Construction site potential safety hazard record acquisition system and method based on artificial intelligence
The construction site safety hazard record acquisition system based on artificial intelligence utilizes a multimodal large model and a construction hazard knowledge graph to automatically identify construction sites, solving the problems of low efficiency and low accuracy of manual inspections in existing technologies. This achieves intelligent construction site safety management and improves the accuracy of hazard identification.
Patent Information
- Application Number
- CN202511682701.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-17
- Publication Date
- 2026-03-27
AI Technical Summary
Current technologies rely on manual inspections for identifying safety hazards at construction sites, which is inefficient and inaccurate, making it difficult to achieve automated and intelligent management.
An AI-based construction site safety hazard record acquisition system is adopted. It uses a multimodal large model combined with a construction hazard knowledge graph to identify construction image data, generate hazard judgment prompt words and identify them, and uses deep learning algorithms to perform feature fusion to generate hazard records.
It has improved the accuracy and timeliness of identifying potential hazards at construction sites, reduced the incidence of safety accidents, and realized the intelligent and automated management of safety at construction sites.
Smart Images
Figure CN121746758A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of safety management in the construction industry, and particularly relates to a construction site safety hidden danger record acquisition system and method based on artificial intelligence. BACKGROUND
[0002] In the field of construction engineering, with the increasingly high requirements of the society on the safety of construction engineering construction, the importance of construction engineering construction safety management is increasingly prominent. Due to the complex construction site environment, a large number of personnel, a variety of equipment, and various potential safety hazards such as unsafe personnel behavior, equipment failure, and environmental factor abnormalities, these may all lead to serious safety accidents.
[0003] At present, in the construction process of construction engineering, the on-site safety analysis mainly depends on manual inspection and experience judgment, which is low in efficiency and greatly influenced by subjective factors. Different patrol personnel may make different judgments on the same safety hidden danger, resulting in low accuracy and reliability of safety analysis. With the vigorous development of computer technology and artificial intelligence technology, image recognition technology is introduced into the construction site to identify potential safety hazards and identify and analyze the construction site. However, the recognition accuracy is not good, the adaptability to complex construction scenes is not enough, and there is a lack of reliable basis and effective disposal suggestions for hidden dangers.
[0004] In order to avoid huge economic losses and safety accidents caused by inaccurate identification and judgment of hidden dangers in the process of engineering construction, the technical personnel in the field have been seeking a construction site safety hidden danger record acquisition method based on artificial intelligence, so as to realize the automatic identification of hidden danger scenes and hidden danger items in the construction site, improve the accuracy of hidden danger identification, the timeliness of processing and the intelligent level of management, and meet the needs of construction safety hidden danger identification. SUMMARY
[0005] The purpose of the present application is to provide a construction site safety hidden danger record acquisition system and method based on artificial intelligence, which can automatically identify hidden danger scenes and hidden danger items in the construction site, provide strong support for the safety management of the construction site, and reduce the incidence of safety accidents.
[0006] The construction site safety hidden danger record acquisition system based on artificial intelligence designed by the present application has the following special features: a data acquisition module for acquiring image data of the construction site; The hidden danger identification module is configured to utilize the multi-modal large model in combination with a construction hidden danger scene identification prompt word to judge a construction hidden danger scene in image data of a construction site, obtain a type of the construction hidden danger scene, query a plurality of hidden danger items and corresponding hidden danger standard attributes contained in the type of the construction hidden danger scene through a pre-constructed construction hidden danger knowledge graph, generate a corresponding hidden danger judgment prompt word for each hidden danger item and corresponding hidden danger standard attribute, and input the hidden danger judgment prompt word and the image data of the construction site into the multi-modal large model for hidden danger identification to obtain a hidden danger identification result. The hidden danger record acquisition module is configured to judge whether there is at least one hidden danger anomaly in the hidden danger identification result, and if so, acquire hidden danger standard attributes from hidden danger items in the hidden danger identification result, and generate a hidden danger record according to the hidden danger standard attributes.
[0007] Further, a specific construction method of the pre-constructed construction hidden danger knowledge graph is as follows: historical construction hidden danger data is acquired, the historical construction hidden danger data is pre-processed to obtain pre-processed historical construction hidden danger data, entity recognition is performed on the pre-processed historical construction hidden danger data through a named entity recognition technology to obtain hidden danger scene entities and hidden danger item entities, attributes of the hidden danger scene entities and the hidden danger item entities are extracted to obtain attributes of the hidden danger scene entities and the hidden danger item entities, a semantic relationship between the hidden danger scene entities and the hidden danger item entities is identified through a relationship extraction algorithm, the extracted entities, attributes and semantic relationships are converted into a graph structure, nodes in the graph structure and edges between the nodes are created, and the nodes and the edges are converged to obtain the pre-constructed construction hidden danger knowledge graph.
[0008] Further, the pre-constructed construction hidden danger knowledge graph includes nodes of hidden danger scene entities and nodes of hidden danger item entities, the nodes of the hidden danger scene entities include a scene name attribute and a scene description attribute, the nodes of the hidden danger item entities include a hidden danger name attribute, a hidden danger level attribute, a hidden danger standard attribute, a hidden danger basis attribute and a hidden danger suggestion attribute, and the nodes of the hidden danger scene entities and the nodes of the hidden danger item entities are associated through the hidden danger level attribute, i.e., edges between the nodes of the hidden danger scene entities and the nodes of the hidden danger item entities.
[0009] Further, the hidden danger identification module utilizes the multi-modal large model in combination with a construction hidden danger scene identification prompt word to judge a construction hidden danger scene in image data of a construction site, and a specific method is as follows: a scene identification prompt word template is acquired, hidden danger scenes and hidden danger standards in the scene identification prompt word template are occupied by variables, corresponding hidden danger scenes and hidden danger standards are acquired through the construction hidden danger knowledge graph, the corresponding hidden danger scenes and hidden danger standards are filled into the above-mentioned positions through variables, a construction hidden danger scene identification prompt word is obtained, the construction hidden danger scene identification prompt word and the image data of the construction site are input into the multi-modal large model, and a type of the construction hidden danger scene is obtained.
[0010] Further, the hidden danger identification module queries the multiple hidden danger items and corresponding hidden danger standard attributes contained in the type of the construction hidden danger scene through the pre-constructed construction hidden danger knowledge graph, generates a corresponding hidden danger judgment prompt word for each hidden danger item and corresponding hidden danger standard attribute, and the specific method is: extracting the node of the hidden danger scene entity from the construction hidden danger knowledge graph according to the type of the construction hidden danger scene, obtaining multiple hidden danger item nodes associated with the node of the hidden danger scene entity and the hidden danger standard attributes corresponding to each hidden danger item; obtaining a hidden danger judgment prompt word template, the hidden danger items and hidden danger standards in the hidden danger judgment prompt word template are occupied by variables, and the corresponding hidden danger items and hidden danger standards are filled into the above-mentioned occupation through variables to obtain a hidden danger judgment prompt word.
[0011] Further, the hidden danger identification module inputs the hidden danger judgment prompt word and the image data of the construction site into the multi-modal large model for hidden danger identification, and the specific method is: encoding the hidden danger judgment prompt word to obtain a semantic feature vector of the prompt word; extracting features from the image data of the construction site to obtain a feature vector of the image; using a multi-modal feature fusion network to perform feature alignment and fusion on the semantic feature vector and the feature vector of the image to obtain a fusion feature representation; classifying and positioning the hidden danger target in the fusion feature representation through the multi-modal large model to output a hidden danger identification result.
[0012] Further, the hidden danger evaluation module judges whether there is at least one hidden danger anomaly in the hidden danger identification result, and the specific method is: obtaining the hidden danger identification confidence of each hidden danger identification of the multi-modal large model, judging whether the hidden danger identification confidence is less than the preset hidden danger identification confidence threshold, if less than the preset hidden danger identification confidence threshold, obtaining the hidden danger attribute corresponding to the hidden danger identification confidence, generating a hidden danger record according to the hidden danger attribute, and if greater than or equal to the preset hidden danger identification confidence threshold, the identification result is invalid, and no hidden danger record is generated.
[0013] Further, the system described above further comprises: a historical hidden danger investigation module; the historical hidden danger investigation module is used for obtaining historical hidden danger records, retrieving corresponding historical hidden danger scenes and hidden danger items through the historical hidden danger records, comparing the hidden danger scenes and hidden danger items in the current hidden danger records with the hidden danger scenes and hidden danger items in the historical hidden danger records, and judging whether the historical hidden danger is excluded.
[0014] The building site safety hidden danger record acquisition method based on artificial intelligence designed to achieve the second purpose of the application has the special feature that it includes the following steps: Obtaining image data of a construction site; The multi-modal large model is used to judge the construction hidden danger scene in the image data of the construction site by combining the construction hidden danger scene identification prompt word, obtain the type of the construction hidden danger scene, query a plurality of hidden danger items and corresponding hidden danger standard attributes contained in the type of the construction hidden danger scene through a pre-constructed construction hidden danger knowledge graph, generate a corresponding hidden danger judgment prompt word for each hidden danger item and corresponding hidden danger standard attribute, and input the hidden danger judgment prompt word and the image data of the construction site into the multi-modal large model for hidden danger identification to obtain a hidden danger identification result. If there is at least one hidden danger exception in the hidden danger identification result, the hidden danger standard attribute is obtained from the hidden danger item in the hidden danger identification result, and the hidden danger record is generated according to the hidden danger standard attribute.
[0015] A computer program product designed for the third aspect of the present application includes computer instructions for causing a computer to execute the above-mentioned artificial intelligence-based construction site safety hidden danger record acquisition method.
[0016] The present application has the following advantages: (1) The artificial intelligence-based construction site safety hidden danger record acquisition system and method uses the hidden danger identification module to identify the scene of the construction image data by using the deep learning algorithm, and combines the prompt word and the image data to input into the multi-modal large model for hidden danger determination, realizes the bidirectional fusion identification of image semantic features and text knowledge features, generates identification prompt words for different construction hidden danger scenes, and the prompt words contain hidden danger names, determination standards and judgment condition examples, so that the multi-modal large model can focus on the hidden danger related feature area in the reasoning stage, and the misidentification rate is reduced.
[0017] (2) The artificial intelligence-based construction site safety hidden danger record acquisition system and method uses the knowledge graph construction module to perform entity recognition and relationship extraction on historical construction hidden danger data to form a graph structure containing hidden danger scene nodes and hidden danger item nodes; in the hidden danger identification process, the system can automatically retrieve related hidden danger items and determination standards based on the construction scene type, provide knowledge constraints and semantic context for the model, realize the fusion from "knowledge recognition" to "knowledge reasoning", and improve the accuracy and professionalism of hidden danger identification. BRIEF DESCRIPTION OF DRAWINGS
[0018] Figure 1 A module schematic diagram of one specific embodiment of the artificial intelligence-based construction site safety hidden danger record acquisition system of the present application is shown.
[0019] Figure 2 A flowchart of one specific embodiment of the artificial intelligence-based construction site safety hidden danger record acquisition method of the present application is shown. DETAILED DESCRIPTION
[0020] Exemplary embodiments of the present disclosure will be described in greater detail below with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it is understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided so that the present disclosure can be more thoroughly understood, and the scope of the present disclosure can be accurately conveyed to those skilled in the art.
[0021] As shown in Figure 1 and Figure 2 The embodiments of the present application disclose an artificial intelligence-based construction site safety hazard record acquisition system and method, which can satisfy the automatic identification of construction site hazard scenes and hazard items.
[0022] Embodiment 1 The present embodiment discloses an artificial intelligence-based construction site safety hazard record acquisition system, which comprises: a data acquisition module for acquiring image data of a construction site; a hazard identification module for judging construction hazard scenes in the image data of the construction site by using a multi-modal large model combined with construction hazard scene identification prompt words, obtaining the type of the construction hazard scene, querying a pre-constructed construction hazard knowledge graph for a plurality of hazard items and corresponding hazard standard attributes contained in the type of the construction hazard scene, generating corresponding hazard judgment prompt words for each hazard item and corresponding hazard standard attribute, and inputting the hazard judgment prompt words and the image data of the construction site into the multi-modal large model for hazard identification to obtain a hazard identification result; a hazard record acquisition module for judging whether there is at least one hazard anomaly in the hazard identification result, and if so, acquiring hazard standard attributes from the hazard items of the hazard identification result, and generating a hazard record according to the hazard standard attributes.
[0023] It should be noted that the multi-modal large model can be a BLIP, GPT-4V, Qwen-VL, etc.
[0024] In this embodiment, the specific construction of the pre-constructed construction hazard knowledge graph method is: obtaining historical construction hazard data, preprocessing the historical construction hazard data to obtain pre-processed historical construction hazard data, it should be noted that preprocessing the historical construction hazard data includes: text segmentation, word segmentation and removing stop words of historical construction hazard data; entity recognition is performed on the pre-processed historical construction hazard data by named entity recognition technology to obtain hazard scene entities and hazard item entities, and the attributes of the hazard scene entities and the hazard item entities are extracted to obtain the attributes of the hazard scene entities and the attributes of the hazard item entities. It should be noted that attribute extraction specifically extracts the context semantics of each recognized entity, binds the extracted attributes to the corresponding entities, and forms an "entity-attribute" pair; the semantic relationship between the hazard scene entities and the hazard item entities is identified by a relationship extraction algorithm, and the extracted entities, attributes and semantic relationships are converted into a graph structure. It should be noted that according to the relationship category, the hazard scene entities, hazard item entities and corresponding relationships are stored in a triple form, the nodes in the graph structure and the edges between the nodes are created, and the nodes and edges are aggregated to obtain the pre-constructed construction hazard knowledge graph.
[0025] In this embodiment, the pre-constructed construction hazard knowledge graph includes: nodes of a plurality of hazard scene entities and nodes of hazard item entities, the nodes of the hazard scene entities include: scene name attributes and scene description attributes, and the nodes of the hazard item entities include: hazard name attributes, hazard level attributes, hazard standard attributes, hazard basis attributes and hazard suggestion attributes. The nodes of the hazard scene entities and the nodes of the hazard item entities are associated through the hazard level attributes, i.e. the edges between the nodes of the hazard scene entities and the nodes of the hazard item entities.
[0026] In this embodiment, the hazard identification module uses a multi-modal large model to combine construction hazard scene identification cues to judge the construction hazard scene in the image data of the construction, and the specific method is: obtaining a scene identification cue template, it should be noted that the template is obtained by clustering and inducing high-frequency prompt sentences from historical hazard records, the hazard scene and hazard standard in the scene identification cue template are occupied by variables, for example:
scene
[0027] The construction hazard scene identification cues are, for example: Please first determine which construction hidden danger scenarios the picture involves, and the corner areas and long-range areas in the picture and the blurred areas are omitted, and the core visible areas are focused on:
Distribution box interior
[0028]
Distribution box exterior
[0029]
Pile hole operation area
[0030]
Foundation pit operation area
[0031] Note that the above prompt words are example-level prompt words, constructed through dynamic context.
[0032] Important note: 1. Only return the construction hidden danger scene name, do not add any explanation 2. Start judging the possible existence of the above
construction hidden danger name
construction hidden danger name
[0033] Among them, the special scene prompt words are as follows: According to the engineering safety related knowledge and experience to identify whether there is a safety hazard, If there is a safety hazard, describe all the violations in the picture in the following fixed format, Only list the violations and confidence, do not describe other content.
[0034] # Violation: 1. Violation 1, confidence 2. Violation 2, confidence 3. Violation 3, confidence It should be noted that the above prompt words are task-level prompt words.
[0035] Please identify whether the following safety hazards exist in the picture:
Fire extinguisher
Judgment condition: When only a single fire extinguisher is identified within the picture range, this hazard is triggered. When the image clearly shows the fire extinguisher dial pointer, this hazard is not judged.
Judgment condition: Only when the visual angle clearly shows that the fire extinguisher dial pointer points to the red area, this hazard is triggered, otherwise this hazard is not triggered. When the pointer points to the yellow area, this hazard is not triggered.
Judgment condition: When the fire extinguisher is clearly identified as falling down (not straight up) within the picture range, this hazard is triggered.
Judgment condition: Only when the visual angle clearly shows that the fire extinguisher dial pointer points to the yellow area, this hazard is triggered, otherwise this hazard is not triggered.
[0036] 3. If there is no hazard in <>, return an empty string directly.
[0037] It should be noted that the above is the output constraint.
[0038] In this embodiment, the hidden danger identification module queries the multiple hidden danger items and corresponding hidden danger standard attributes contained in the type of the construction hidden danger scene through the pre-constructed construction hidden danger knowledge graph, generates a corresponding hidden danger judgment prompt word for each hidden danger item and corresponding hidden danger standard attribute, and the specific method is: extracting the node of the hidden danger scene entity from the construction hidden danger knowledge graph according to the type of the construction hidden danger scene, obtaining multiple hidden danger item nodes associated with the node of the hidden danger scene entity and hidden danger standard attributes corresponding to each hidden danger item; obtain a hidden danger judgment prompt word template, the hidden danger items and hidden danger standards in the hidden danger judgment prompt word template are occupied by variables, and the corresponding hidden danger items and hidden danger standards are filled into the above-mentioned placeholders through variables to obtain a hidden danger judgment prompt word.
[0039] The hidden danger judgment prompt word, for example:
Fire extinguisher
Judgment condition: trigger this hidden danger when only a single fire extinguisher is identified within the image range. When the image clearly shows the fire extinguisher dial pointer, do not judge this hidden danger.
Judgment condition: trigger this hidden danger only when the fire extinguisher dial pointer is clearly visible in the view angle and points to the red area, otherwise do not trigger this hidden danger. Do not trigger this hidden danger when the pointer points to the yellow area.
Judgment condition: trigger this hidden danger when the fire extinguisher falls down (not straight up) within the image range.
Judgment condition: trigger this hidden danger only when the fire extinguisher dial pointer is clearly visible in the view angle and points to the yellow area, otherwise do not trigger this hidden danger.
[0040] 3. If there is no hidden danger in <>, return an empty string directly.
[0041] In this embodiment, the hidden danger identification module inputs the hidden danger judgment prompt word and the image data of the construction site into the multi-modal large model for hidden danger identification, and the specific method is: performing encoding processing on the hidden danger judgment prompt word to obtain a semantic feature vector of the prompt word; performing feature extraction on the image data of the construction site to obtain a feature vector of the image; using a multi-modal feature fusion network to perform feature alignment and fusion on the semantic feature vector and the feature vector of the image to obtain a fusion feature representation; classifying and positioning the hidden danger target in the fusion feature representation through the multi-modal large model to output a hidden danger identification result.
[0042] It should be noted that the multi-modal large model comprises: a visual feature extraction layer configured to encode features of input construction site image data and extract visual semantic features; a text feature extraction layer configured to encode semantic features of input scene recognition prompt words and hidden danger judgment prompt words; a multi-modal feature fusion layer configured to realize alignment and fusion of text and image semantics through a cross-attention mechanism or a transformer structure; a recognition output layer configured to output a hidden danger recognition result based on the fused feature representation.
[0043] In this embodiment, the hidden danger evaluation module determines whether there is at least one hidden danger anomaly in the hidden danger recognition result. The specific method is as follows: obtaining hidden danger recognition confidence of the multi-modal large model, determining whether the hidden danger recognition confidence is less than a preset hidden danger recognition confidence threshold, if less than the preset hidden danger recognition confidence threshold, obtaining a hidden danger attribute corresponding to the hidden danger recognition confidence, generating a hidden danger record according to the hidden danger attribute, and if greater than or equal to the preset hidden danger recognition confidence threshold, the recognition result is invalid, and no hidden danger record is generated.
[0044] It should be noted that the format of the result output by the multi-modal large model is as follows: 1. Hidden danger behavior, confidence.
[0045] 2. Hidden danger behavior, confidence.
[0046] For example: analysis result of the proprietary scene disc type scaffold: # Violation behavior: 1. Missing load-bearing support acceptance notice board; 0.95 2. Missing or too high from the ground of the scaffold body sweeping rod; 0.80 The hidden danger recognition confidence is obtained by analyzing the returned result.
[0047] In this embodiment, the hidden danger record comprises: hidden danger scene, hidden danger item, hidden danger level, hidden danger basis, hidden danger suggestion and corresponding construction image data.
[0048] Based on the above system, optionally, the system further comprises a historical hidden danger investigation module; the historical hidden danger investigation module is configured to obtain historical hidden danger records, retrieve corresponding historical hidden danger scenes and hidden danger items through the historical hidden danger records, compare the hidden danger scenes and hidden danger items in the current hidden danger record with the hidden danger scenes and hidden danger items in the historical hidden danger records, and determine whether the historical hidden danger is excluded.
[0049] Embodiment 2 The embodiment discloses a construction site safety hidden danger record acquisition method based on artificial intelligence, and the method comprises the following steps: Step 1, acquire construction image data; Step 2, utilize a multi-modal large model to combine a construction hidden danger scene identification prompt word to judge a construction hidden danger scene in the image data of the construction, obtain a type of the construction hidden danger scene, acquire a plurality of hidden danger items and their discrimination standards contained in the type of the construction scene from a pre-constructed knowledge graph based on the type of the construction hidden danger scene, generate a corresponding identification prompt word for each hidden danger item, and input the prompt word and the construction image data into the multi-modal large model for identification to obtain a hidden danger identification result; Step 3, judge whether at least one hidden danger anomaly exists in the hidden danger identification result, if yes, acquire a hidden danger attribute according to the corresponding hidden danger item, and generate a hidden danger record according to the hidden danger attribute.
[0050] Embodiment 3 The embodiment discloses a computer program product, for example, computer program instructions, when executed by a computer, the operation of the computer can call or provide the method and / or technical solution according to the present application. Those skilled in the art should understand that the form of computer program instructions in computer readable medium includes but is not limited to source file, executable file, installation package file and the like, and accordingly, the way of computer program instructions executed by computer includes but is not limited to: the computer directly executes the instruction, or the computer compiles the instruction and then executes the corresponding compiled program, or the computer reads and executes the instruction, or the computer reads and installs the instruction and then executes the corresponding installed program. Here, the computer readable medium can be any available computer readable storage medium or communication medium accessible to the computer. Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above described program and module can refer to the corresponding process description in the foregoing method embodiment, which will not be described here.
[0051] The algorithms and displays provided herein are not inherently related to any particular computer, virtual system, or other apparatus. Various general purpose systems can be used with these teachings, with or without accompanying software. Those skilled in the art will recognize that structures embodied by these descriptions might be subjected to numerous modifications, and yet, are described in terms of structures tailored to flow of data to achieve the best mode of the application.
[0052] In the description provided herein, numerous specific details are set forth. However, it is understood that embodiments of the application can be practiced without these specific details. In some instances, well-known methods, structures and techniques have not been shown in detail in order not to obscure an understanding of this description.
[0053] Similarly, it is to be understood that the embodiments of the present application can be readily combined with one another and / or other concepts, to produce further embodiments of the present application. Similarly, it should be understood that, in order to concisely discuss the various aspects of the present application, some embodiments of the present application have been shown to include certain features, while other embodiments have been shown to include other features. However, this is in no way intended to imply that the features of the various embodiments of the present application cannot be combined. Indeed, the present application expressly contemplates and fully encompasses the various combinations of the features of the various embodiments of the present application. Moreover, it should be understood that any of the features of the present application can be used in any combination with any other feature of the present application.
Claims
1. A construction site safety hazard recording and acquisition system based on artificial intelligence, characterized in that, include: The data acquisition module is used to acquire image data of the construction site; The hazard identification module is used to judge the construction hazard scenes in the image data of the construction site by using a multimodal big data model combined with construction hazard scene identification prompts, to obtain the type of construction hazard scene, to query multiple hazard items and corresponding hazard standard attributes contained in the type of construction hazard scene by using a pre-built construction hazard knowledge graph, to generate corresponding hazard judgment prompts for each hazard item and corresponding hazard standard attribute, and to input the hazard judgment prompts and the image data of the construction site into the multimodal big data model for hazard identification, and to obtain the hazard identification result; The hazard record acquisition module is used to determine whether there is at least one hazard anomaly in the hazard identification result. If so, the hazard standard attribute is obtained from the hazard item in the hazard identification result, and a hazard record is generated based on the hazard standard attribute.
2. The construction site safety hazard record acquisition system based on artificial intelligence according to claim 1, characterized in that, The specific construction method of the pre-constructed construction hazard knowledge graph is as follows: acquire historical construction hazard data, preprocess the historical construction hazard data to obtain preprocessed historical construction hazard data, perform entity recognition from the preprocessed historical construction hazard data using named entity recognition technology to obtain hazard scene entities and hazard item entities, extract attributes from hazard scene entities and hazard item entities to obtain attributes of hazard scene entities and hazard item entities, identify semantic relationships between hazard scene entities and hazard item entities using relation extraction algorithms, transform the extracted entities, attributes, and semantic relationships into a graph structure, create edges between nodes in the graph structure, and aggregate the nodes and edges to obtain the pre-constructed construction hazard knowledge graph.
3. The construction site safety hazard record acquisition system based on artificial intelligence according to claim 2, characterized in that, The pre-constructed construction hazard knowledge graph includes nodes of several hazard scenario entities and nodes of hazard item entities. The nodes of the hazard scenario entities include: scenario name attribute and scenario description attribute. The nodes of the hazard item entities include: hazard name attribute, hazard level attribute, hazard standard attribute, hazard basis attribute, and hazard suggestion attribute. The nodes of the hazard scenario entities and the nodes of the hazard item entities are associated through the hazard level attribute, which is the edge between the nodes of the hazard scenario entities and the nodes of the hazard item entities.
4. The construction site safety hazard record acquisition system based on artificial intelligence according to claim 1, characterized in that, The hazard identification module uses a multimodal large model combined with construction hazard scene identification prompts to judge construction hazard scenes in construction image data. The specific method is as follows: obtain a scene identification prompt template. The hazard scene and hazard standard in the scene identification prompt template are placed by variables. Obtain the corresponding hazard scene and hazard standard through the construction hazard knowledge graph. Fill the above placeholders with the corresponding hazard scene and hazard standard through variables to obtain the construction hazard scene identification prompt. Input the construction hazard scene identification prompt and the construction image data into the multimodal large model to obtain the type of construction hazard scene.
5. The construction site safety hazard record acquisition system based on artificial intelligence according to claim 1, characterized in that, The hazard identification module queries the construction hazard knowledge graph to find multiple hazard items and corresponding hazard standard attributes contained in the type of the construction hazard scenario. For each hazard item and corresponding hazard standard attribute, a corresponding hazard judgment prompt word is generated. The specific method is as follows: extract the node of the hazard scenario entity from the construction hazard knowledge graph according to the type of the construction hazard scenario, and obtain multiple hazard item nodes associated with the node of the hazard scenario entity and the hazard standard attributes corresponding to each hazard item. Obtain the hazard assessment prompt word template. In the hazard assessment prompt word template, the hazard items and hazard standards are used as placeholders by variables. Fill the corresponding hazard items and hazard standards into the placeholders through variables to obtain the hazard assessment prompt words.
6. The construction site safety hazard record acquisition system based on artificial intelligence according to claim 1, characterized in that, The hazard identification module inputs hazard judgment prompts and construction site image data into a multimodal large model for hazard identification. Specifically, the method involves: encoding the hazard judgment prompts to obtain semantic feature vectors; extracting features from the construction site image data to obtain image feature vectors; aligning and fusing the semantic feature vectors with the image feature vectors using a multimodal feature fusion network to obtain a fused feature representation; and classifying and locating the hazard targets in the fused feature representation using the multimodal large model to output the hazard identification result.
7. The construction site safety hazard record acquisition system based on artificial intelligence according to claim 1, characterized in that, The hazard assessment module determines whether there is at least one hazard anomaly in the hazard identification results. The specific method is as follows: obtain the hazard identification confidence of each hazard in the multimodal large model, determine whether the hazard identification confidence of each hazard is less than a preset hazard identification confidence threshold. If it is less than the preset hazard identification confidence threshold, obtain the hazard attribute of the corresponding hazard identification confidence and generate a hazard record based on the hazard attribute. If it is greater than or equal to the preset hazard identification confidence threshold, the identification result is invalid and no hazard record is generated.
8. The construction site safety hazard record acquisition system based on artificial intelligence according to claim 1, characterized in that, Also includes: Historical hidden danger investigation module; The historical hazard investigation module is used to obtain historical hazard records, retrieve corresponding historical hazard scenarios and hazard items through the historical hazard records, compare the hazard scenarios and hazard items in the current hazard records with the hazard scenarios and hazard items in the historical hazard records, and determine whether the historical hazard has been eliminated.
9. A method for obtaining records of safety hazards at construction sites based on artificial intelligence, characterized in that, Includes the following steps: Acquire image data of the construction site; By using a multimodal large model combined with construction hazard scene identification prompts, the construction hazard scenes in the image data of the construction site are judged to obtain the type of construction hazard scene. By querying the multiple hazard items and corresponding hazard standard attributes contained in the type of construction hazard scene through a pre-constructed construction hazard knowledge graph, corresponding hazard judgment prompts are generated for each hazard item and corresponding hazard standard attribute. The hazard judgment prompts and the image data of the construction site are input into the multimodal large model for hazard identification to obtain the hazard identification result. Determine whether there is at least one abnormal hazard in the hazard identification result. If so, obtain the standard attribute of the hazard from the hazard item in the hazard identification result, and generate a hazard record based on the standard attribute of the hazard.
10. A computer program product, characterized in that, It includes computer instructions, which are used to cause a computer to execute the artificial intelligence-based construction site safety hazard record acquisition method as described in claim 9.