Intelligent construction site hidden danger study and judgment and decision generation method and system

Abstract

The invention provides an intelligent construction site hidden danger research and judgment and decision generation method and system, and relates to the field of application of artificial intelligence in intelligent construction site safety management, and the method comprises the steps: obtaining an original image containing construction site potential safety hazards, carrying out the fine-grained analysis of the original image through a fine-tuned visual large model, and obtaining a fine-grained analysis result; identifying and extracting entities, entity attributes and relationships among the entities, and generating structured metadata; performing multi-hop reasoning on the structured metadata converted into the triple form by utilizing a pre-constructed construction site safety domain knowledge graph, mining and generating a complete risk chain; and on the basis of the generated risk chain, combining with dynamic context information, constructing prompt words, inputting the prompt words into a language large model, and generating a potential safety hazard rectification report containing potential hazard description, risk analysis, standard basis and specific rectification measures. According to the method, deep recognition and analysis of potential safety hazards are realized from perception to cognition to decision making, and the full-chain intellectualization of the multi-step cognition reasoning and decision suggestion generation process is realized.

Description

Technical Field

[0001] This invention relates to the application of artificial intelligence in the safety management of smart construction sites, specifically to a method and system for assessing and generating decisions regarding potential hazards at smart construction sites. Background Technology

[0002] Currently, smart construction site safety management has become an important direction for the digital transformation of the construction industry. Existing technologies largely rely on fixed cameras for real-time video monitoring and use simple computer vision algorithms (such as object detection) to identify pre-set safety hazards (such as not wearing a safety helmet), or use knowledge graphs and natural language processing technologies to extract safety knowledge from text data, structure it, and perform risk analysis, decision support, and intelligent question answering. While these solutions achieve a certain degree of automation, their limitations are becoming increasingly apparent. 1. Superficial perception: Existing systems can usually only identify isolated, explicit targets (such as "safety helmets"), but cannot deeply understand the state of objects in the scene (such as "safety helmets not being worn properly"), spatial relationships (such as "materials piled up in fire lanes"), and complex contextual information. They are powerless to deal with hazards that require comprehensive judgment (such as "welding work next to flammable materials").

[0003] 2. Lack of cognitive reasoning: Most systems lack true "cognitive" ability. They simply match the identification results with a static rule base and cannot perform multi-step reasoning like human experts to uncover potential and cascading risks. For example, after identifying "oil on the ground", they cannot further associate it with "slipping risk" and the deeper safety hazards such as "secondary fall from height".

[0004] 3. Weak decision support: The generated alarms or rectification suggestions are mostly templated and vague statements, lacking specificity and operability; they fail to effectively combine specific safety regulations and historical rectification cases, resulting in poor practicality of the generated measures and difficulty in guiding on-site personnel to carry out effective rectification.

[0005] 4. Insufficient flexibility: It relies heavily on pre-set fixed monitoring points, making it difficult to cover all corners of the construction site, especially temporary work sites and hidden corners; the system currently lacks effective support for flexible and proactive hazard investigation methods such as safety officers taking photos on the go.

[0006] In summary, existing smart construction site safety management technologies have significant shortcomings in terms of the depth and breadth of hazard identification, the level of intelligence in decision-making, and the practicality of the generated measures. Summary of the Invention

[0007] To address the aforementioned issues, this invention proposes a method and system for intelligent construction site hazard assessment and decision generation. From "perception" to "cognition" and then to "decision," it achieves intelligent full-chain processing of in-depth identification and analysis of safety hazards, multi-step cognitive reasoning, and decision suggestion generation.

[0008] According to some embodiments, the present invention adopts the following technical solution: A method for intelligent construction site hazard assessment and decision generation includes: The original image containing construction site safety hazards is acquired, and a fine-tuned visual model is used to perform fine-grained analysis on the original image to identify and extract entities, entity attributes and relationships between entities, and generate structured metadata. By utilizing a pre-built knowledge graph of construction site safety, multi-hop reasoning is performed on structured metadata converted into triples to mine and generate a complete risk chain; Based on the generated risk chain and combined with dynamic context information, prompt words are constructed and input into the language model to generate a safety hazard rectification report that includes hazard description, risk analysis, regulatory basis, and specific rectification measures.

[0009] According to some embodiments, the present invention adopts the following technical solution: A smart construction site hazard assessment and decision generation system includes: The visual perception and structured data generation module is configured to: acquire original images containing construction site safety hazards, perform fine-grained analysis on the original images using a finely tuned visual big model, identify and extract entities, entity attributes and relationships between entities, and generate structured metadata. The knowledge graph reasoning and risk chain mining module is configured to: use a pre-built knowledge graph in the field of construction site safety to perform multi-hop reasoning on structured metadata converted into triples, and mine and generate a complete risk chain; The generative decision report output module is configured to: based on the generated risk chain and combined with dynamic context information, construct prompt words and input them into the language big model to generate a safety hazard rectification report that includes hazard description, risk analysis, regulatory basis and specific rectification measures.

[0010] According to some embodiments, the present invention adopts the following technical solution: A computer program product includes a computer program that, when executed by a processor, implements the aforementioned method for assessing and generating intelligent construction site hazards.

[0011] According to some embodiments, the present invention adopts the following technical solution: A non-transitory computer-readable storage medium is provided for storing computer instructions, which, when executed by a processor, implement the aforementioned method for assessing and generating intelligent construction site hazards.

[0012] According to some embodiments, the present invention adopts the following technical solution: An electronic device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the aforementioned method for assessing and generating intelligent construction site hazards.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention forms a complete closed loop from hazard discovery and in-depth analysis to generating actionable decisions through a collaborative architecture of "visual big model (perception) → domain knowledge graph (cognition) → language big model (decision)". It simulates the judgment and thinking process of human safety experts and breaks through the limitation of the separation of perception, cognition and decision-making in traditional solutions.

[0014] This invention utilizes a finely tuned visual large model for fine-grained analysis, which can not only identify isolated entities in images, but also accurately extract their entity attributes (such as state and position) and complex relationships between entities, thereby gaining a deep understanding of the scene context. This solves the problem of "superficial perception" pointed out in the background technology, enabling the system to handle complex hazards that require comprehensive judgment, such as "improper wearing of safety helmets" and "materials piled up in fire lanes".

[0015] This invention uses a pre-built knowledge graph in the field of construction site safety to perform multi-hop reasoning on structured metadata. Starting from the initial hidden danger, it can uncover a complete chain of potential and interconnected risks, making up for the core shortcoming of existing technologies that "lacks cognitive reasoning". This enables the system to have a deep risk assessment capability similar to that of human experts.

[0016] This invention, based on a risk chain generated through reasoning and combined with dynamic contextual information, drives a large language model to generate a report containing specific hazard descriptions, risk analysis, regulatory basis, and rectification measures. The generated decision support content is highly targeted, based on clear evidence, and with specific steps, greatly improving the problems of "weak decision support" and "poor practicality of measures" in the background technology, and can effectively guide on-site rectification. Attached Figure Description

[0017] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0018] Figure 1 This is a flowchart of a smart construction site hazard assessment and decision generation method as shown in Example 1.

[0019] Figure 2 This is a diagram illustrating the online reasoning and offline construction process of the knowledge graph in Example 1. Figure 3 This is a diagram of the multi-hop reasoning process of the knowledge graph in Example 1. Detailed Implementation

[0020] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0021] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0022] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0023] Example 1 One embodiment of the present invention provides a method for smart construction site hazard assessment and decision generation, comprising: S1. Obtain the original image containing construction site safety hazards, and use the fine-tuned visual large model to perform fine-grained analysis on the original image to identify and extract entities, entity attributes and relationships between entities, and generate structured metadata. S2. Utilize a pre-built knowledge graph of construction site safety to perform multi-hop reasoning on structured metadata converted into triples, and mine and generate a complete risk chain; S3. Based on the generated risk chain and combined with dynamic context information, construct prompt words and input them into the language model to generate a safety hazard rectification report that includes hazard description, risk analysis, regulatory basis and specific rectification measures.

[0024] As one embodiment, the present invention provides a smart construction site hazard assessment and decision generation method. Safety officers can obtain in-depth risk chain analysis reports, rectification measures, and supporting data through on-the-spot photo capture, based on the fusion of computer vision, knowledge graph, and natural language processing technologies. This supports safety officers on construction sites in flexibly and proactively conducting hazard identification. Figure 1 As shown, the specific implementation process is as follows: 1. Fine-grained visual perception and structured data generation Using a large visual model fine-tuned based on LoRA technology, entities, their state attributes, location attributes, and relationship attributes between entities are identified from the original image. These are then linked to direct risk descriptions to generate structured JSON data containing entities, states, locations, relationships, direct risks, and confidence levels. Specifically: Image data of construction site safety hazards were collected and labeled to generate a structured description dataset. Based on this dataset, the basic visual model was fine-tuned using LoRA technology.

[0025] Specifically, images of various safety hazards at construction sites were collected, including typical scenarios such as inadequate protection, equipment malfunctions, and operational violations. These images needed to encompass diverse lighting conditions, angles, and complex situations. Blurry and occluded images underwent enhancement and noise reduction, while excessively low-quality images were discarded. Each image was annotated with structured descriptive text. The annotated data underwent a process of initial annotation by annotators, review by safety personnel, and consistency checks to ensure an accuracy rate of ≥95%. High-quality content with clearly defined hazards, clear descriptions, and high risk was selected as the training dataset, while a portion was used as the test dataset. The structured descriptive text can be referenced as follows: { "entity": "cable", "state": "Scattered and messy after mopping", "position": "Entrance to the foundation pit", "relation": "near: water accumulation", "immediate_risk": "electrocution", "confidence": 0.90 } The JSON annotations are then converted into natural language descriptions to provide further content descriptions for the large visual model. For example, the image shows that the cable is located at the entrance of the foundation pit, close to the water accumulation, which poses a risk of electric shock, with a confidence level of 90%.

[0026] Qwen2.5vl was selected as the basic large-scale visual model. The large-scale visual model was then fine-tuned to address security risks using the high-quality dataset obtained above.

[0027] The large model was fine-tuned using Low Rank Adaptation (LoRA), and the hyperparameter configuration was confirmed based on ablation experiments. The visual large model was guided by prompt words to generate JSON structured output, and the recognition results were verified by manual review on the test set. The evaluation requirements were that the entity and attribute (state, location, etc.) recognition accuracy was greater than or equal to 92%, and the relation extraction F1 score was ≥0.89.

[0028] Using a finely tuned visual model, deep analysis is performed on images uploaded by safety officers. Through prompt word engineering, the model is guided to identify entities in the images (such as "rebar" and "scaffolding") and accurately extract their state attributes (such as "scattered" and "loose") and various relational attributes (such as "close to: edge protection"), directly associating them with basic surface risks (such as "tripping") and generating the confidence score of the current entry (such as "0.9"). Finally, the model outputs structured metadata in a uniform format, such as JSON, laying the foundation for subsequent knowledge graph processing.

[0029] 2. Knowledge Graph Deep Reasoning and Risk Chain Mining like Figure 2 As shown, the explanation will cover both offline construction and online inference: (1) Offline construction of knowledge graph Using a large language model, knowledge is extracted from safety specification documents and historical rectification reports to form a triplet knowledge base with (entity, relation, attribute) as the basic unit, and stored in a graph database, i.e., a knowledge graph. Specifically: Construct a knowledge graph specifically for the field of construction site safety; this graph integrates a large amount of data on safety hazards at construction sites, national safety production regulations, industry standards, corporate safety rules and regulations, historical accident cases, and other diverse knowledge, storing entities, attributes, and relationships in the form of triples.

[0030] Specifically, the Qwen3 language model is used to parse documents such as the "Construction Safety Inspection Standard," extracting rules such as "prohibited behaviors," "mandatory measures," and "applicable scenarios" to form triples of (entity, constrained behavior, and regulatory clause). Historical rectification reports are used as text to extract complete causal chains of "hazardous phenomena" → "direct causes" → "root causes" → "rectification measures" → "verification results." Combined with labeled data from the fine-tuning of the visual model, a knowledge graph is constructed, primarily involving entity type design, entity attribute design, and relationship type design.

[0031] The knowledge graph in the field of construction site safety includes entity types, entity attributes, and relationship types. The entity types include hazard entities, risk type entities, standard and measure entities, and operation type entities. The entity attributes include ID, name, status, location, and confidence level. The relationship types include causal relationships, transmission relationships, and subordinate relationships.

[0032] Specifically, entity type design includes hazard entities, such as workers, equipment (e.g., scaffolding), materials (e.g., steel bars), and environment (e.g., oil spills); risk types, such as slipping, falling, electric shock, and being struck by objects, which are also entities and target nodes for reasoning; norms and measures, such as immediate cleanup, setting up warning signs, and Article X of the "Safety Standard"; and abstract concepts, such as work type entities like working at height and hot work, which can be used to classify and analyze hazards at a higher level.

[0033] Entity attribute design is mainly based on entity type. Different entity types may contain different entity attributes, but in general, they still have common entity attributes, such as ID, name, status, location, confidence level, etc.

[0034] The design of relationship types includes causal relationships, such as (oil stains on the ground) -- [leads to] --> (slip risk); transmission relationships, such as (slip risk) -- [may trigger] --> (fall risk), which is the key to realizing multi-hop reasoning and discovering potential chain risks; and subordinate relationships, such as (corrective measures) -- [belong to] --> (the responsibilities of a certain department), which facilitates the clarification of responsibilities.

[0035] The construction of a knowledge graph uses a graph database to store these entities and relationships. Each entity is a node, and each relationship is a directed edge. Importing the above data into the graph database completes the construction of the knowledge graph.

[0036] (2) Online reasoning based on knowledge graphs Upon receiving the structured metadata generated by the visual big data model, the knowledge graph's reasoning engine performs predefined or dynamic multi-hop reasoning. Starting with the triples obtained from the structured metadata, it uses a sequential initiation and chain query approach to gradually discover hidden risks through the relationships between entities and connections.

[0037] Specifically, after receiving the JSON information, it is converted into the triple structure required by the knowledge graph, as shown below: Attributes: (Cable, Status, Scattered on the floor) Attributes: (Cable, located at the pit entrance) Relationship: (Cable, Nearby, Water Accumulation) Relationship: ("Cable - proximity - water accumulation", resulting in electric shock) When processing such triples, a sequential, chained query approach is adopted. Based on the reasoning process of the knowledge graph, hidden risks are gradually discovered through the relationships between entities and connections. Although single-query multi-hops support merged queries, splitting them into independent queries is more beneficial for error isolation and interpretability requirements. The reasoning engine will enter the relevant fields based on predefined rules, semantic relationships, or dynamic rules. Figure 3 The following in-depth reasoning is shown: Spatial attribute confirmation (Query 1, Query 2): By determining that the "cable" is located at the "foundation pit entrance", and further identifying that the entrance is the "main construction passage / fire passage", a scenario context is provided for the subsequent risk chain.

[0038] Attribute Hazard Reasoning (Query 3, Query 4): Based on the "functional classification" of the channel, the "channel blockage" hazard is inferred and linked to the "impact on access to fire-fighting equipment" hazard.

[0039] Causal risk chain construction (queries 5 and 6): Combining the two conditions of "water accumulation" and "scattered floor after mopping", the potential hazards of "electrical short circuit" and "fire" are inferred through the dynamic path of the knowledge graph and the predefined path.

[0040] Rectification Regulations Inquiry (Inquiry 7): Based on the potential hazard scenario, inquire about the regulations and measures for use in the subsequent generative decision-making process.

[0041] Through the above multi-hop reasoning, the final reasoning result, i.e., the complete risk chain, is obtained as follows: “(1) Core hidden danger: Cables are dragging on the ground and are close to water accumulation, blocking the entrance to the foundation pit; (2) Risk Analysis: The direct risk is electric shock; the derivative risk is blocked passageway affecting fire emergency response; the superimposed risk is electrical short circuit causing fire; (3) Regulatory basis: Violation of Article Z of the "Technical Specification for Temporary Power Supply Safety at XXXX Construction Site"; (4) Corrective measures: Immediately clean up and install anti-slip mats. To better illustrate the reasoning logic, here is another example of reasoning: From the fact that "there is oil on the ground" (entity 1) and "it is located at the top of the stairs" (location attribute), we can deduce that "there is a high risk of slipping" (risk 1), and further associate it with "Article X of the Technical Specification for Safety of High-Altitude Operations in Building Construction" (basis) and "clean up immediately and install anti-slip mats" (measure). This reasoning ability can discover hidden and complex risk chains.

[0042] 3. Generative Decision Output and Corrective Actions By combining the complete results derived from the knowledge graph (including hazard descriptions, risk chains, related regulatory clauses, and rectification measures) with dynamic contextual information (such as weather, time, or other textual information from safety officers), dynamic prompt words are constructed. A safety hazard rectification opinion report is generated through a large language model. This report lists the problems, clarifies the root causes of the risks, cites specific regulations, and provides targeted rectification steps based on the actual situation.

[0043] Taking the results of the above multi-hop reasoning process as an example, and combining dynamic context information, the following prompt word template is constructed: "You are an experienced, meticulous, and rigorous construction safety risk inspector. Based on the reference information provided by the user, please list the core hidden dangers, explain the root causes of the risks, refer to the relevant standards and regulations, and combine them with rectification measures to provide a hidden danger analysis report and rectification suggestions."

[0044] Requirements: 1. Strictly adhere to the hidden dangers and risks information provided by users, and do not expand the scope without authorization. 2. Strictly adhere to the specifications provided by the user. 3. The corrective measures must include, but are not limited to, the corrective measures provided by the user. 4. The output must include specific core hazards, risk analysis, regulatory basis, and rectification measures. The following is reference information provided by the user: Core vulnerability: {Filling in the blanks with inference results} Risk Analysis: {Fill in using reasoning results} Standard basis: {Fill in using reasoning results} Corrective measures: {Fill in the blanks with inference results} Other information: {dynamic context information}.

[0045] Specifically, the system accepts the complete risk chain derived from the knowledge graph, the relevant basis for regulatory measures, and dynamic contextual information (such as weather, time, or other textual information from the safety officer). Based on the QWen3 language model, it uses prompt words to generate a hazard report and rectification suggestions. Continuing from the previous example, the generated safety hazard rectification report includes a hazard description, risk analysis, regulatory basis, and specific rectification measures as follows: I. Description of Core Hidden Dangers Upon inspection, serious safety hazards were found at the entrance area of ​​the foundation pit at the construction site: temporary power cables were not laid in accordance with regulations, were scattered haphazardly on the ground, and some cables were close to water accumulation areas, while also blocking the main entrance passage of the foundation pit. II. Risk Analysis This hidden danger constitutes a comprehensive security threat with multiple overlapping risks, as detailed below: Direct risk: Danger of electric shock to personnel 1) If the cable is dragged directly on the ground, it is easily scratched by hard objects, which will cause damage to the insulation layer.

[0046] 2) Cables near or submerged in water greatly increase the risk of leakage. Once the insulation is damaged, the water and damp ground will become live conductors, seriously threatening the lives of workers in the vicinity.

[0047] Derivative risk: Emergency access blocked Scattered cables blocked the pit entrance, obstructing safety passages. In the event of emergencies such as electric shock, fire, or pit collapse, this would severely hinder the rapid evacuation of personnel and the effective access of emergency rescue personnel and equipment, potentially causing secondary disasters or exacerbating the accident's consequences.

[0048] Additional risks: electrical fire risk Cable insulation, when aged and damaged by dragging, rolling, or soaking in water, is highly susceptible to short circuits and the generation of electric arcs. If flammable materials are present nearby, there is a very high probability of igniting a fire, resulting in a combined accident of electric shock and fire.

[0049] III. Standard Basis This potential hazard directly violates the basic regulations for safety management at construction sites: 1) Violation of Article Z of the "Safety Technical Specifications for Temporary Power Supply at XXXX Construction Site" which stipulates that "cable trunk lines are strictly prohibited from being laid openly along the ground, and should be laid overhead or underground to avoid mechanical damage and media corrosion".

[0050] 2) Violation of relevant regulations regarding keeping fire and safety exits unobstructed.

[0051] IV. Corrective Measures To ensure immediate elimination of risks and prevent accidents from occurring, the following targeted corrective measures have been formulated: Immediate cleanup and isolation (short-term measures) 1) Immediately disconnect the power supply to the area and have a qualified electrician perform the operation.

[0052] 2) Clear all cables blocking the pit entrance to ensure the passage is completely unobstructed.

[0053] 3) Set up anti-slip mats and temporary warning lines around the waterlogged area to prevent people from accidentally entering and to provide dry working conditions for electrical work.

[0054] Standardized installation (fundamental measure) 1) Thoroughly rectify the cables that are dragged on the ground, and adopt overhead laying (such as using insulated hook brackets) or conduit protection in accordance with the specifications to ensure that the cables are kept away from water accumulation areas.

[0055] 2) The height of the overhead cable or the depth of its burial must meet safety standards, and special protective measures must be taken at the crossing points (such as adding steel sleeves or protective trenches).

[0056] System maintenance and education (long-term prevention) 1) Conduct a comprehensive inspection of all temporary power lines at the construction site, learn from this experience, and eliminate similar potential hazards.

[0057] 2) Provide specialized safety training and briefings to workers, especially electricians and operators of frequently moving equipment, emphasizing temporary electrical safety regulations and eliminating the concept of "temporary."

[0058] Example 2 One embodiment of the present invention provides a smart construction site hazard assessment and decision generation system, comprising: The visual perception and structured data generation module is configured to: acquire original images containing construction site safety hazards, perform fine-grained analysis on the original images using a finely tuned visual big model, identify and extract entities, entity attributes and relationships between entities, and generate structured metadata. The knowledge graph reasoning and risk chain mining module is configured to: use a pre-built knowledge graph in the field of construction site safety to perform multi-hop reasoning on structured metadata converted into triples, and mine and generate a complete risk chain; The generative decision report output module is configured to: based on the generated risk chain and combined with dynamic context information, construct prompt words and input them into the language big model to generate a safety hazard rectification report that includes hazard description, risk analysis, regulatory basis and specific rectification measures.

[0059] Example 3 One embodiment of the present invention provides a computer program product, including a computer program, which, when executed by a processor, implements the aforementioned method for assessing and generating intelligent construction site hazards.

[0060] Example 4 In one embodiment of the present invention, a non-transitory computer-readable storage medium is provided for storing computer instructions. When the computer instructions are executed by a processor, the method for assessing and generating intelligent construction site hazards is implemented.

[0061] Example 5 One embodiment of the present invention provides an electronic device, including: a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory, so that the electronic device executes the aforementioned method for assessing and generating intelligent construction site hazards.

[0062] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0063] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0064] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A method for intelligent construction site hazard assessment and decision generation, characterized in that, include: The original image containing construction site safety hazards is acquired, and a fine-tuned visual model is used to perform fine-grained analysis on the original image to identify and extract entities, entity attributes and relationships between entities, and generate structured metadata. By utilizing a pre-built knowledge graph of construction site safety, multi-hop reasoning is performed on structured metadata converted into triples to mine and generate a complete risk chain; Based on the generated risk chain and combined with dynamic context information, prompt words are constructed and input into the language model to generate a safety hazard rectification report that includes hazard description, risk analysis, regulatory basis, and specific rectification measures.

2. The method for intelligent construction site hazard assessment and decision generation as described in claim 1, characterized in that, The generation of structured metadata utilizes a large visual model fine-tuned based on LoRA technology to identify entities, their state attributes, location attributes, and relationship attributes between entities from the original image, and associates them with direct risk descriptions to generate structured JSON data containing entities, state, location, relationships, direct risks, and confidence levels.

3. The method for intelligent construction site hazard assessment and decision generation as described in claim 1, characterized in that, The process of constructing a knowledge graph in the field of construction site safety includes: extracting knowledge from safety specification documents and historical rectification reports using a large language model, forming a triplet knowledge base with (entity, relation, attribute) as the basic unit, and storing it in a graph database.

4. The method for intelligent construction site hazard assessment and decision generation as described in claim 1, characterized in that, The knowledge graph for construction site safety includes entity types, entity attributes, and relationship types. The entity types include hazard entities, risk type entities, standard and measure entities, and operation type entities. The entity attributes include ID, name, status, location, and confidence level. The relationship types include causal relationships, transmission relationships, and subordinate relationships.

5. The method for intelligent construction site hazard assessment and decision generation as described in claim 1, characterized in that, The multi-hop reasoning starts with the triples obtained from the transformation of the structured metadata, and uses a sequential initiation and chain query approach to gradually discover hidden risks through the relationship between entities and connections.

6. The method for intelligent construction site hazard assessment and decision generation as described in claim 1, characterized in that, It also includes model training, collecting and labeling image data of construction site safety hazards, generating a structured description dataset, and using LoRA technology to fine-tune the basic visual model based on the dataset.

7. A smart construction site hazard assessment and decision generation system, characterized in that, include: The visual perception and structured data generation module is configured to: acquire original images containing construction site safety hazards, perform fine-grained analysis on the original images using a finely tuned visual big model, identify and extract entities, entity attributes and relationships between entities, and generate structured metadata. The knowledge graph reasoning and risk chain mining module is configured to: use a pre-built knowledge graph in the field of construction site safety to perform multi-hop reasoning on structured metadata converted into triples, and mine and generate a complete risk chain; The generative decision report output module is configured to: based on the generated risk chain and combined with dynamic context information, construct prompt words and input them into the language big model to generate a safety hazard rectification report that includes hazard description, risk analysis, regulatory basis and specific rectification measures.

8. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the method for intelligent construction site hazard assessment and decision generation as described in any one of claims 1-6.

9. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium is used to store computer instructions, which, when executed by a processor, implement a smart construction site hazard assessment and decision generation method as described in any one of claims 1-6.

10. An electronic device, characterized in that, include: The device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the method for assessing and generating intelligent construction site hazards as described in any one of claims 1-6.

Images