An intelligent construction site construction safety supervision method and system based on artificial intelligence

Abstract

The application discloses a kind of wisdom construction site construction safety supervision method and system based on artificial intelligence, it is related to wisdom construction site technical field, including edge perception layer, cloud collaborative layer, grid management platform and collaborative governance terminal.Edge perception layer is deployed in construction site, realizes local real-time analysis by multimodal data acquisition unit and edge computing unit, independently maintains regulatory function when network interruption;Cloud collaborative layer carries out deep semantic understanding and root cause analysis by large model inference engine;Grid management platform realizes regional granularity division and responsibility binding;Collaborative governance terminal completes early warning reception and rectification feedback.The application solves the problems of strong network dependence, insufficient algorithm adaptability and serious data island in the prior art, and realizes low-latency, high-reliability, low-cost and easy-to-promote intelligent safety management of construction site.

Description

Technical Field

[0001] This invention relates to the field of smart construction site safety management technology, specifically to a smart construction site construction safety supervision method and system based on artificial intelligence. Background Technology

[0002] With the rapid development of the construction industry, the scale and complexity of engineering projects are constantly expanding, posing severe challenges to on-site safety management. Traditional construction safety supervision mainly relies on manual inspections and fixed-point duty, which suffers from problems such as incomplete supervision coverage, inadequate risk identification, and delayed response to hidden dangers. In recent years, some construction sites have introduced video surveillance systems, which have freed up manpower to some extent, but are still limited by the monitoring angle and scope, making it difficult to achieve comprehensive, all-weather intelligent supervision.

[0003] Existing systems mostly adopt a centralized cloud processing model, requiring the real-time uploading of massive amounts of video data from construction sites to the cloud for intelligent identification and analysis. However, large construction sites have dispersed work areas and frequently change construction phases, making it difficult to guarantee continuous and stable network coverage. This results in the timeliness of risk identification and the reliability of alarms being constrained by network conditions. In remote construction sites or in the event of a temporary network outage, the system may become paralyzed.

[0004] Therefore, this invention proposes a smart construction site safety supervision method and system based on artificial intelligence to solve the aforementioned problems. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a smart construction site safety supervision method and system based on artificial intelligence, which solves the problem of not being able to predict accident-prone areas during monitoring.

[0006] To achieve the above objectives, the present invention provides the following technical solution: An AI-based smart construction site safety monitoring system includes an edge perception layer deployed at the construction site, comprising a multimodal data acquisition unit and an edge computing unit. The multimodal data acquisition unit is used to collect video data, radar point cloud data, acoustic data, environmental sensor data, and personnel positioning data from the construction site in real time. The edge computing unit has a built-in lightweight AI recognition model for real-time analysis and processing of the collected multimodal data locally, identifying safety hazards and generating early warning information, and independently maintaining on-site monitoring functions even in the event of a network outage. The cloud-based collaboration layer includes a data fusion center, a large-scale model inference engine, and a knowledge graph library. The data fusion center is used to receive and integrate structured data uploaded by the edge perception layer, eliminating the heterogeneity of multi-source data. The large-scale model inference engine performs deep semantic understanding and root cause analysis of complex construction scenarios based on a multimodal large language model, and generates rectification suggestions by combining construction safety specifications in the knowledge graph library. The grid-based management platform includes a region division module, a responsibility binding module, and a hierarchical early warning module. The region division module is used to divide the work surface into granular areas based on construction drawings to determine grid areas. The responsibility binding module is used to configure responsible personnel and patrol points for each grid area and establish the relationship between personnel, area, and equipment. The hierarchical early warning module is used to generate differentiated early warning instructions based on the risk level and push them to the smart terminals of the corresponding responsible personnel. The collaborative governance terminals, including smart safety helmets, mobile law enforcement terminals, and on-site exposure screens, are used to receive early warning information, execute rectification tasks, and feed back the processing results to the cloud, forming a closed-loop management system.

[0007] Preferably, the edge computing unit adopts a heterogeneous computing architecture, integrating a GPU acceleration module and an NPU inference module, supporting real-time analysis of multiple video streams; the lightweight AI recognition model undergoes knowledge distillation and model pruning optimization to reduce computing resource consumption while maintaining recognition accuracy.

[0008] Preferably, the multimodal data acquisition unit includes: A high-definition camera array covers the main work areas of the construction site and supports infrared night vision and wide dynamic range imaging; Millimeter-wave radar is used for predicting the trajectory of falling objects from high altitudes and detecting people in dangerous areas; A voiceprint sensor array is used to capture abnormal sound signals and identify the type of risk. The environmental sensor suite includes a dust sensor, a noise sensor, a temperature and humidity sensor, and an anemometer. UWB positioning base stations are used to track the location of people and equipment in real time.

[0009] Preferably, the large model inference engine adopts a retrieval-enhanced generation architecture. When performing risk analysis, it first retrieves relevant safety regulations from the knowledge graph database, and then inputs the retrieval results and on-site perception data into the multimodal large language model to generate rectification suggestions with legal basis.

[0010] Preferably, the grid management platform also includes a patrol task generation module, which is used to dynamically generate differentiated patrol tasks based on the risk level of the grid area, increasing the patrol frequency in high-risk areas and decreasing the patrol frequency in low-risk areas, thereby achieving optimized resource allocation.

[0011] Preferably, the collaborative governance terminal also includes an insurance data interface, which is used to synchronize data such as hazard discovery, rectification process, and accident records to the insurance institution's system in real time, supporting differentiated pricing of insurance rates based on dynamic risk assessment.

[0012] A smart construction site safety supervision method based on artificial intelligence includes the following steps: S1, Grid-based area division: Obtain construction drawings, divide the construction drawings into polygonal granularity, and determine several grid areas; establish a risk source database, determine the risk level of each grid area and the corresponding inspection standards based on the risk source database; assign responsible personnel and patrol points to each grid area, and bind the patrol points to patrol cards. S2, Multimodal Data Acquisition and Edge Processing: Multimodal data acquisition units are deployed at the construction site to collect video, radar, sound waves, environmental and positioning data in real time; the edge computing unit preprocesses and analyzes the collected data in real time, and calls a lightweight AI recognition model to identify safety hazards; if the recognition result reaches the warning threshold, the edge computing unit immediately generates a level one warning message locally, and provides immediate reminders through the on-site exposure screen and smart safety helmet, while uploading the warning message and on-site footage to the cloud collaboration layer; S3, Cloud-based Deep Analysis and Root Cause Tracing: The cloud-based collaboration layer receives structured data uploaded by the edge perception layer and eliminates the heterogeneity of multi-source data through the data fusion center; the large model inference engine performs deep semantic understanding of complex construction scenarios to identify hidden risks that are difficult to judge by conventional algorithms; combined with construction safety specifications in the knowledge graph library, it performs root cause analysis of risks and generates specific rectification suggestions; S4, Collaborative Governance and Closed-Loop Management: The grid management platform generates differentiated early warning instructions based on risk levels and pushes them to the smart terminals of the responsible personnel in the corresponding grid area; after receiving the early warning, the responsible personnel rush to the site to verify and handle the situation, and upload before-and-after comparison photos of the rectification through mobile law enforcement terminals; the system automatically verifies the rectification effect, and closes the early warning work order after confirming that the hidden danger has been eliminated, forming a closed-loop management of "discovery-early warning-handling-verification"; S5, Data Feedback and Model Optimization: Collect historical early warning data, rectification records and accident cases to build a safety knowledge graph for construction sites; use new sample data to incrementally train the lightweight AI recognition model and the large model inference engine to continuously improve recognition accuracy and inference ability.

[0013] Preferably, the lightweight AI recognition model in S2 includes the following detection tasks: Personal protective equipment (PPE) inspection: Identifying violations such as not wearing a helmet, not wearing reflective clothing, and smoking; Dangerous area intrusion detection: Determines whether personnel or equipment have entered a prohibited area based on electronic fences; Falling object detection: Identifying the risk of falling objects from heights by combining visual optical flow analysis and millimeter-wave radar trajectory prediction; Structural deformation detection: Monitoring displacement and deformation of structures such as deep foundation pits and scaffolding based on visual SLAM technology; Early smoke and fire identification: By fusing video smoke features and infrared thermal imaging, the false alarm rate is reduced.

[0014] Preferably, the independent operation mechanism of the edge computing unit in S2 during network interruption includes: local storage of collected data and identification results; automatic synchronization of data with the cloud collaboration layer when the network is restored; and sending emergency notifications to on-site management personnel via audible and visual alarms and 4G / 5G backup channels if a serious security risk occurs during the network outage.

[0015] Preferably, the risk root cause analysis method of the large model inference engine in S3 includes: extracting key information of abnormal events and matching it with construction project application materials, construction plans, and technical briefing records; determining the specific location coordinates and responsible parties of the risk source; assessing the degree of responsibility for the accident based on the degree of difference; and generating a complete risk assessment report that includes risk description, cause analysis, rectification measures, and responsible persons.

[0016] This invention provides a smart construction site safety supervision method and system based on artificial intelligence. Compared with existing technologies, it has the following advantages: (1) Adopting a cloud-edge-device collaborative architecture, the AI ​​recognition capability is brought forward to the edge nodes of the construction site. Risk identification and alarms do not rely on the cloud network, and the on-site monitoring function can be maintained independently even if the cloud connection is disconnected. The edge computing unit has a built-in lightweight model to achieve millisecond-level response, which significantly improves the timeliness of handling emergencies.

[0017] (2) By integrating multi-source data such as video, radar, sound waves, and environmental sensors, the limitations of single visual perception in complex environments are overcome. Millimeter-wave radar can penetrate dust to accurately detect the location of personnel, voiceprint recognition can capture abnormal sound signals, infrared thermal imaging can improve the accuracy of smoke and fire identification, and multimodal collaboration can significantly reduce the false alarm rate and the missed alarm rate. Attached Figure Description

[0018] Figure 1 This is a block diagram of the system in this invention; Figure 2 This is a diagram of the multimodal data acquisition unit in this invention; Figure 3 This is a flowchart of the monitoring process in this invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] This invention provides the following technical solutions: Please see Figures 1-3 An AI-based smart construction site safety monitoring system includes an edge perception layer deployed at the construction site, comprising a multimodal data acquisition unit and an edge computing unit. The multimodal data acquisition unit is used to collect video data, radar point cloud data, acoustic data, environmental sensor data, and personnel positioning data from the construction site in real time. The edge computing unit has a built-in lightweight AI recognition model for real-time analysis and processing of the collected multimodal data locally, identifying safety hazards and generating early warning information, and independently maintaining on-site monitoring functions even in the event of a network outage. The cloud-based collaboration layer includes a data fusion center, a large-scale model inference engine, and a knowledge graph library. The data fusion center is used to receive and integrate structured data uploaded by the edge perception layer, eliminating the heterogeneity of multi-source data. The large-scale model inference engine performs deep semantic understanding and root cause analysis of complex construction scenarios based on a multimodal large language model, and generates rectification suggestions by combining construction safety specifications in the knowledge graph library. The grid-based management platform includes a region division module, a responsibility binding module, and a hierarchical early warning module. The region division module is used to divide the work surface into granular areas based on construction drawings to determine grid areas. The responsibility binding module is used to configure responsible personnel and patrol points for each grid area and establish the relationship between personnel, area, and equipment. The hierarchical early warning module is used to generate differentiated early warning instructions based on the risk level and push them to the smart terminals of the corresponding responsible personnel. The collaborative governance terminals, including smart safety helmets, mobile law enforcement terminals, and on-site exposure screens, are used to receive early warning information, execute rectification tasks, and feed back the processing results to the cloud, forming a closed-loop management system.

[0021] The edge computing unit adopts a heterogeneous computing architecture, integrating a GPU acceleration module and an NPU inference module, supporting real-time analysis of multiple video streams; the lightweight AI recognition model has undergone knowledge distillation and model pruning optimization, reducing computing resource consumption while maintaining recognition accuracy.

[0022] The multimodal data acquisition unit includes: A high-definition camera array covers the main work areas of the construction site and supports infrared night vision and wide dynamic range imaging; Millimeter-wave radar is used for predicting the trajectory of falling objects from high altitudes and detecting people in dangerous areas; A voiceprint sensor array is used to capture abnormal sound signals and identify the type of risk. The environmental sensor suite includes a dust sensor, a noise sensor, a temperature and humidity sensor, and an anemometer. UWB positioning base stations are used for real-time tracking of personnel and equipment locations.

[0023] The large model inference engine adopts a retrieval-enhanced generation architecture. When conducting risk analysis, it first retrieves relevant safety regulations from the knowledge graph database, and then inputs the retrieval results and on-site perception data into a multimodal large language model to generate rectification suggestions with legal basis.

[0024] The grid management platform also includes a patrol task generation module, which dynamically generates differentiated patrol tasks based on the risk level of the grid area. The patrol frequency is increased in high-risk areas and decreased in low-risk areas to achieve optimized resource allocation.

[0025] The collaborative governance terminal also includes an insurance data interface, which is used to synchronize data such as hazard discovery, rectification process, and accident records to the insurance institution's system in real time, supporting differentiated pricing of insurance rates based on dynamic risk assessment.

[0026] A smart construction site safety supervision method based on artificial intelligence includes the following steps: S1, Grid-based area division: Obtain construction drawings, divide the construction drawings into polygonal granularity, and determine several grid areas; establish a risk source database, determine the risk level of each grid area and the corresponding inspection standards based on the risk source database; assign responsible personnel and patrol points to each grid area, and bind the patrol points to patrol cards. Obtain construction drawings and import CAD files into the grid management platform 300. Use the polygon tool to divide the work surfaces of the construction drawings into N grid areas. The division principles include: first-level grids based on construction sections, second-level grids based on floors or functional areas, and third-level grids based on critical and high-risk engineering work surfaces. Establish a risk source database, containing 12 major categories of risk sources such as high-altitude operations, temporary power supply, hoisting and lifting, and foundation pit engineering. Each risk source corresponds to a specific risk description, possible accident types, prevention and control measures, and inspection standards. Based on the risk source database, determine the risk level of each grid area and the corresponding inspection standards.

[0027] Assign responsible personnel and patrol points to each grid area, and bind the patrol points to NFC patrol cards. The patrol cards support NFC near-field communication. When patrol personnel arrive at the point, they use a smart safety helmet or mobile device to scan the card, and the system automatically records the patrol time and personnel information.

[0028] S2, Multimodal Data Acquisition and Edge Processing: Multimodal data acquisition units are deployed at the construction site to collect video, radar, sound waves, environmental and positioning data in real time; the edge computing unit preprocesses and analyzes the collected data in real time, and calls a lightweight AI recognition model to identify safety hazards; if the recognition result reaches the warning threshold, the edge computing unit immediately generates a level one warning message locally, and provides immediate reminders through the on-site exposure screen and smart safety helmet, while uploading the warning message and on-site footage to the cloud collaboration layer; Multimodal data acquisition units are deployed at the construction site according to grid areas to ensure comprehensive coverage of key areas. Edge computing units poll the video streams from each camera and utilize lightweight AI recognition models for real-time analysis. Recognition tasks include: Personal protective equipment (PPE) detection: Based on the YOLOv8s model, it identifies behaviors such as not wearing a helmet, not wearing reflective clothing, and smoking; Dangerous area intrusion detection: Based on the semantic segmentation model DeepLabV3+, the electronic fence is dynamically divided to determine whether personnel or equipment have crossed the boundary; Falling object detection: By integrating video optical flow analysis and millimeter-wave radar point cloud data, the trajectory of falling objects is predicted and the impact point range is calculated, providing early warning; Structural deformation detection: Based on visual SLAM technology, feature points of deep foundation pit support structures are extracted to monitor displacement and deformation; Early smoke and fire identification: By fusing visible light smoke characteristics with infrared thermal imaging temperature anomalies, the false alarm rate is reduced.

[0029] If the identification result reaches the warning threshold, the edge computing unit immediately generates a level-one warning message locally, triggers an audible and visual alarm via the on-site exposure screen, and sends a voice reminder to personnel wearing smart safety helmets in the area. Simultaneously, the warning message and on-site footage are compressed and uploaded to the cloud collaboration layer. If the network is interrupted, the edge computing unit activates local storage, which can continuously record data for 7 days, automatically synchronizing once the network is restored.

[0030] S3, Cloud-based Deep Analysis and Root Cause Tracing: The cloud-based collaboration layer receives structured data uploaded by the edge perception layer and eliminates the heterogeneity of multi-source data through the data fusion center; the large model inference engine performs deep semantic understanding of complex construction scenarios to identify hidden risks that are difficult to judge by conventional algorithms; combined with construction safety specifications in the knowledge graph library, it performs root cause analysis of risks and generates specific rectification suggestions.

[0031] S4, Collaborative Governance and Closed-Loop Management: The grid management platform generates differentiated early warning instructions based on risk levels and pushes them to the smart terminals of the responsible personnel in the corresponding grid area; after receiving the early warning, the responsible personnel rush to the site to verify and handle the situation, and upload before-and-after comparison photos of the rectification through mobile law enforcement terminals; the system automatically verifies the rectification effect, and closes the early warning work order after confirming that the hidden danger has been eliminated, forming a closed-loop management of "discovery-early warning-handling-verification"; The tiered early warning module of the grid management platform determines the risk level based on the risk assessment report and generates differentiated early warning instructions: Level 1 Warning (Red): High-risk operations such as hoisting and lifting must be suspended immediately starting today; Level 2 Warning (Orange): Rectification completed during the shift; Level 3 Warning (Yellow): Rectification must be completed within 24 hours; Level 4 Warning (Blue): A notification message sent to the safety officer.

[0032] Upon receiving an alert, responsible personnel rush to the site for verification and handling. They use mobile enforcement terminals to take photos before rectification, record the handling process, and upload photos after rectification. The system automatically verifies the rectification effect using image comparison algorithms. Once the hazard is confirmed to be eliminated, the alert work order is closed. If rectification is not completed within the specified time or is deemed unqualified, the alert is automatically escalated and pushed to the next higher-level manager. The handling process of all alert events is stored on the blockchain for evidence preservation, ensuring tamper-proof accountability.

[0033] S5, Data Feedback and Model Optimization: Collect historical early warning data, rectification records and accident cases to build a safety knowledge graph for construction sites; use new sample data to incrementally train the lightweight AI recognition model and the large model inference engine to continuously improve recognition accuracy and inference ability.

[0034] The lightweight AI recognition model in S2 includes the following detection tasks: Personal protective equipment (PPE) inspection: Identifying violations such as not wearing a helmet, not wearing reflective clothing, and smoking; Dangerous area intrusion detection: Determines whether personnel or equipment have entered a prohibited area based on electronic fences; Falling object detection: Identifying the risk of falling objects from heights by combining visual optical flow analysis and millimeter-wave radar trajectory prediction; Structural deformation detection: Monitoring displacement and deformation of structures such as deep foundation pits and scaffolding based on visual SLAM technology; Early smoke and fire identification: By fusing video smoke features and infrared thermal imaging, the false alarm rate is reduced.

[0035] The independent operation mechanism of the edge computing unit in S2 during network interruption includes: local storage of collected data and identification results; automatic synchronization of data with the cloud collaboration layer when the network is restored; and sending emergency notifications to on-site management personnel via audible and visual alarms and 4G / 5G backup channels if a serious security risk occurs during the network outage.

[0036] The risk root cause analysis method of the S3 large model inference engine includes: extracting key information of abnormal events and matching it with construction project application materials, construction plans, and technical briefing records; determining the specific location coordinates and responsible parties of the risk source; assessing the degree of responsibility for the accident based on the degree of difference; and generating a complete risk assessment report that includes risk description, cause analysis, rectification measures, and responsible persons.

[0037] Furthermore, any content not described in detail in this specification is existing technology known to those skilled in the art.

[0038] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0039] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A smart construction site safety supervision system based on artificial intelligence, characterized in that: It includes an edge perception layer deployed at the construction site, comprising a multimodal data acquisition unit and an edge computing unit. The multimodal data acquisition unit is used to collect video data, radar point cloud data, acoustic data, environmental sensor data, and personnel positioning data from the construction site in real time. The edge computing unit has a built-in lightweight AI recognition model, which is used to perform real-time analysis and processing of the collected multimodal data locally, identify safety hazards and generate early warning information, and independently maintain on-site monitoring functions even when the network is interrupted. The cloud-based collaboration layer includes a data fusion center, a large-scale model inference engine, and a knowledge graph library. The data fusion center is used to receive and integrate structured data uploaded by the edge perception layer, eliminating the heterogeneity of multi-source data. The large-scale model inference engine performs deep semantic understanding and root cause analysis of complex construction scenarios based on a multimodal large language model, and generates rectification suggestions by combining construction safety specifications in the knowledge graph library. The grid-based management platform includes a region division module, a responsibility binding module, and a hierarchical early warning module. The region division module is used to divide the work surface into granular areas based on construction drawings to determine grid areas. The responsibility binding module is used to configure responsible personnel and patrol points for each grid area and establish the relationship between personnel, area, and equipment. The hierarchical early warning module is used to generate differentiated early warning instructions based on the risk level and push them to the smart terminals of the corresponding responsible personnel. The collaborative governance terminals, including smart safety helmets, mobile law enforcement terminals, and on-site exposure screens, are used to receive early warning information, execute rectification tasks, and feed back the processing results to the cloud, forming a closed-loop management system.

2. The intelligent construction site safety supervision system based on artificial intelligence according to claim 1, characterized in that: The edge computing unit adopts a heterogeneous computing architecture, integrating a GPU acceleration module and an NPU inference module, supporting real-time analysis of multiple video streams; the lightweight AI recognition model has undergone knowledge distillation and model pruning optimization, reducing computing resource consumption while maintaining recognition accuracy.

3. The intelligent construction site safety supervision system based on artificial intelligence according to claim 1, characterized in that: The multimodal data acquisition unit includes: A high-definition camera array covers the main work areas of the construction site and supports infrared night vision and wide dynamic range imaging; Millimeter-wave radar is used for predicting the trajectory of falling objects from high altitudes and detecting people in dangerous areas; A voiceprint sensor array is used to capture abnormal sound signals and identify the type of risk. The environmental sensor suite includes a dust sensor, a noise sensor, a temperature and humidity sensor, and an anemometer. UWB positioning base stations are used to track the location of people and equipment in real time.

4. The intelligent construction site safety supervision system based on artificial intelligence according to claim 1, characterized in that: The large model inference engine adopts a retrieval-enhanced generation architecture. When conducting risk analysis, it first retrieves relevant safety regulations from the knowledge graph database, and then inputs the retrieval results and on-site perception data into a multimodal large language model to generate rectification suggestions with legal basis.

5. The intelligent construction site safety supervision system based on artificial intelligence according to claim 1, characterized in that: The grid management platform also includes a patrol task generation module, which dynamically generates differentiated patrol tasks based on the risk level of the grid area. The patrol frequency is increased in high-risk areas and decreased in low-risk areas to achieve optimized resource allocation.

6. The intelligent construction site safety supervision system based on artificial intelligence according to claim 1, characterized in that: The collaborative governance terminal also includes an insurance data interface, which is used to synchronize data such as hazard discovery, rectification process, and accident records to the insurance institution's system in real time, supporting differentiated pricing of insurance rates based on dynamic risk assessment.

7. A smart construction site safety supervision method based on artificial intelligence, characterized in that, This method applies to an AI-based smart construction site safety monitoring system as proposed in claims 1-6, and includes the following steps: S1, Grid-based area division: Obtain construction drawings, divide the construction drawings into polygonal granularity, and determine several grid areas; establish a risk source database, determine the risk level of each grid area and the corresponding inspection standards based on the risk source database; assign responsible personnel and patrol points to each grid area, and bind the patrol points to patrol cards. S2, Multimodal Data Acquisition and Edge Processing: Multimodal data acquisition units are deployed at the construction site to collect video, radar, sound waves, environmental and positioning data in real time; the edge computing unit preprocesses and analyzes the collected data in real time, and calls a lightweight AI recognition model to identify safety hazards; if the recognition result reaches the warning threshold, the edge computing unit immediately generates a level one warning message locally, and provides immediate reminders through the on-site exposure screen and smart safety helmet, while uploading the warning message and on-site footage to the cloud collaboration layer; S3, Cloud-based Deep Analysis and Root Cause Tracing: The cloud-based collaboration layer receives structured data uploaded by the edge perception layer and eliminates the heterogeneity of multi-source data through the data fusion center; the large model inference engine performs deep semantic understanding of complex construction scenarios to identify hidden risks that are difficult to judge by conventional algorithms; combined with construction safety specifications in the knowledge graph library, it performs root cause analysis of risks and generates specific rectification suggestions; S4, Collaborative Governance and Closed-Loop Management: The grid management platform generates differentiated early warning instructions based on risk levels and pushes them to the smart terminals of the responsible personnel in the corresponding grid area; after receiving the early warning, the responsible personnel rush to the site to verify and handle the situation, and upload before-and-after comparison photos of the rectification through mobile law enforcement terminals; the system automatically verifies the rectification effect, and closes the early warning work order after confirming that the hidden danger has been eliminated, forming a closed-loop management of "discovery-early warning-handling-verification"; S5, Data Feedback and Model Optimization: Collect historical early warning data, rectification records and accident cases to build a safety knowledge graph for construction sites; use new sample data to incrementally train the lightweight AI recognition model and the large model inference engine to continuously improve recognition accuracy and inference ability.

8. The method for supervising construction safety at smart construction sites based on artificial intelligence according to claim 7, characterized in that: The lightweight AI recognition model in S2 includes the following detection tasks: Personal protective equipment (PPE) inspection: Identifying violations such as not wearing a helmet, not wearing reflective clothing, and smoking; Dangerous area intrusion detection: Determines whether personnel or equipment have entered a prohibited area based on electronic fences; Falling object detection: Identifying the risk of falling objects from heights by combining visual optical flow analysis and millimeter-wave radar trajectory prediction; Structural deformation detection: Monitoring displacement and deformation of structures such as deep foundation pits and scaffolding based on visual SLAM technology; Early smoke and fire identification: By fusing video smoke features and infrared thermal imaging, the false alarm rate is reduced.

9. The method for supervising construction safety at smart construction sites based on artificial intelligence according to claim 7, characterized in that: The independent operation mechanism of the edge computing unit in S2 during network interruption includes: local storage of collected data and identification results; automatic synchronization of data with the cloud collaboration layer when the network is restored; and sending emergency notifications to on-site management personnel via audible and visual alarms and 4G / 5G backup channels if a serious security risk occurs during the network outage.

10. The method for supervising construction safety at smart construction sites based on artificial intelligence according to claim 7, characterized in that: The risk root cause analysis method of the S3 large model inference engine includes: extracting key information of abnormal events and matching it with construction project application materials, construction plans, and technical briefing records; determining the specific location coordinates and responsible parties of the risk source; assessing the degree of responsibility for the accident based on the degree of difference; and generating a complete risk assessment report that includes risk description, cause analysis, rectification measures, and responsible persons.

Images