meet a structured format that the system can parse, and be directly linked to basic attributes such as equipment model, installation location, and affiliated unit through coding rules. This ensures that inspection tasks are accurately located to specific objects, avoiding task allocation errors caused by identical or similar names.

[0037] Standard inspection items include, but are not limited to, equipment status monitoring items and environmental parameter collection items.

[0038] In this context, "site" refers to the specific physical location where production, construction, operation, or activities actually take place. Site inspection involves conducting safety and quality checks at the actual production, construction, operation, or activity site. Sites requiring inspection include, but are not limited to, those in the construction sector, industrial production sector, transportation sector, or other scenarios. In the construction sector, the site includes all areas within the construction site's perimeter wall, including the construction area, material storage area, temporary power facilities, and worker living quarters. Inspection requests can be made by on-site construction personnel or construction supervisors via user terminals, or by unrelated personnel requesting inspections of construction sites, completed sites, or sites currently in use. Inspection requests can be made using any smart device capable of communicating with the digital management platform, such as a user's mobile terminal or PC.

[0039] Inspection tasks can be generated through a preset algorithm that automatically generates corresponding task content based on the information to be inspected in the inspection request. For example, when the information to be inspected is a safety inspection of electrical equipment in a factory workshop, the preset algorithm will generate a specific inspection task based on this information, including a list of specific equipment to be inspected, the inspection items, and the standards.

[0040] Understandably, the on-site construction period is determined by the construction type, which can be either maintenance or new construction. Therefore, when the project is new construction, the period is longer. Safety or quality inspections are required during and after the construction process. Thus, the inspection request can be initiated by the digital management platform after it receives the on-site inspection request for the first time, based on the construction type determined by the inspection request. If the construction type corresponding to the inspection request is new construction, the digital management platform can generate inspection instructions for the construction site during the construction period according to the preset inspection time interval to ensure safety and quality during construction. Alternatively, the inspection time interval can be preset when the on-site construction personnel or construction supervisors issue an inspection request on the corresponding user terminal.

[0041] The information to be inspected also includes information about the requesting party and the inspected party. The information to be inspected includes, but is not limited to, the name of the organization / individual, contact information (telephone / email), document number (such as business license number, ID card number), and the relationship between the requesting party and the inspected party (such as regulatory department, partner, complainant, etc.). The information about the inspected party includes, but is not limited to, the name of the on-site construction unit, on-site address, contact person and contact information, inspection items, inspection scope, and inspection time.

[0042] S2, based on the inspection task, obtain the current status and current location of at least one inspector, determine the inspector whose current status is idle and whose current location is closest to the inspection object as the target inspector, and send the inspection task and the information to be inspected to the inspection terminal bound to the target inspector so that the inspection terminal will remind the target inspector when it receives the inspection task and the information to be inspected.

[0043] Specifically, after the digital management platform generates an inspection instruction, it retrieves the inspector information from its own database and determines the inspector's current status and location based on the inspector information. The inspector whose current status is idle and whose current location is closest to the inspection target is identified as the target inspector. After the target inspector is identified, the inspection request carrying the information to be inspected is sent to the inspection terminal corresponding to the target inspector, so that the inspection terminal can remind the target inspector to conduct on-site inspection based on the information to be inspected when it receives the information to be inspected.

[0044] In this embodiment, the digital management platform achieves accurate matching and task assignment of target inspection personnel through the following steps: First, it obtains personnel coordinates in real time by accessing the positioning module of the inspection terminal and collects the current task progress through the status reporting interface, i.e., obtains the current status of the inspection personnel; Second, based on the GIS coordinates of the inspection object and the real-time location of each inspection personnel, it calculates the straight-line distance between each inspection personnel and the inspection object, corrects the actual reachable path by combining regional road network data, and generates a distance ranking list; For cases where all statuses are idle and the distances are the same, it introduces performance indicators such as historical task response speed and hidden danger detection rate for secondary ranking to ensure that tasks are assigned to personnel with the best overall efficiency; Finally, it encrypts and sends the inspection task package to the target terminal through the MQTT message push protocol, triggers sound and light reminders and vibration alarms, and records the sending timestamp and reception status. If the reception is not confirmed within 10 minutes, it automatically switches to the second-best personnel, forming a closed-loop control mechanism for task assignment.

[0045] The positioning module of the terminal to be inspected includes, but is not limited to, GPS, Beidou or indoor RFID positioning.

[0046] The inspector's current task progress includes idle, in progress, and completed.

[0047] The inspection task package includes, but is not limited to, information to be inspected, a list of objects to be inspected, and urgency indicators.

[0048] Understandably, the digital management platform pre-stores inspector information and updates the status of each inspector in real time based on the inspection tasks and their attendance information to determine their current status. Inspector status is categorized as present or absent, with present including idle and busy. In practical applications, there may be situations where no inspectors are available. In this case, the current number of existing inspection tasks for each inspector is obtained, and the inspector with the fewest existing tasks is designated as the target inspector. If the inspection task corresponding to the inspection instruction is an urgent task, the inspector closest to the inspection task can be designated as the target inspector based on the inspection location specified in the instruction. When all inspectors have the same number of existing tasks, or multiple inspectors have the same and fewest existing tasks, the inspector who completes their existing tasks first is designated as the target inspector, and so on. In short, any method that minimizes the time required to complete the inspection task specified in the inspection instruction can be adopted.

[0049] It's easy to see that after receiving an inspection request, the digital management platform can obtain the location information of each inspector, determine the corresponding inspection location information based on the inspected party information corresponding to the inspection request, and determine the target inspector by combining the inspector's location, the inspected party's location information, and the inspector's current status. Specifically, after determining that the inspector's current status is idle, the platform determines the inspector closest to the inspected party based on the inspected party's location information and the inspector's current location. Then, the inspector who is currently idle and closest to the inspected party is designated as the target inspector. This helps improve inspection efficiency, responds to inspection requests as quickly as possible, and thus ensures on-site safety and quality.

[0050] In one feasible approach, before sending the information to be inspected to the inspection terminal of the target inspector, the method further includes: acquiring historical inspection data within a first preset time period, including but not limited to inspection items, inspection order, hazard discovery rate, corresponding hazard rectification success rate, on-site environmental data, and risk diffusion coefficient; using a linear regression algorithm to calculate the effective weight of each inspection item based on the historical inspection data, and determining the inspection strategy corresponding to the information to be inspected based on the effective weight; and sending the information to be inspected and the inspection strategy to the inspection terminal bound to the target inspector.

[0051] Specifically, the first preset time period can be dynamically configured according to the risk level of the inspection object, such as setting it to 7 days for objects with extremely high risk and 30 days for objects with general risk.

[0052] The multiple linear regression algorithm was used to explore the correlation between "historical inspection strategies, on-site environmental data, hazard detection rate, risk diffusion coefficient, and rectification success rate" and to calculate the effective weight of each inspection item.

[0053] The specific method for calculating the effective weights is as follows: Using the effective weights of the inspection items as the dependent variable (Y), and the inspection frequency (X1), inspection duration (X2), temperature and humidity deviation values ​​(X3), dust concentration exceeding the standard multiple (X4), hazard detection rate (X5), risk diffusion coefficient (X6), and rectification success rate (X7) from the historical inspection strategy as independent variables, a multiple linear regression model is constructed. ,in, For constant terms, These are the regression coefficients (i.e., weighted contribution values) of each influencing factor. This represents the random error term. During model training, historical data from the past 6 months (sample size ≥ 500 groups) was used. Z-score standardization was used to eliminate dimensional differences, stepwise regression was used to screen for significant variables (P < 0.05), and 5-fold cross-validation was employed to optimize hyperparameters. Finally, the models were sorted according to the absolute value of their regression coefficients; for example, when... (Risk Diffusion Coefficient) > (Hazard detection rate) > When the temperature and humidity deviation is significant, it indicates that the risk diffusion coefficient has the strongest impact on the weight of inspection items, and resource allocation for this type of inspection item should be prioritized when generating inspection strategies. For variables with negative regression coefficients (such as rectification success rate)... It is necessary to combine business logic to determine whether it is a suppressive factor (such as a high rectification success rate may reduce the priority of the inspection item), and adjust the weight calculation results through elasticity coefficients to ensure that the model output meets the actual regulatory needs.

[0054] In this application embodiment, if the inspection item is the temperature inspection of electrical equipment, and the risk diffusion coefficient of the corresponding hidden danger decreases by 0.2 and the rectification success rate increases by ≥18% for every 10% increase in the hidden danger detection rate of the inspection item in a high temperature environment with a temperature greater than 35°C, then the initial weight of the inspection item is increased by 25%.

[0055] In this embodiment of the application, after the inspection strategy is generated, the skill tags such as special equipment operation certificate and high-altitude operation qualification of each inspector can be automatically associated with the current task requirements. If the qualification matching degree of the target inspector is less than 80%, the backup personnel scheduling mechanism is triggered.

[0056] The inspection strategy distribution process uses the MQTT protocol for asynchronous communication. The task package includes an encrypted electronic form template (supporting offline completion), a historical hazard heat map, and emergency contact information. After receiving the package, the inspection terminal provides dual alerts via TTS voice and vibration, and uploads the reception status to the blockchain in real time to ensure the traceability of task transmission. During the execution of the corresponding inspection tasks by the target inspectors, the user terminal's built-in GPS module uploads location coordinates every 5 minutes. The digital management platform compares the inspection area boundaries using the GeoHash algorithm. When it detects that the target inspectors have deviated from the preset route by more than 10 meters, it automatically pushes a route correction prompt, combined with electronic fence technology to prevent the inspection area from being missed.

[0057] In another feasible approach, the inspection strategy can also be upgraded based on the evolution path of the hazard. Specifically, this includes: retrieving a preset number of evolution records for similar hazards from a case library; generating similar evolution path data based on these records; the evolution records include on-site environmental data and changes in the risk diffusion coefficient at each evolution; and the evolution path data includes the hazard progression chain, the corresponding evolution cycle, and risk diffusion coefficient change curves under different environmental parameters. The strategy also involves: acquiring real-time inspection data and current environmental data at the inspection site; matching the real-time inspection data, real-time on-site environmental data, and evolution path data to determine the target evolution path, remaining evolution cycle, and real-time trend of the risk diffusion coefficient for the current hazard; generating an early warning time window and early warning level based on the remaining evolution cycle and risk diffusion coefficient change trend; generating preventative inspection recommendations based on the target evolution path and real-time on-site environmental data; and distributing the early warning time window, early warning level, preventative inspection recommendations, and risk diffusion coefficient change trend chart to user terminals and inspection terminals.

[0058] Specifically, the extraction of evolution records from the case library employs a sliding time window mechanism, defaulting to selecting complete lifecycle data of similar hazards within the past 12 months (with a sample size of no less than 30 records). Each record includes on-site environmental data such as temperature, humidity, and dust concentration collected every 5 minutes, a risk diffusion coefficient change curve, and manually labeled evolution stage tags, such as nascent stage, development stage, and outbreak stage. Evolution path data is constructed using a directed graph model, with nodes representing hazard states, such as "loose scaffolding bolts" in the nascent stage, "excessive lateral displacement" in the development stage, and "structural instability" in the outbreak stage. Edge weights correspond to state transition probabilities (derived from historical data statistics). The real-time matching stage uses a Dynamic Time Warping (DTW) algorithm to non-linearly align the current environmental data sequence with historical evolution paths. When the similarity is ≥85%, the target evolution path is locked, and the remaining evolution cycle is predicted using an LSTM neural network (with an error range controlled within ±15%).

[0059] The warning levels can be divided into four levels: Level I (red warning) is when the remaining period is <2 hours and the diffusion coefficient growth rate is >0.3 / hour; Level II (orange warning) is when the remaining period is 2-4 hours and the diffusion coefficient growth rate is 0.1-0.3 / hour; Level III (yellow warning) is when the remaining period is 4-8 hours and the diffusion coefficient growth rate is <0.1 / hour; and Level IV (blue warning) is when the remaining period is >8 hours.

[0060] The preventive inspection recommendation generation module is linked to the equipment maintenance database, automatically recommending suitable detection tools such as ultrasonic flaw detectors for steel structure crack hazards, and suggesting an inspection frequency of once every 30 minutes for Level I early warnings. It also uses knowledge graph technology to link successful intervention measures from similar cases.

[0061] For example, if the same type of hidden danger occurs ≥2 times, and the hidden danger is not subject to repeated rectification, and the risk diffusion coefficient is ≥1.5 at least once, the inspection strategy upgrade will be automatically triggered. The inspection strategy upgrade includes: adding requirements for specialized testing equipment, using real-time humidity data to correct the risk diffusion coefficient, expanding the inspection scope, and adjusting the frequency of re-inspection.

[0062] The requirements for specialized testing equipment have been increased. For example, for electrical hazards, an infrared thermometer with a resolution of ≥0.1℃ and a measurement range of -20℃ to 500℃ must be used to detect the temperature of the circuit and record the real-time humidity data.

[0063] Expand the scope of inspection. For example, in response to potential hazards in distribution boxes, extend the inspection scope from the distribution box itself to the transmission lines and downstream electrical equipment connected to it. Among them, the length of the transmission line is ≤50 meters, and the downstream electrical equipment includes motors and lighting devices. It is specified that "the operating parameters of the equipment within the extended scope must be collected simultaneously", and the operating parameters include voltage and current.

[0064] Adjust the frequency of follow-up inspections, such as changing the original monthly frequency to once every two weeks, and adding a "root cause investigation of potential hazards" inspection item during the follow-up inspections; the "root cause investigation of potential hazards" inspection item includes the detection of the aging degree of the line and the assessment of the impact of environmental temperature and humidity on the equipment.

[0065] S3 receives the target image corresponding to the information to be inspected, collected and uploaded by the inspection personnel using the inspection terminal, and obtains on-site environmental data by connecting to the smart construction site sensor.

[0066] Specifically, the terminal's built-in high-definition camera module supports autofocus and image stabilization. Target image acquisition must meet preset resolution (e.g., 1920×1080 pixels) and angle requirements (e.g., a 45° overhead shot of the device's front). During acquisition, timestamps, GPS coordinates, and personnel ID watermarks are automatically added. Images are transmitted encrypted via 4G / 5G networks (using the AES-256 algorithm), and large files are processed using a segmented upload mechanism to ensure transmission stability in weak network environments. Smart construction site sensor access uses standardized Modbus protocol or LoRaWAN wireless communication to collect real-time environmental data.

[0067] The on-site environmental data includes, but is not limited to, temperature and humidity, PM2.5 concentration, noise levels, and equipment operating parameters. The sampling frequency for temperature and humidity can be once per minute, for PM2.5 concentration once every 5 minutes, and for noise levels once every 30 seconds. Equipment operating parameters are key technical indicators of the equipment during operation, including but not limited to temperature, pressure, vibration amplitude, current, rotational speed, cumulative operating time, wear of key components, tower crane tilt angle, and scaffolding stress.

[0068] After arriving at the location of the inspected party based on the information to be checked, the inspection personnel use the inspection terminal or the user terminal of the inspection personnel to collect on-site target images during the inspection process.

[0069] During the process of acquiring target images, the target inspector can adjust the image acquisition range, size, and other information as needed. This adjustment can be made on the human-computer interaction interface of the inspection terminal or the user terminal of the target inspector. At the same time, the equipment used for acquiring target images needs to be connected to the digital management platform to ensure that the target images can be transmitted to the digital management platform.

[0070] The target image refers to the on-site image collected by the inspection personnel using the inspection terminal or user terminal during the on-site inspection process, based on their own experience and preliminary judgment that there may be problems. The target image can also be an image of any location on the site collected by the target inspection personnel. If the target inspection personnel cannot determine whether there is an anomaly after preliminary judgment, the image can also be collected and transmitted to the digital management platform. The digital management platform can then analyze the target image to determine whether there is an abnormal object, thereby improving the quality of on-site inspection.

[0071] S4, using target detection technology to determine the target region of the target image, and determining the reference object within the target region, wherein the reference object is a reference object of known size or properties.

[0072] Specifically, target detection technology is used to detect target images and filter out irrelevant areas to obtain the target area. The target area can be obtained by marking or dividing the image after the inspection personnel have collected the on-site image to form an identifier in the on-site image. The target detection technology processes the target image based on the identifier or division mark in the target image to obtain the target area, and determines the reference object of known size in the target area. Then, the size of each object in the target image is calculated based on the reference object.

[0073] Object detection techniques can be algorithms that can separate the background from the target, such as the YOLO algorithm or the Faster R-CNN object detection algorithm.

[0074] The target detection technology extracts key regions (such as equipment surfaces, work areas, etc.) in the target image through a multi-scale feature fusion network, thereby achieving rapid localization of the target region and controlling the localization accuracy error within ±5 pixels. The reference object recognition module compares known reference objects in the image, such as standard-sized safety helmets, tape measure markings, equipment nameplates, etc., through a template matching algorithm, automatically extracts the pixel size of the reference object, and calculates the actual physical size by combining the camera intrinsic parameters. When the confidence level of the reference object recognition is <90%, the manual auxiliary annotation process is triggered.

[0075] S5, determine abnormal objects within the target area based on reference objects and preset inspection standards.

[0076] Specifically, firstly, a scaling factor is calculated using the known physical dimensions of the reference object and the image pixel dimensions. The scaling factor = actual size of the reference object / pixel dimensions of the reference object. This scaling factor is used to convert the pixel dimensions of the object to be detected within the target area into its actual physical dimensions. Secondly, the system retrieves the parameter thresholds for the corresponding objects from the preset inspection standard library. These thresholds include hard indicators such as safety passage width ≥ 1.2m and scaffold pole spacing ≤ 1.5m, as well as attribute requirements such as the color of equipment operation status indicator lights and the integrity of protective devices. The structured comparison engine calculates the deviation between the actual measured value and the standard threshold. The deviation rate = (measured value - standard value) / standard value × 100%. When the absolute value of the deviation rate is > 5% or the attribute status does not conform to the standard, the system automatically marks it as an abnormal object and classifies the risk level according to the degree of deviation. For example, abnormal objects with a deviation rate of 5%-10% are classified as minor abnormalities, those with a deviation rate of 10%-20% are classified as moderate abnormalities, and those with a deviation rate > 20% are classified as severe abnormalities.

[0077] For morphological anomalies such as tilting or deformation of objects, an edge detection algorithm is used to extract the object's contour features, and the Hu moment matching degree is calculated against a standard template. If the matching degree is <85%, it is judged as a morphological anomaly. During the anomaly judgment process, the reference object ID, measurement timestamp, and standard reference number are recorded simultaneously, and a comparison image with annotated boxes is generated for manual review. When the system's judgment confidence level is <90%, it is automatically pushed to the quality inspection expert terminal to trigger a secondary review process.

[0078] S6. Calculate the risk expansion coefficient of the abnormal object based on the on-site environmental data, and determine the risk level and rectification plan based on the risk diffusion coefficient.

[0079] Specifically, the risk diffusion coefficient is calculated using a weighted summation model, with the formula: Risk Diffusion Coefficient = ω1 × Temperature and Humidity Influence Factor + ω2 × Dust Concentration Exceedance Multiple + ω3 × Wind Speed ​​Level + ω4 × Personnel Density Coefficient, where ω1-ω4 are the weight values ​​of each environmental parameter (determined through a logistic regression model trained on historical accident data, e.g., dust concentration weight set to 0.35, temperature and humidity influence factor weight set to 0.25). The temperature and humidity influence factor is calculated based on the deviation rate between the measured value and the safety threshold (e.g., for every 1°C increase in temperature above the 35°C threshold, the factor increases by 0.1). The dust concentration exceedance multiple = measured PM2.5 value / national standard value. The wind speed level is classified according to the Beaufort scale (wind speed coefficient of level 6 and above is 1.2). The personnel density coefficient = real-time number of people in the inspection area / safe carrying capacity. After calculation, the system maps the risk diffusion coefficient to five risk levels: <0.2 is Level I (no risk), 0.2-0.4 is Level II (low risk), 0.4-0.6 is Level III (medium risk), 0.6-0.8 is Level IV (high risk), and >0.8 is Level V (extremely high risk). The rectification plan generation module automatically matches predefined handling strategies according to the risk level. For example, Level V risk immediately triggers a work stoppage order and dispatches an emergency team (response time ≤30 minutes); Level IV risk requires temporary protective measures to be completed within 2 hours; and Level III risk generates a daily review plan. Simultaneously, the system associates case studies to recommend rectification tools for similar hazards (e.g., infrared thermal imager detection for high-risk electrical hazards) and rectification processes (including work permit approval nodes), and visualizes the time nodes and resource allocation of rectification tasks through a Gantt chart. When the risk diffusion coefficient increases by >0.3 within 1 hour, the system automatically upgrades the risk level and adds rectification resource allocation instructions.

[0080] S7 combines the abnormal object, risk diffusion coefficient, risk level and rectification plan to generate a set of target inspection reports corresponding to each abnormal object according to the preset template.

[0081] Specifically, the preset template is a template pre-stored in the digital management platform. It is used to generate information such as the abnormal objects identified by the target inspectors on-site, the information of the abnormal objects, and the risk diffusion coefficient, risk level and rectification plan of the abnormal objects, which is easy for users to check whether there are abnormal objects on site.

[0082] The preset template can include a basic report information area, an abnormal object details area, a risk assessment area, and a rectification plan area.

[0083] The basic information area of ​​the report includes, but is not limited to, automatically filled report number, inspection time, and inspector ID.

[0084] The abnormal object details area includes, but is not limited to, the associated abnormal object ID, location coordinates, and reference object size data.

[0085] The risk assessment area includes, but is not limited to, risk diffusion coefficients, risk level determination criteria, and visualization curves.

[0086] The rectification plan area includes, but is not limited to, a phased task list, a resource requirement list, and acceptance criteria.

[0087] During the data filling phase, the API interface is used to extract abnormal object attributes such as type and size deviation rate from the image analysis module, retrieve on-site environmental data such as temperature, humidity and dust concentration from the environmental database to automatically calculate the risk diffusion coefficient, and combine it with the preset risk level matrix to match the predefined rectification strategy library to generate a handling process that includes the person responsible for rectification, the completion deadline and backup plan.

[0088] The risk level prediction based on the risk diffusion coefficient can be as follows: if the risk diffusion coefficient is ≥1.8, the risk level prediction is "immediate rectification"; if 1.2≤risk diffusion coefficient<1.8, the risk level prediction is "time-limited observation", and the observation period for "time-limited observation" is ≤72 hours; if the risk diffusion coefficient is <1.2, the risk level prediction is "routine monitoring".

[0089] In one feasible approach, after the target inspection report is generated, the electronic signature of the target inspector is added using a digital signature algorithm, and both PDF and XML versions are generated simultaneously. The PDF is used for manual review, while the XML format can be used for system archiving and data analysis. If the report contains anomalies with a risk level of IV or higher, a cross-departmental collaborative process is automatically triggered, pushing the information to the safety management department and construction unit terminals via a message middleware. At the same time, equipment operation permissions in the relevant work areas are locked until rectification is completed.

[0090] The target inspection report set must include at least one target inspection report, because the target image is an image collected by the target price inspectors based on the on-site inspection and after the target inspectors have determined that there may be problems. Therefore, there is at least one on-site construction problem in the target image.

[0091] Among them, the target price inspection report is an analysis report that determines the safety or quality problems existing on site corresponding to the target image based on the analysis of the target image.

[0092] Object detection technologies can employ common algorithms such as YOLO and Faster R-CNN. Preset inspection criteria can be based on industry standards, internal company regulations, etc., such as the normal appearance and markings of specific equipment.

[0093] The preset template can also include basic information about the inspection, such as inspection time, location, and inspectors, as well as descriptions of abnormal objects, judgment criteria, and suggested measures.

[0094] Visual analysis reports can be presented in the form of charts, graphs, etc., making it easy for users to intuitively understand the security status and trends.

[0095] It is understandable that there may be more than one object in the target image that does not conform to the preset inspection criteria. Therefore, the target inspection report set generated based on the target image shall include at least one target inspection report.

[0096] It is conceivable that, in order to improve the efficiency of determining inspection results, for matters that can be directly judged by the target inspectors during the on-site inspection process to determine whether they meet the standards, there is no need to collect target images. Instead, the target inspectors can input the inspection results, inspected party information, and other necessary information for the target inspection report into the inspection terminal or the target inspector's user terminal during the on-site inspection process, and upload the inspection information to the digital management platform. The digital management platform determines the inspection items based on the inspection information uploaded by the target inspectors, and then calls the corresponding inspection standards. By comparing and analyzing the information uploaded by the target inspectors with the preset inspection standards, the severity of the problems existing on-site is determined. Finally, the target inspection report is generated by combining the inspection information uploaded by the target personnel and the corresponding inspection standards.

[0097] In one feasible approach, after generating a set of target inspection reports for each anomaly object according to a preset template by combining the anomaly object, risk diffusion coefficient, risk level, and rectification plan, the target inspection reports can be stored in a digital management platform through the following steps: obtaining the entry identifier of each case library in the preset case library set; matching each target inspection report in the target inspection report set with each entry identifier to determine the target entry identifier corresponding to each target inspection report; receiving the rectification results corresponding to each target inspection report, including but not limited to rectification completion time, rectification success rate, and hazard recurrence rate; and storing the target inspection report, the corresponding on-site environmental data, and the rectification results in the case library corresponding to the target entry identifier, so that the corresponding target inspection report can be obtained using the target entry identifier.

[0098] Specifically, after obtaining the corresponding target inspection report based on the target image, the digital management platform obtains the entry identifier of each case library stored in its database. Then, it matches each target inspection report in the target inspection report with the entry identifier of each case library, determines the target entry identifier that matches each target inspection report from the case library, and receives the rectification results corresponding to each target inspection report, including rectification completion time, rectification success rate, and hazard recurrence rate. Then, it stores each target inspection report, the corresponding rectification results, and the on-site environmental data into the case library of the corresponding target entry identifier. Subsequently, the corresponding target inspection report, the rectification results corresponding to each target inspection report, and the on-site environmental data at the time of inspection can be obtained through the target entry identifier.

[0099] The case library can be categorized based on basic information such as the name or affiliation of the construction party corresponding to the site. For example, different companies could have different case libraries, and the target inspection reports corresponding to different construction parties could be grouped into one case library. Alternatively, it can be categorized based on the different inspection items, such as fire protection measures and construction safety measures, and the target inspection reports corresponding to different inspection items could be grouped into a case library for the same inspection item. It can also be categorized based on the type of site, such as construction sites, schools, residential buildings, and office buildings, and the inspection reports corresponding to sites belonging to the same type could be grouped into one case library.

[0100] The entry identifier can be a unique number or code used to distinguish different case libraries.

[0101] Matching can be achieved by comparing information such as keywords and problem types in the report with the features of the case library represented by the entry identifier. For example, it can be determined by calculating the similarity between the target detection report and each entry identifier. Based on the type of entry identifier, corresponding content is extracted from each target price inspection report, and the similarity between this content and each entry identifier is calculated, thereby determining the target entry identifier that matches each target inspection report.

[0102] The target price inspection report can also be used as training data for various models in the digital management platform to increase the amount of training data for each model, thereby improving the accuracy of the digital management platform in analyzing the target images collected on-site by the target inspection personnel.

[0103] Inspection standards vary depending on the type of site and the specific inspection items. Therefore, the digital management platform contains a database of all inspection standards, including those for the same or different inspection items for all site types. For example, the standard floor height varies depending on the building type. The standard floor height for ordinary commercial housing is 2.8~3.0m, while some high-end residences may reach 3.1~3.3m. The floor height for loft or duplex residences is usually 4.5~5.8m, which can be divided into two floors, but it is important to note that the net height of each floor after division must be ≥2.4m. The floor height for ordinary office buildings is generally 3.5~4.2m, with standard floors (middle floors) often having a height of 3.8~4.0m, and the ground floor (lobby) can be appropriately increased to 4.5~6.0m, etc. Existing paper or electronic standards often use terms like "equivalent" or "similar" to describe the same evaluation standard applied to different site types or inspection items. Directly importing these standards into a digital management platform increases its computational load, reducing inspection efficiency. Therefore, before importing standards into the digital management platform, they should be manually broken down into separate standards for different site types, different items, different items within the same site type, and the same items across different site types. These broken standards are then stored in the digital management platform for analysis of anomalies in target images. During on-site inspections, inspectors may need to initially identify anomalies before collecting images, or they may collect images even when anomalies cannot be determined at the inspection site. The digital management platform then analyzes and processes these images using pre-set inspection standards and reference points, improving processing efficiency.

[0104] Meanwhile, when new inspection standards are generated due to standard updates, the new inspection standards can be directly imported into the digital management platform. The digital management platform uses natural language processing technology to compare the new inspection standards with the inspection standards pre-stored in the original digital management platform to determine the on-site type and inspection items corresponding to the new inspection standards. In this way, the new inspection standards replace the original pre-stored inspection standards, thereby improving the quality and accuracy of inspections.

[0105] In one feasible approach, the entry identifier adopts a composite coding structure, formatted as "hazard type code - inspection area code - risk level code," such as "EL-03-A" representing an electrical hazard, area 3, and risk level A. Unique identifiers and associated metadata, including creation time and covered hazard types, are extracted in batches from the case database using SQL queries. The matching phase employs the bidirectional longest common subsequence (LCS) algorithm, calculating semantic similarity between the target inspection report's core fields (hazard type, inspection object, risk level, etc.) and the entry identifier metadata. When the similarity is ≥80%, the target entry identifier is automatically bound. If multiple matching results exist, such as hazard types across regions, the optimal match is selected by weighted sorting based on the risk diffusion coefficient. Rectification results are received via an API interface, supporting both real-time uploads from inspection terminals and supplementary entries via the web. Data is filtered and validated according to rules such as the rectification completion time needing to be later than the inspection time and the success rate ranging from 0-100%, before being associated with the inspection report. During the storage phase, a database sharding and table partitioning strategy is adopted. The data is routed to the corresponding physical database according to the region code of the entry identifier, and the rectification results and inspection reports are stored in a relational database. At the same time, data fingerprints are automatically generated and stored in the blockchain evidence storage system. When subsequent cases are called, the integrity of the data is verified by fingerprint comparison. If tampering is detected, an alarm is triggered and access to the corresponding case database is locked.

[0106] S8: Input the target inspection report set into the pre-trained data statistical model to obtain a visualization analysis report corresponding to the target inspection report set generated based on preset statistical rules.

[0107] Specifically, after determining the corresponding set of target inspection reports based on the target image analysis, the set of target inspection reports is input into a pre-trained data statistical model. The data statistical model classifies and statistically analyzes each target inspection report in the target detection report according to preset statistical rules to form a corresponding visual analysis report.

[0108] Among them, the preset statistical rules can be the requirements entered in the user terminal that makes the request when the inspection request is issued, such as statistical analysis according to a certain category, such as statistical analysis of inspection results according to inspection items, statistical analysis of inspection results according to inspection location, etc.

[0109] S9 sends the target inspection report set and visualization analysis report to the user terminal that issued the inspection request, so that the user terminal can display the visualization analysis report when it receives it.

[0110] Specifically, after statistically analyzing the target inspection report set using a data statistical model to generate a visualization report, the corresponding target price inspection report set and the corresponding visualization analysis report are sent to the user terminal that issued the inspection request. This allows the user terminal to display each target price inspection report and its corresponding visualization analysis report within the target inspection report set. This enables the user who issued the inspection request to view the target inspection reports on their terminal and take appropriate measures based on the maintenance or rectification suggestions in the target inspection reports to ensure on-site safety and quality.

[0111] In one possible approach, after the digital management platform completes the analysis of the target image and obtains a set of target inspection reports, the user can obtain inspection reports from the existing case library in the digital management platform by establishing a communication connection with the platform. This includes the following steps: receiving a case retrieval request carrying request information, including but not limited to inspection type, inspection address, and current site environment data; calculating a first similarity between the request information and the entry identifier of each case library using a cosine similarity algorithm, and identifying at least one case library with the highest first similarity and greater than or equal to a preset similarity threshold as the target case library; calculating a second similarity between the request information and each target inspection report in the target case library, and calculating the scene fit between the request information and each target inspection report in the target case library; weighting and summing the second similarity and scene fit corresponding to each target inspection report according to a preset weight ratio to obtain a comprehensive matching degree corresponding to each target inspection report; identifying at least one target inspection report with the highest comprehensive matching degree and greater than or equal to a preset matching threshold as the target case, and sending the target case to the user terminal corresponding to the case retrieval request.

[0112] Specifically, when a user needs to obtain cases, they send a case acquisition request to the digital management platform on their user terminal, providing request information such as the case content and type they wish to obtain. Upon receiving the case acquisition request carrying the request information, the digital management platform calculates the first similarity between the request information and the entry identifier of each case library using a preset similarity calculation method. It then identifies at least one case library with the highest first similarity as the target case library. Next, it calculates the second similarity between each target inspection report in the target case library and the request information, identifies at least one target inspection report with the highest second similarity as the target case, and sends the target case to the user terminal corresponding to the case acquisition request.

[0113] The requested information includes the requester's information, the party to be obtained's information, and the relationship between the requester and the party to be obtained. The requester's information includes, but is not limited to, the name of the organization / individual, contact information such as telephone / email, business license number, and ID number. The relationship between the requester and the party to be obtained includes, but is not limited to, on-site construction personnel, on-site management personnel, regulatory departments, partners, and complainants. The party to be obtained's information includes, but is not limited to, the name of the on-site construction unit, the case content, and the case type.

[0114] Another way to determine the target case library is to preset a corresponding case library similarity threshold in the digital management platform. By comparing the first similarity between the request information and the entry identifier of each case library with the case library similarity threshold, the case library corresponding to the first similarity higher than the case library similarity threshold is determined as the target case library.

[0115] The target case can be determined by setting a corresponding case similarity threshold in the digital management platform. By comparing the second similarity between the request information and the target inspection report in each target case with the case similarity threshold, the target inspection report corresponding to the second similarity higher than the case similarity threshold is determined as the target case.

[0116] The first and second similarity scores are calculated using Natural Language Processing (NPL). They can calculate the edit distance between strings, i.e., the minimum number of single-character insertions, edits, and replacements required to transform the request information into the database identifier and the content of the case studies. Alternatively, they can use the cosine similarity formula, representing the request information, the database identifiers of each case database, and the text of each case in the target case database as word frequency vectors. The cosine of the angle between the request information and the corresponding vectors of the database identifiers of each case database and the text of each case in the target case database is then calculated to obtain the corresponding similarity score.

[0117] Understandably, when a user sends a case retrieval request through their corresponding user terminal, to ensure the security of case information in the case library, the data access permission level of the user terminal sending the request is determined based on the request information. This data access permission level is determined by the relevance between the user terminal sending the request and each case library. Pre-defined processing rules corresponding to the data access permission level are then used to anonymize the target case. Specifically, when the digital management platform receives a case retrieval request, it determines the data access permission of the corresponding user terminal based on the request information. It then determines the corresponding data access permission by calculating the relevance between the request information and each case library. Finally, it anonymizes the target case determined by similarity calculation according to the pre-defined processing rules corresponding to the data access permission level, ensuring that the private information of each case in the case library is not leaked.

[0118] The data access permission level can be determined by the relationship between the requester and the party to be accessed in the request information, or by the similarity between the request information and the entry identifier of each case library and / or the target inspection report in the case library. The similarity is directly proportional to the data access permission level, and the higher the similarity, the higher the data access permission level.

[0119] Data access permission levels can also be determined based on information such as the case content and type that the user wants to obtain, as specified in the request information. For example, if the request information indicates that the customer wants a lot of content and the content is not highly related to each other, the corresponding data access permission is low. When the target case is sent to the user's corresponding user terminal, the information is anonymized according to the target case.

[0120] Information anonymization methods include, but are not limited to, replacement, masking, generalization, deletion, encryption, and data synthesis. Replacement involves replacing real data with fictitious data while maintaining consistent formatting; for example, replacing the name Zhang San with Li Si, or the phone number 13812345678 with 12322222222. Masking involves retaining some data while covering the rest with symbols such as * or #; for example, replacing the ID number 320103199001011234 with 320103********1234, or the email address user@example.com with u###@example.com. Generalization reduces data precision to prevent identification of individuals; for example, generalizing the age 28 to 20-30 years old, or generalizing the geographical location Beijing Chaoyang District to North China. Deletion directly removes sensitive fields, preserving the information's place in the target report. Encryption protects data using symmetric encryption (AES) or asymmetric encryption (RSA), requiring a decryption key to view the content. Synthetic data is created using templates such as GAN and VAE to produce fake data with statistical characteristics similar to real data. The Faker library is used to generate fake names, addresses, etc., and SDV is used to generate structured data.

[0121] The request information includes the inspection type (such as standardized classifications like "scaffolding safety" and "temporary power supply"), the GIS coordinate code of the inspection address, the current environmental data sequence (temperature, humidity, wind speed, dust concentration, etc., collected every 5 minutes), and the user role identifier (for access control). During the first similarity calculation, the request information is matched with the feature vector of the database identifier using cosine similarity. The feature vector of the database identifier includes dimensions such as the hazard type code (e.g., EL represents electrical hazards), the regional risk level (Levels A / B / C), and baseline values ​​of environmental parameters. The preset similarity threshold is set to 0.75 (a valid match is considered when the similarity is ≥0.75). If multiple case databases have the same and highest similarity values, a secondary sort is performed based on the number of cases in the case database (prioritizing databases with more than 100 cases) and the most recent update time (prioritizing databases updated within the last 30 days). The second similarity calculation targets inspection reports in the target case library. It extracts hazard description text (vectorized using TF-IDF), risk diffusion coefficient curves, and rectification plan keywords from the reports as features, comparing them with the semantic vector of the request information. The similarity calculation formula is cosθ = (A·B) / (|A||B|) × 100%. Scene adaptability is calculated using Euclidean distance based on environmental parameters. After normalizing the current environmental data with historical environmental data in the cases, multi-dimensional spatial distances are calculated (e.g., temperature and humidity deviation weighted at 0.3, dust concentration deviation weighted at 0.4, and wind speed deviation weighted at 0.3). Smaller distance values ​​indicate higher adaptability (range 0-100%). The comprehensive matching degree uses a weighted summation formula: Comprehensive Matching Degree = Second Similarity × 0.6 + Scene Adaptability × 0.4. A preset matching threshold of 80 points is set. When the comprehensive matching degree is ≥ 80 points, the top 3 reports in descending order of score are selected as the target case candidate set. Before distribution, the system automatically anonymizes the cases (hiding sensitive fields such as personnel names and specific coordinates from the original report), encrypts the transmission using HTTPS, and generates a watermarked PDF file on the user's terminal (the watermark includes the user ID and viewing timestamp). If the request information includes real-time image data, the system additionally enables image feature comparison (calculating the number of feature points that match between the case report's attached image and the current image based on the SIFT algorithm). When the number of matching points is greater than 50, the scene fit weight is increased to 0.5. For cases with no matching results (overall matching scores < 80 points), a manual case recommendation process is automatically triggered, forwarding the request information to the security expert's terminal and prompting for additional search criteria.

[0122] In one feasible approach, when a user encounters a problem at the site or in another location and is unsure of the answer, they can ask a question to the digital management platform to obtain the corresponding answer. This includes the following steps: S801, receiving the target question; S802, inputting the target question into a pre-trained intelligent question-answering model to obtain a set of answers corresponding to the target question; S803, calculating the third similarity between the target question and each answer in the answer set, and determining the answer with a third similarity greater than or equal to a preset similarity threshold as the target answer; S804, sending the target answer to the user terminal corresponding to the target question.

[0123] Specifically, when a user encounters a question, they can establish a communication connection between their user terminal and the digital management platform. The user then inputs the target question on the human-computer interaction interface of their terminal and submits it to the digital management platform. Upon receiving the target question, the platform inputs it into a pre-trained intelligent question-answering model. The model analyzes the target question to obtain a set of corresponding answers, calculates the third similarity between each answer and the target question, compares the third similarity of each answer with a preset similarity threshold, and identifies the answer with a third similarity greater than the threshold as the target answer. This target answer is then sent back to the user terminal that sent the target question, allowing the user to easily view the answer upon receipt.

[0124] The training data for the intelligent question-answering model includes information such as target inspection reports and inspection standards. The intelligent question-answering model is trained based on the target inspection reports and inspection standards so that it can provide a relatively accurate answer when it receives a target question, and the answer is accompanied by evidence.

[0125] The process by which an intelligent question-answering model determines the answer combination based on the target question is as follows: The target question is preprocessed to standardize its text, handling special characters, punctuation, extra spaces, and case sensitivity. Then, the target question is segmented into units that the intelligent question-answering model can understand, such as words, phrases, and single characters. The segmented target question is then converted into vectors, and the positional encoding of these vectors is performed. Finally, an autoregressive algorithm is used to generate the corresponding answers, and the answers are parsed, evaluated, and filtered to obtain the corresponding answer set.

[0126] It should be noted that the methods and system embodiments provided in the above embodiments belong to the same concept, and their specific implementation process can be found in the system embodiments, which will not be repeated here.

[0127] Reference Figure 2 This is a schematic diagram of the structure of a digital system for on-site safety inspection and consultation disclosed in an embodiment of this application. Figure 2As shown, the system includes a digital management platform, user terminals, and inspection terminals. The digital management platform further includes a receiving module, a processing module, a data acquisition module, a target determination module, an anomaly determination module, a risk calculation module, a report generation module, a statistics module, and a sending module.

[0128] The receiving module is used to receive an inspection request sent by a user terminal, which carries inspection information, including but not limited to the inspection object, inspection scope, and inspection time.

[0129] The processing module is used to generate an inspection instruction based on the inspection request, and obtain the current status and current location of at least one inspector based on the inspection instruction. The inspector whose current status is idle and whose current location is closest to the inspection object is identified as the target inspector. The inspection information to be inspected is sent to the inspection terminal bound to the target inspector so that the inspection terminal can remind the target inspector when it receives the inspection information.

[0130] The acquisition module is used to receive the target image corresponding to the information to be inspected, which is collected and uploaded by the inspection personnel using the inspection terminal, and to obtain on-site environmental data by connecting to the smart construction site sensor.

[0131] The target determination module is used to determine the target region of the target image using target detection technology, and to determine the reference object within the target region, wherein the reference object is a reference object with known size or properties.

[0132] The anomaly determination module is used to determine abnormal objects within the target area based on the reference object and preset inspection standards.

[0133] The risk calculation module is used to calculate the risk diffusion coefficient of the abnormal object based on the on-site environmental data, and to determine the risk level and rectification plan based on the risk diffusion coefficient.

[0134] The report generation module is used to combine the abnormal object, risk diffusion coefficient, risk level and rectification plan to generate a set of target inspection reports corresponding to each abnormal object according to a preset template.

[0135] The statistics module is used to input the target inspection report set into a pre-trained data statistics model to obtain a visualization analysis report corresponding to the target inspection report set generated based on preset statistical rules.

[0136] The sending module is used to send the target inspection report set and the visualization analysis report to the user terminal that issued the inspection request, so that the user terminal can display the visualization analysis report when it receives it.

[0137] By dividing the digital management platform into multiple modules, each responsible for a specific function, the system architecture becomes clearer, facilitating development, maintenance, and expansion. The modules collaborate to complete the digital process of on-site safety inspections and consultations, improving the overall performance and reliability of the system and providing enterprises with a more efficient and accurate safety management solution.

[0138] This application also discloses an electronic device 500. (See reference...) Figure 3 , Figure 3 This is a schematic diagram of the structure of an electronic device 500 disclosed in an embodiment of this application. The electronic device 500 may include: at least one processor 501, at least one network interface 504, a user interface 503, a memory 505, and at least one communication bus 502.

[0139] The communication bus 502 is used to enable communication between these components.

[0140] The user interface 503 may include a display screen and a camera. Optionally, the user interface 503 may also include a standard wired interface and a wireless interface.

[0141] The network interface 504 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface).

[0142] The processor 501 may include one or more processing cores. The processor 501 connects to various parts of the server using various interfaces and lines, and performs various server functions and processes data by running or executing instructions, programs, code sets, or instruction sets stored in memory 505, and by calling data stored in memory 505. Optionally, the processor 501 may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). The processor 501 may integrate one or a combination of several of the following: Central Processing Unit (CPU), Graphics Processing Unit (GPU), and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content to be displayed on the screen; and the modem handles wireless communication. It is understood that the modem may also not be integrated into the processor 501 and may be implemented as a separate chip.

[0143] The memory 505 may include random access memory (RAM) or read-only memory. Optionally, the memory 505 may include a non-transitory computer-readable storage medium. The memory 505 may be used to store instructions, programs, code, code sets, or instruction sets. The memory 505 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-described method embodiments, etc.; the data storage area may store data involved in the above-described method embodiments, etc. Optionally, the memory 505 may also be at least one storage device located remotely from the aforementioned processor 501. (Refer to...) Figure 3 The memory 505, which serves as a computer storage medium, may include an operating system, a network communication module, a user interface module, and an application program for a digital method of on-site safety inspection and consultation.

[0144] exist Figure 3 In the illustrated electronic device 500, the user interface 503 is mainly used to provide an input interface for the user and to acquire user input data; while the processor 501 can be used to call an application program stored in the memory 505 for a digital method of on-site safety inspection and consultation. When executed by one or more processors 501, the electronic device 500 performs one or more of the methods described in the above embodiments. It should be noted that, for the foregoing method embodiments, for the sake of simplicity, they are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, because according to this application, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also understand that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0145] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0146] In the various embodiments provided in this application, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some service interface; the indirect coupling or communication connection between apparatuses or units may be electrical or other forms.

[0147] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0148] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0149] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory 505 and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned memory 505 includes various media capable of storing program code, such as a USB flash drive, external hard drive, magnetic disk, or optical disk.

[0150] The above description is merely an exemplary embodiment of this disclosure and should not be construed as limiting the scope of this disclosure. Any equivalent changes and modifications made in accordance with the teachings of this disclosure shall still fall within the scope of this disclosure. Other embodiments of this disclosure will be readily apparent to those skilled in the art upon consideration of the specification and the disclosure of practical truths.

[0151] This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not described in this disclosure. The specification and embodiments are to be considered exemplary only, and the scope and spirit of this disclosure are defined by the claims.

Claims

1. A digital method for on-site safety inspection and consultation, characterized in that, The method includes a digital management platform, user terminals, and inspection terminals, and is applied to the digital management platform. Receive an inspection request sent by a user terminal carrying information to be inspected, and generate an inspection task based on the inspection request. The information to be inspected includes, but is not limited to, the inspection object, the inspection scope, and the inspection time. Based on the inspection task, the current status and current location of at least one inspector are obtained. The inspector whose current status is idle and whose current location is closest to the inspection object is identified as the target inspector. The inspection task and the information to be inspected are sent to the inspection terminal bound to the target inspector so that the inspection terminal can remind the target inspector when it receives the inspection task and the information to be inspected. Receive the target image corresponding to the information to be inspected, which is collected and uploaded by the inspection personnel using the inspection terminal, and obtain on-site environmental data by connecting to the smart construction site sensor; The target region of the target image is determined using target detection technology, and a reference object within the target region is determined. The reference object is a base object with known size or properties. Based on the reference objects and preset inspection standards, abnormal objects within the target area are identified; The risk diffusion coefficient of the abnormal object is calculated based on the on-site environmental data, and the risk level and rectification plan are determined based on the risk diffusion coefficient. Based on the aforementioned abnormal objects, risk diffusion coefficients, risk levels, and rectification plans, a set of target inspection reports corresponding to each abnormal object is generated according to a preset template. The target inspection report set is input into a pre-trained data statistical model to obtain a visualization analysis report corresponding to the target inspection report set generated based on preset statistical rules. The target inspection report set and the visualization analysis report are sent to the user terminal that issued the inspection request, so that the user terminal can display the visualization analysis report when it receives it.

2. The digital method for on-site safety inspection and consultation according to claim 1, characterized in that, After generating a set of target inspection reports for each abnormal object according to a preset template, by combining the abnormal object, risk diffusion coefficient, risk level, and rectification plan, the process further includes: Retrieve the entry identifier of each case library in the preset case library set; Match each target inspection report in the target inspection report with each of the inbound identifiers to determine the target inbound identifier corresponding to each target inspection report; Receive the rectification results corresponding to each target inspection report, including but not limited to rectification completion time, rectification success rate, and hazard recurrence rate; Each target inspection report, the corresponding on-site environmental data, and the rectification results are stored in a case library with a corresponding target entry identifier, so that the corresponding target inspection report can be obtained using the target entry identifier.

3. The digital method for on-site safety inspection and consultation according to claim 2, characterized in that, The method further includes: Receive a case retrieval request carrying request information, including but not limited to inspection type, inspection address, and current site environment data; The first similarity between the request information and the entry identifier of each case library is calculated using the cosine similarity algorithm. At least one case library with the highest first similarity and greater than or equal to a preset similarity threshold is determined as the target case library. Calculate the second similarity between the request information and each target inspection report in the target case library, and calculate the scenario fit between the request information and each target inspection report in the target case library; The second similarity and scene adaptability corresponding to each target inspection report are weighted and summed according to a preset weight ratio to obtain the comprehensive matching degree corresponding to each target inspection report. At least one target inspection report with the highest overall matching degree and greater than or equal to the preset matching threshold is identified as a target case, and the target case is sent to the user terminal corresponding to the case acquisition request.

4. The digital method for on-site safety inspection and consultation according to claim 3, characterized in that, Before sending the target case to the user terminal corresponding to the case acquisition request, the process also includes: The user role that issued the case retrieval request is determined based on the request information; The data access permission level is determined based on the relevance of the user role to the target case library; The target case is anonymized using the preset processing rules corresponding to the data acquisition permission level.

5. The digital method for on-site safety inspection and consultation according to claim 1, characterized in that, The method further includes: Receive target questions and current environment data sent by user terminals; The target question is input into a pre-trained intelligent question-answering model to obtain an initial set of answers corresponding to the target question; Calculate the third similarity between the on-site environmental data and the current environmental data for each answer in the initial answer set, and determine the feasibility of each answer based on the third similarity. Calculate the fourth similarity between the target question and each answer in the answer set, and determine at least one answer whose fourth similarity is greater than or equal to a preset similarity threshold and is feasible as the target answer set; The target answer set is sent to the user terminal corresponding to the target question.

6. The digital method for on-site safety inspection and consultation according to claim 5, characterized in that, The method further includes: Receive follow-up questions and current inspection task information sent by the user terminal; Based on the follow-up questions and current environmental data, the target answer is dynamically adjusted to obtain the final answer; Obtain the target image corresponding to the current inspection task, and determine the target tool based on the size data of the reference object in the target image; The final answer and target tools will be sent to the user's terminal.

7. The digital method for on-site safety inspection and consultation according to claim 1, characterized in that, The step of sending the information to be checked to the inspection terminal bound to the target inspector includes: Obtain historical inspection data within a first preset time period. The historical inspection data includes, but is not limited to, inspection items, inspection order, hazard discovery rate, corresponding hazard rectification success rate, on-site environmental data, and risk diffusion coefficient. The effective weights of each inspection item are calculated using a linear regression algorithm based on historical inspection data, and the inspection strategy corresponding to the information to be inspected is determined based on the effective weights. The information to be checked and the inspection strategy are sent to the inspection terminal bound to the target inspector.

8. The digital method for on-site safety inspection and consultation according to claim 7, characterized in that, The method further includes: Obtain the number of times the same type of hidden danger occurs at the same inspection address within the second preset time period and the risk diffusion coefficient at each occurrence; If the number of occurrences of the same type of hidden danger is greater than or equal to the preset number, and there is at least one risk diffusion coefficient greater than or equal to the preset diffusion threshold, then an upgraded inspection strategy shall be adopted for the hidden danger. The upgraded inspection strategy includes, but is not limited to, increasing the requirements for steering detection equipment, expanding the inspection scope, and adjusting the frequency of re-inspection. The information to be checked and the upgraded inspection strategy will be sent to the inspection terminal bound to the target inspector.

9. The digital method for on-site safety inspection and consultation according to claim 1, characterized in that, The method further includes: The evolution records of a preset number of similar hazards are obtained from the case library. Evolution path data of the same type is generated based on the evolution records. The evolution records include the on-site environmental data and the change of risk diffusion coefficient at each evolution. The evolution path data includes the hazard progression chain, the corresponding evolution cycle, and the risk diffusion coefficient change curve under different environmental parameters. Acquire real-time inspection data and current environmental data at the inspection site; The real-time inspection data, real-time on-site environmental data, and evolution path data are matched to determine the target evolution path, remaining evolution cycle, and real-time trend of risk diffusion coefficient corresponding to the current hidden danger. Based on the remaining evolution cycle and the changing trend of the risk diffusion coefficient, an early warning time window and an early warning level are generated. Based on the target evolution path and real-time on-site environmental data, preventive inspection recommendations are generated; The warning time window, warning level, preventive inspection recommendations, and risk diffusion coefficient trend chart are sent to user terminals and inspection terminals.

10. A digital system for on-site safety inspection and consultation using any one of claims 1-9, characterized in that, The system includes a digital management platform, user terminals, and inspection terminals; The digital management platform includes a receiving module, a processing module, a data acquisition module, a target determination module, an anomaly determination module, a risk calculation module, a report generation module, a statistics module, and a sending module. The receiving module is used to receive an inspection request sent by a user terminal, which carries inspection information, including but not limited to the inspection object, inspection scope, and inspection time. The processing module is used to generate an inspection instruction based on the inspection request, and obtain the current status and current location of at least one inspector based on the inspection instruction. The inspector whose current status is idle and whose current location is closest to the inspection object is identified as the target inspector. The inspection information to be inspected is sent to the inspection terminal bound to the target inspector so that the inspection terminal will remind the target inspector when it receives the inspection information to be inspected. The acquisition module is used to receive the target image corresponding to the information to be inspected, which is collected and uploaded by the target inspector using the inspection terminal, and to obtain on-site environmental data by connecting to the smart construction site sensor; The target determination module is used to determine the target region of the target image using target detection technology, and to determine the reference object within the target region, wherein the reference object is a reference object with known size or properties; An anomaly determination module is used to determine abnormal objects within the target area based on the reference object and preset inspection standards; The risk calculation module is used to calculate the risk diffusion coefficient of the abnormal object based on the on-site environmental data, and to determine the risk level and rectification plan based on the risk diffusion coefficient. The report generation module is used to combine the abnormal object, risk diffusion coefficient, risk level and rectification plan to generate a set of target inspection reports corresponding to each abnormal object according to a preset template. The statistics module is used to input the target inspection report set into a pre-trained data statistics model to obtain a visualization analysis report corresponding to the target inspection report set generated based on preset statistical rules. The sending module is used to send the target inspection report set and the visualization analysis report to the user terminal that issued the inspection request, so that the user terminal can display the visualization analysis report when it receives it.

Images