Construction safety intelligent management and control method and system based on potential safety hazard library

Abstract

The invention provides a construction safety intelligent management and control method and system based on a potential safety hazard library. The method comprises the following steps: establishing a potential safety hazard standard library for potential safety hazard classification and level-to-level management; constructing an intelligent detection and recognition model according to the potential safety hazard standard library, arranging process image data in a construction stage, performing data labeling, and training the intelligent recognition model by using the process image data; installing the detection equipment, and recording the installation and control area of the detection equipment; configuring an intelligent detection strategy according to classification and level-to-level management requirements of the potential safety hazard standard library; acquiring a video stream from the detection equipment, calling the intelligent identification model for detection to obtain potential safety hazards, automatically recording detection results, and storing the detection results in a database according to categories; and monitoring the potential safety hazard processing result according to the potential safety hazard standard library. The system comprises a model training module, a strategy execution module and an intelligent detection module. According to the invention, the major goals of improving the safety level and reducing accidents are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of information data processing, and in particular to a construction safety intelligent management and control method and system based on a potential safety hazard library. Background Art

[0002] In engineering management, safety management is the basis for the smooth implementation of the project and the premise and guarantee for obtaining benefits. Therefore, safety management is the top priority of engineering management. During the construction process, safety management personnel of each unit need to conduct safety inspections in accordance with regulations. Safety accidents, inadequate safety precautions, and safety hazards found by each unit of the project need to be recorded and tracked in detail, and safety logs should be filled in according to the records of safety inspections, so that safety management work is recorded and traceable to ensure that safety management goals are achieved. In order to meet the principles and goals of engineering safety management and achieve the requirements of timely prevention, dynamic control, and comprehensive standardized control of production safety, safety management personnel of each unit need to conduct safety inspections and records frequently. The entire process depends on people to complete. The existing system can only play the role of auxiliary records. The differences in the degree of understanding and execution of standards by different participants will lead to a lack of standardization in the inspection process, and in some scenarios that are beyond the reach of human power, it will cause the problem of incomplete coverage of safety control. After the inspection finds safety hazards, a large amount of human resources are also required to drive rectification. The lack of standardized processing procedures makes it difficult to achieve lean, fast and flat on-site safety management. Frequent inspections and rectification supervision require a lot of manpower and material resources to complete, which inevitably causes huge resource expenditure and is not conducive to project cost control.

[0003] Prior art 1, application number: CN202410248759.5 discloses a substation construction safety control system and method based on binocular visual ranging. The system includes: a safety control object determination subsystem, which is used to determine the safety control object at the substation construction site; a target hazard source determination subsystem, which is used to determine the target hazard source at the substation construction site; a real-time distance detection subsystem, which is used to calibrate the binocular camera, and obtain the real-time distance between the safety control object and the target hazard source through the binocular camera; a control subsystem, which is used to determine the warning information for control based on the real-time distance and the safety distance against the warning template. Although the binocular camera is calibrated, the real-time distance between the safety control object and the target hazard source in the substation construction site is obtained through the binocular camera, and then the safety distance is introduced to compare the warning template with the real-time distance to determine the warning information for control, which improves the timeliness and comprehensiveness of the control. However, the detection accuracy and efficiency of the control are low.

[0004] Prior art 2, application number: CN 202410088354.X discloses a safety risk control method and system based on knowledge graph, including the following steps: collecting power grid infrastructure construction safety data involved in infrastructure accident cases; indexing the obtained power grid infrastructure construction safety data, and fusing different indexes of text, sound, picture and video data through similarity calculation; extracting entities from the fused data to obtain entity words included in the above multimodal fusion data, and the entity relationship between the above entity words; based on the above entity relationship and the above multimodal fusion data, constructing a knowledge graph of power grid infrastructure construction safety control with multimodal characteristics; based on the obtained knowledge graph, constructing risk identification indicators, although the power grid infrastructure construction risk identification is carried out, and safety risk control is implemented. However, it is impossible to automatically classify the image, resulting in poor accuracy of the control result.

[0005] Prior art three, application number: CN 202410315275.8 discloses an active power grid infrastructure safety management and control system based on multi-factor algorithm analysis, which includes: factor data acquisition module, weight acquisition module, comprehensive situation acquisition module, and early warning management module. By acquiring the construction safety monitoring data of the power grid infrastructure site and preprocessing the collected construction safety monitoring data, the infrastructure safety monitoring factor data is obtained; the historical construction safety monitoring data of the power grid infrastructure project is obtained, and the weight corresponding to the infrastructure safety monitoring factor data is analyzed; the infrastructure safety monitoring factor data and its corresponding weight are input into the multi-factor algorithm model to obtain a comprehensive infrastructure safety monitoring index; an early warning is issued according to the comprehensive infrastructure safety monitoring index, thereby improving the comprehensiveness of safety supervision of the infrastructure site, although it solves the problem of incomplete safety supervision of the infrastructure site in the prior art. However, its intelligence level is low, resulting in poor power grid detection effect.

[0006] At present, the existing technologies 1, 2 and 3 have the problem that the intelligent level of construction safety control measures is low, resulting in poor control accuracy and efficiency. Therefore, the present invention provides a construction safety intelligent control method and system based on a potential safety hazard library. Summary of the invention

[0007] In order to solve the above technical problems, the present invention provides a construction safety intelligent management and control method based on a potential safety hazard database, comprising the following steps:

[0008] Extract hidden danger classifications and items from safety production standardization specifications, and establish a standard library of hidden dangers for classification and hierarchical management of hidden dangers; build an intelligent detection and recognition model based on the standard library of hidden dangers, organize process image data during the construction phase, perform data annotation, and use process image data to train the intelligent recognition model;

[0009] Install testing equipment in the area to be tested on the construction site, and record the installation and control areas of the testing equipment; configure intelligent testing strategies according to the classification and hierarchical management requirements of the safety hazard standard library;

[0010] Obtain video streams from detection equipment and call intelligent recognition models for detection to obtain safety hazards. Automatically record detection results and store them in the database by category. Supervise safety hazard handling results based on the safety hazard standard library.

[0011] Optionally, the process of establishing a standard library of security risks includes the following steps:

[0012] Confirm the information types contained in the safety hazard standard library, establish a regulatory document database, and mark the document source and scope of application;

[0013] Mark key paragraphs of the regulatory document database; establish a classification system based on information types; design the structure of the safety hazard standard library, determine the primary and foreign key relationships, realize the association between data, and establish indexes;

[0014] Add the user interface of the safety hazard standard library, including the main menu, query interface and information entry interface.

[0015] Optionally, the information type includes hidden danger project code, hidden danger project name, hidden danger point, basis, hidden danger project classification attribute and hidden danger level.

[0016] Optionally, the process of building an intelligent recognition model includes the following steps:

[0017] Collect process image data at different construction stages under different conditions, perform image scaling operations on all process image data to ensure that each frame of the image has the same resolution and size, and use a consistent cropping strategy to highlight key construction areas; perform image enhancement operations at the same time; annotate the process impact data after the enhancement operation, distinguish the types and characteristics of safety, and obtain the pre-processed process impact data set;

[0018] The preprocessed process impact data set is divided into a basic data set, a specific data set and a validation set; the basic data set is used to conduct preliminary training on the intelligent recognition model, the specific data set is used to fine-tune the intelligent recognition model, and the validation set is used to evaluate the performance of the intelligent recognition model and record the accuracy, precision and recall rate indicators of the intelligent recognition model in identifying different types of hidden dangers;

[0019] A classifier including feature extraction layer, context information module, feature fusion layer and classification layer is used as the framework of the intelligent recognition model to identify safety hazards. The basic data set machine is used to conduct preliminary training of the intelligent recognition model, and then a specific data set is used to fine-tune the intelligent recognition model. The verification set is used to evaluate the performance of the intelligent recognition model.

[0020] Optionally, the training process of the intelligent recognition model includes the following steps:

[0021] Obtain the basic data set, start the randomly disrupted data loading strategy, and input the basic data set after the data loading strategy into the intelligent recognition model for preliminary training;

[0022] Obtain a specific data set, fine-tune the initially trained intelligent recognition model, and adopt an adaptive small batch size strategy, using 1 / 3 of the specific data set in the early stage of fine-tuning, and increasing the batch size to 2 / 3 of the specific data set in the later stage of training;

[0023] The fine-tuned intelligent recognition model is evaluated using a validation set, and a variety of evaluation indicators are used to comprehensively analyze the performance of the intelligent recognition model.

[0024] Optional, intelligent detection strategy, including area division, frame extraction frequency and real-time detection requirements.

[0025] The optional, intelligent detection strategy process includes the following steps:

[0026] Acquire image data through a detection device, the image data being a static photo or a video stream; preprocess the image data to obtain preprocessed image data;

[0027] Through edge detection, key information is extracted from the preprocessed image data, the image data set consisting of the key information is marked, and the model of the intelligent detection strategy is trained to obtain a trained model for testing and inference;

[0028] When a new image is input, the previously learned knowledge is used to identify the objects or patterns in the image and output a category label and location information.

[0029] Optionally, the process of obtaining a trained model includes the following steps:

[0030] Use edge detection algorithm to process the preprocessed image, calculate the gradient strength and direction of the image, use non-maximum suppression to remove non-edge points, and perform double threshold processing to determine the edge; identify and extract specific patterns through image segmentation, use morphological operations to analyze the structural characteristics of the image, use color histogram to analyze the color distribution in the image, and extract information in the color space;

[0031] The extracted features are labeled, a labeled data set is created, and the training set divided by the labeled data is input into the model for training. Multiple epochs of training are performed, and each epoch will be run on the training set.

[0032] The performance is evaluated on the validation set, and the model performance is evaluated on the test set, and the accuracy, recall and F1-score indicators are calculated.

[0033] Optionally, the process of automatically recording the test results and storing them in a database by category includes the following steps:

[0034] Start the detection equipment, obtain the monitoring video stream of the construction site in real time, and ensure the latest dynamics are captured through high-frequency frame acquisition; according to the intelligent detection strategy, extract key frames from the video stream at the set frame extraction frequency; perform basic image preprocessing on the extracted frames;

[0035] In the video stream processing, the previously trained intelligent recognition model is loaded, and the image data of the input video stream is received for analysis; the pre-processed image frames are input into the intelligent recognition model for real-time detection; the intelligent recognition model identifies the potential safety hazards;

[0036] Automatically record the identified safety hazard information and store the detection results in the database by category; check whether the detected hazards meet the corresponding safety specifications and requirements through the safety hazard standard library; summarize the detected hazards and trigger automatic alarms or remind relevant management personnel according to the severity and urgency.

[0037] The present invention provides a construction safety intelligent management and control system based on a potential safety hazard database, comprising:

[0038] The model training module is responsible for extracting hidden danger classifications and items from the safety production standardization specifications, and establishing a standard library of hidden dangers for safety hazard classification and hierarchical management; building an intelligent detection and recognition model based on the standard library of hidden dangers, and organizing the process image data of the construction stage, performing data annotation, and using the process image data to train the intelligent recognition model;

[0039] The strategy execution module is responsible for installing detection equipment in the area to be detected on the construction site and recording the installation and control areas of the detection equipment; configuring intelligent detection strategies according to the classification and hierarchical management requirements of the safety hazard standard library;

[0040] The intelligent detection module is responsible for obtaining the video stream from the detection equipment and calling the intelligent recognition model for detection, automatically recording the detection results and storing them in the database by category; and supervising the safety hazard handling results according to the safety hazard standard library.

[0041] The present invention establishes a standard library of safety hazards and an intelligent recognition model, constructs a standard library of safety hazards, clarifies the classification and hierarchical management of hazards, and provides standardized assessment principles and management requirements; completes the construction and training of the intelligent detection and recognition model, and can identify safety hazards in the construction stage. Significance: Through standardized classification and management of hidden dangers, it is helpful to improve the safety management capabilities of the construction site, clarify responsibilities and processing procedures; the trained intelligent recognition model can effectively identify various types of safety hazards, improve the accuracy and efficiency of detection, and provide data support for subsequent monitoring and decision-making. Deploy detection equipment and configure intelligent detection strategies, reasonably install detection equipment in the area to be detected, ensure coverage of key areas of the construction site, and improve the comprehensiveness of monitoring; configure flexible detection strategies according to the hidden danger standard library, including regional division, frame extraction frequency and real-time detection settings. Significance: By scientifically arranging detection equipment, the monitoring coverage rate is maximized to ensure that potential safety hazards can be discovered and handled in a timely manner; according to the characteristics of the actual construction site, the detection strategy is flexibly configured to adapt to different detection needs and environmental changes, and the adaptability of the system is improved. Implement intelligent detection and supervision of processing results. By acquiring video streams in real time and calling intelligent recognition models, safety hazards can be detected in a timely manner and the results can be automatically recorded. According to the safety hazard standard library, the results of hazard treatment after detection are supervised to ensure that rectification is implemented. Significance: Real-time monitoring and timely recording of safety hazards can reduce potential accident risks and improve the overall safety level of the construction site; supervise the handling of hazards to ensure that rectification measures are effectively implemented, promote the traceability and effectiveness of safety hazard management, and achieve lean safety management.

[0042] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description and the accompanying drawings.

[0043] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0045] Figure 1 This is a flow chart of a construction safety intelligent management and control method based on a potential safety hazard database in Example 1 of the present invention;

[0046] Figure 2 A process diagram for establishing a standard library of potential safety hazards in Example 2 of the present invention;

[0047] Figure 3This is a diagram of the construction process of the intelligent recognition model in Example 3 of the present invention;

[0048] Figure 4 is a process diagram of an image enhancement operation in Embodiment 4 of the present invention;

[0049] Figure 5 A process diagram showing how the process after preprocessing affects the partitioning of a data set in Example 5 of the present invention;

[0050] Figure 6 This is a diagram of the training process of the intelligent recognition model in Example 6 of the present invention;

[0051] Figure 7 This is a process diagram of the intelligent detection strategy in Example 7 of the present invention;

[0052] Figure 8 This is a process diagram of obtaining a trained model in Example 8 of the present invention;

[0053] Fig. 9 A process diagram of automatically recording test results and storing them in a database by category in Example 9 of the present invention;

[0054] Fig.10 A process diagram of storing the detection results in a database by category in Embodiment 10 of the present invention;

[0055] Fig.11 A process diagram of supervising the safety hazard processing results according to the safety hazard standard library in Example 11 of the present invention;

[0056] Fig.12 This is a block diagram of the construction safety intelligent management and control system based on the safety hazard database in Example 12 of the present invention. DETAILED DESCRIPTION

[0057] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0058] The terms used in the embodiments of the present application are only for the purpose of describing specific embodiments, and are not intended to limit the embodiments of the present application. The singular forms of "a", "said" and "the" used in the embodiments of the present application are also intended to include plural forms, unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used in this article refers to and includes any or all possible combinations of one or more associated listed items.

[0059] When the following description refers to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present application as described in detail. In the description of the present application, it should be understood that the terms "first", "second", "third", etc. are only used to distinguish similar objects, and do not have to be used to describe a specific order or sequence, nor can they be understood as indicating or implying relative importance. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood according to the specific circumstances.

[0060] Example 1: Figure 1 As shown, an embodiment of the present invention provides a construction safety intelligent management and control method based on a potential safety hazard library, comprising the following steps:

[0061] S100: Extract hidden danger classifications and items from safety production standardization specifications, and establish a standard library of hidden dangers for safety hazard classification and hierarchical management; build an intelligent detection and recognition model based on the standard library of hidden dangers, organize process image data in the construction stage, perform data annotation, and use process image data to train the intelligent recognition model;

[0062] S200: Install detection equipment in the area to be detected on the construction site, and record the installation and control areas of the detection equipment; configure intelligent detection strategies according to the classification and hierarchical management requirements of the safety hazard standard library, including area division, frame extraction frequency, and real-time detection requirements;

[0063] S300: Obtain the video stream from the detection device and call the intelligent recognition model for detection, obtain the existence of safety hazards, automatically record the detection results and store them in the database by category; supervise the safety hazard processing results according to the safety hazard standard library.

[0064] The working principle and beneficial effects of the above technical solution are as follows: First, this embodiment extracts hidden danger classification and items from the safety production standardization specification, establishes a standard library of hidden dangers for safety hazard classification and hierarchical management; constructs an intelligent detection and recognition model based on the standard library of hidden dangers, organizes the process image data of the construction stage, performs data annotation, and uses the process image data to train the intelligent recognition model; secondly, installs detection equipment in the area to be detected on the construction site, and records the installation and control area of ​​the detection equipment; configures the intelligent detection strategy according to the classification and hierarchical management requirements of the standard library of hidden dangers, including regional division, frame extraction frequency and real-time detection requirements; finally, obtains the video stream from the detection equipment and calls the intelligent recognition model for detection, obtains the existence of hidden dangers, automatically records the detection results and stores them in the database by category; supervises the processing results of hidden dangers according to the standard library of hidden dangers. Step S100 of the above scheme establishes a standard library of hidden dangers and an intelligent recognition model, constructs a standard library of hidden dangers, clarifies the classification and hierarchical management of hidden dangers, and provides standardized assessment principles and management requirements; completes the construction and training of the intelligent detection and recognition model, and can identify hidden dangers in the construction stage. Significance: Through standardized hazard classification and management, it is helpful to improve the safety management capabilities of the construction site, clarify responsibilities and processing procedures; the trained intelligent recognition model can effectively identify various safety hazards, improve the accuracy and efficiency of detection, and provide data support for subsequent monitoring and decision-making. Step S200 deploys detection equipment and configures intelligent detection strategies, reasonably installs detection equipment in the area to be detected, ensures that the key areas of the construction site are covered, and improves the comprehensiveness of monitoring; configures flexible detection strategies according to the hidden danger standard library, including area division, frame extraction frequency and real-time detection settings. Significance: Through the scientific arrangement of detection equipment, the monitoring coverage rate is maximized to ensure that potential safety hazards can be discovered and handled in a timely manner; according to the characteristics of the actual construction site, the detection strategy is flexibly configured to adapt to different detection needs and environmental changes, and the adaptability of the system is improved. Step S300 implements intelligent detection and supervision processing results. By acquiring video streams in real time and calling intelligent recognition models, safety hazards can be detected in time and the results can be automatically recorded; according to the safety hazard standard library, the hidden danger handling results after detection are supervised to ensure rectification and implementation. Significance: Real-time monitoring and timely recording of safety hazards can reduce potential accident risks and improve the overall safety level of the construction site; supervise the handling of hidden dangers, ensure that corrective measures are effectively implemented, promote the traceability and effectiveness of safety hazard management, and achieve lean safety management.

[0065] This embodiment solves the problem of lack of standardization of on-site safety inspections by establishing a standard library of safety hazards. It replaces manual inspections with automatic camera shooting and artificial intelligence detection, improves coverage and coverage period, reduces human resource investment, and realizes lean, fast and flat processing of safety hazards on construction sites, achieving the effect of reducing costs and increasing efficiency. It plays a positive role in promoting the practice of enterprise safety production standardization, guides and regulates the majority of enterprises to independently promote the construction of safety production standardization, strengthens the foundation of enterprise safety production, guides enterprises to develop scientifically and safely, and realizes the organic unity of enterprise production quality, efficiency and safety, which can produce extensive and practical social and economic benefits.

[0066] In summary, this embodiment, through the implementation of these three steps, from the construction of the hidden danger standard library, the configuration of the intelligent detection equipment, to the supervision and management of the detection results, constitutes a complete intelligent construction safety management and control system. It not only improves the efficiency and accuracy of construction site safety management, but also promotes the standardization of the construction process, and ultimately achieves the major goals of improving safety levels and reducing accidents.

[0067] Example 2: Figure 2 As shown, based on Example 1, the process of establishing a standard library of potential safety hazards provided by the embodiment of the present invention includes the following steps:

[0068] S101: Confirm the information types contained in the safety hazard standard library, establish a regulatory document database, and mark the file source and scope of application; the information types include hazard project code, hazard project name, hazard point, basis, hazard project classification attributes and hazard level, etc.; among them, the hazard project code is the safety hazard project classification code, the hazard project name is the name of the safety hazard project, the hazard point is the description of the safety hazard, the basis is the relevant laws and regulations on the construction safety inspection standards of the relevant countries, industries and enterprises, the hazard project classification attributes are industry, project type, inspection sub-item and hazard item, and the hazard level is the level of the hazard item;

[0069] S102: Mark the key paragraphs of the regulatory document database; establish a classification system based on information types; design the structure of the safety hazard standard library, including the hazard table, the law and regulation table, and the modification record table, etc., determine the primary and foreign key relationships, realize the association between data, and establish indexes;

[0070] S103: Add the user interface of the safety hazard standard library, including the main menu, query interface and information entry interface; verify the entered data; among them, the information entry module allows users to add new hazard items, supports batch uploading and editing of existing data; the query and retrieval module maps the hazard item fields, and performs rapid retrieval based on keywords, categories and levels; the report generation module provides data export function to generate hazard analysis reports and statistical reports.

[0071] The working principle and beneficial effects of the above technical solution are as follows: This embodiment first confirms the information type contained in the safety hazard standard library, establishes a legal document database, and marks the file source and applicable scope; the information type includes the hidden danger project code, hidden danger project name, hidden danger point, basis, hidden danger project classification attribute and hidden danger level, etc.; among them, the hidden danger project code is the safety hazard project classification code, the hidden danger project name is the name of the safety hazard project, the hidden danger point is the description of the safety hazard, the basis is the relevant legal and regulatory documents on the construction safety inspection standards of the relevant countries, industries and enterprises, the hidden danger project classification attributes are industry, project type, inspection sub-item and hidden danger item, and the hidden danger level is the level of the hidden danger item; secondly, the regulations are The key paragraphs of the document database are marked; a classification system is established through information types; the structure of the safety hazard standard library is designed, including a hazard table, a legal and regulatory table, and a modification record table, etc., the primary and foreign key relationships are determined, the association between data is realized, and an index is established; finally, the user interface of the safety hazard standard library is added, including the main menu, the query interface, and the information entry interface; the entered data is verified; among them, the information entry module allows users to add new hazard items, supports batch uploading and editing of existing data; the query and retrieval module maps the hazard item fields, and performs rapid retrieval based on keywords, categories, and levels; the report generation module provides data export functions to generate hazard analysis reports and statistical reports. Step S101 of the above scheme confirms the information type and establishes a legal document database, clarifies the key information types that the safety hazard standard library needs to include, and ensures that the content structure of the library is reasonable; by establishing a legal document database, relevant legal and regulatory documents can be easily managed and referenced, providing a basis for subsequent data processing and user queries. Significance: Ensuring the comprehensiveness and accuracy of information helps to form an authoritative standard library that users can rely on to make safety decisions; it is convenient for users to obtain relevant standards when conducting hazard assessments, thereby ensuring that safety production complies with national and industry regulations. Step S102: Mark key paragraphs and establish a classification system. By marking important paragraphs in the regulatory document database, key information can be quickly located to improve the availability of information. Design the structure of the safety hazard standard library, including hazard tables, legal and regulatory tables, and modification record tables, etc., to ensure that the relationship between data is clear, and establish indexes to improve query efficiency. Significance: Improve the efficiency of the use of regulatory information, enable relevant personnel to quickly find the required information, and support the hazard identification and management process; the clear data structure promotes subsequent data entry, retrieval and analysis, and lays the foundation for the systematic development of the standard library. Step S103: User interface design and data entry processing. Designing a user-friendly interface (including the main menu, query interface, and information entry interface) simplifies system operation and makes it easier for users to access and use the database; provide an information entry verification mechanism to ensure the authenticity and consistency of the data, while supporting batch uploading of new hazard items, enhancing the flexibility of the system.Significance: It enhances user experience, improves user work efficiency, and facilitates safety managers to quickly find and enter hidden danger information in daily work; the generated report generation function makes hidden danger analysis and statistics efficient and standardized, provides important basis for decision-making, and helps to plan and implement safety production measures;.

[0072] In summary, each step of establishing the safety hazard standard library in this embodiment achieves specific goals in terms of its technical effect and forms a complementary relationship in practical sense, so that the standard library effectively serves the needs of safety hazard management, making the entire construction process not only technically feasible, but also an important guarantee for safety management. Through systematic and standardized information management, the safety awareness and risk prevention and control capabilities of the construction industry can be effectively improved, and the goal of reducing accidents and ensuring the safety of life and property can be ultimately achieved.

[0073] Examples of standard libraries of security risks:

[0074]

[0075]

[0076] Example 3: Figure 3 As shown, based on Example 1, the construction process of the intelligent recognition model provided by the embodiment of the present invention includes the following steps:

[0077] S104: Collect process image data at different construction stages under different conditions, perform image scaling operations on all process image data to ensure that each frame of image has the same resolution and size, and use a consistent cropping strategy to highlight key construction areas; perform image enhancement operations at the same time; annotate the process impact data after the enhancement operation, distinguish the types and features of safety, and obtain the pre-processed process impact data set;

[0078] S105: Divide the preprocessed process impact data set into a basic data set, a specific data set and a verification set; the basic data set performs preliminary training on the intelligent recognition model, the specific data set performs fine-tuning on the intelligent recognition model, and the verification set evaluates the performance of the intelligent recognition model, and records the accuracy, precision and recall rate of the intelligent recognition model in identifying different types of hidden dangers;

[0079] S106: A classifier including a feature extraction layer, a context information module, a feature fusion layer, and a classification layer is used as the framework of an intelligent recognition model to identify potential safety hazards. A basic data set is used to perform preliminary training on the intelligent recognition model, and then a specific data set is used to fine-tune the intelligent recognition model. The validation set is used to evaluate the performance of the intelligent recognition model.

[0080] Among them, the framework of the classifier is as follows:

[0081] Input layer: The input data accepts images or video frames as input, which can be real-time monitoring images of the construction site; basic image preprocessing (such as normalization, resizing, etc.) is performed at the input layer to ensure that the input data is suitable for subsequent processing.

[0082] Feature extraction layer

[0083] Convolutional layer (CNN): Multiple convolutional layers are stacked to extract image features. Standard convolution operations, batch normalization, and activation functions (such as ReLU) can be used to enhance model performance;

[0084] Attention mechanism: Place the SE module or CBAM module in the feature extraction channel and introduce it between the convolution layer and the pooling layer. This module will weight the convolution feature map, automatically focus on important feature areas, and help the model improve its sensitivity to specific hidden dangers (such as small objects or details).

[0085] Context Information Module

[0086] Region Proposal Network (RPN): Add RPN after the convolutional layer to propose regions for possible objects. RPN generates proposal boxes based on convolutional features to quickly focus on possible hidden danger areas;

[0087] Panoptic segmentation layer: After RPN, a panoptic segmentation module can be introduced to use masks to distinguish the detected areas (such as different safety hazards) from the background in detail, thereby improving the fine-grained classification capability.

[0088] Feature fusion layer

[0089] Feature Pyramid Network (FPN): Use FPN to perform multi-scale feature fusion on convolutional features. FPN connects features between different layers to ensure that features of multiple scales can be transmitted to the classification layer together, thereby improving the recognition rate of hidden dangers of different sizes.

[0090] Classification layer

[0091] Fully connected layer: After flattening the feature map output by FPN, it is transformed nonlinearly through multiple fully connected layers to generate the final output feature vector;

[0092] Softmax layer: Use the Softmax layer to classify the output features and output the classification probabilities of different safety hazards;

[0093] The output layer outputs the probability of each category, including the detection confidence of each potential safety hazard. A threshold mechanism can be adopted to determine whether a potential safety hazard has occurred.

[0094] The working principle and beneficial effects of the above technical solution are as follows: This embodiment first collects process image data of different construction stages under different conditions, performs image scaling operations on all process image data, ensures that each frame of image reaches the same resolution and size, and uses a consistent cropping strategy to highlight key construction areas; performs image enhancement operations at the same time; annotates the process impact data after the enhancement operation, distinguishes the types and characteristics of safety, and obtains a preprocessed process impact data set; secondly, the preprocessed process impact data set is divided into a basic data set, a specific data set and a verification set; the basic data set performs preliminary training on the intelligent recognition model, the specific data set performs fine adjustment on the intelligent recognition model, the verification set evaluates the performance of the intelligent recognition model, and records the accuracy, precision and recall rate of the intelligent recognition model in identifying different categories of hidden dangers; finally, a classifier including a feature extraction layer, a context information module, a feature fusion layer and a classification layer is used as the framework of the intelligent recognition model to identify safety hazards; the basic data set machine is used to perform preliminary training on the intelligent recognition model, and then the specific data set is used to perform fine adjustment on the intelligent recognition model, and the verification set evaluates the performance of the intelligent recognition model. Step S104 of the above scheme is data collection and preprocessing, which collects diversified data: a large amount of process image data is collected under different conditions (different lighting, weather, construction stages, etc.), thereby improving the generalization ability of the model; by scaling all images, each frame of the image is ensured to have a consistent resolution and size, so that the data input is processed at the same scale; a consistent cropping strategy helps to highlight key construction areas, remove unnecessary background information, and ensure that the model focuses on important features; through various image enhancement operations (such as rotation, flipping, brightness adjustment, noise addition, etc.), the diversity of data can be increased and the robustness of the model in practical applications can be improved; labeling the enhanced image data and distinguishing different types and characteristics of safety hazards is the basic significance of establishing an effective supervised learning model: ensuring that the model can learn a variety of construction scenarios during training, and improving the model's ability to identify different construction conditions and safety hazards; providing high-quality, clearly labeled data sets for subsequent training and verification, which is a key factor for successful model training. Step S105: Dataset division and model training. The preprocessed dataset is divided into a basic dataset, a specific dataset, and a validation set, so that different data sources can be used for step-by-step training and validation. The basic dataset is used for preliminary training to enable the model to obtain a preliminary learning effect. The specific dataset is used for fine-tuning to further optimize the accuracy of the model on specific types of hidden dangers. The validation set is used to evaluate the accuracy, precision, and recall of the model in identifying different types of hidden dangers, so as to quantify the performance of the model. Significance: Through structured dataset division, it is ensured that the model is subject to hierarchical training and evaluation strategies to improve the generalizability and accuracy of the model. Through various evaluation indicators, the performance of the model in different scenarios can be clearly understood, guiding the subsequent model adjustment and optimization.Step S106: Model framework selection and training. A classifier including a feature extraction layer, a context information module, a feature fusion layer, and a classification layer is used to effectively extract image features and identify safety hazards. The basic data set is used for preliminary training to enable the model to understand general features. The specific data set is fine-tuned to optimize the model's performance in detecting specific hazards. The validation set is used to conduct a comprehensive evaluation of the model, monitor model performance, and tune the model. Significance: By using a structured model architecture, it is ensured that the model can fully utilize image features and pay attention to context information, thereby improving recognition accuracy. The step-by-step training and tuning method enhances the model's specific recognition ability for construction safety hazards and improves the reliability and effectiveness of the model in practical applications.

[0095] In summary, the series of steps in this embodiment form a perfect system of cyclic iteration. Based on high-quality data sets, through step-by-step training, adjustment and verification, the intelligent recognition model can quickly and efficiently identify safety hazards in construction. It not only helps to improve the safety of the construction site, but also reduces unnecessary accidents and losses in project management, and ultimately provides strong technical support for improving construction efficiency and safety assurance.

[0096] Example 4: Figure 4 As shown, based on Example 3, the process of image enhancement operation provided by the embodiment of the present invention includes the following steps:

[0097] S1041: Select an angle of -25° to +25° to rotate the image of the key construction area; crop a sub-area from the image of the original process image data, and the width and height of the cropped area are between 10% and 90% of the original image to highlight the key area;

[0098] S1042: Increase or decrease the brightness value of each pixel by a random value ranging from -30 to +30; select a contrast factor between 1.0 and 2.0 to enhance the contrast of the image by changing the difference between pixels in the image; apply Gaussian blur to the image, using a blur kernel of size 5x5;

[0099] S1043: Adjust the hue, saturation and brightness of the image. Each type of adjustment is adjusted within the range of + / -10. Select horizontal flip or vertical flip. The flip ratio is set to 50%.

[0100] The working principle and beneficial effects of the above technical solution are as follows: the embodiment first selects an angle of -25° to +25° to rotate the image of the key construction area; cuts out a sub-area from the image of the original process image data, and the width and height of the cut area are between 10% and 90% of the original image to highlight the key area; secondly, the brightness value of each pixel is increased or decreased by a random value in the range of -30 to +30; the contrast coefficient is selected between 1.0 and 2.0 to enhance the contrast of the image by changing the difference between pixels in the image; Gaussian blur is applied to the image, using a blur kernel of size 5x5; finally, the hue, saturation and brightness of the image are adjusted, and each type of adjustment is adjusted within the range of + / -10, and horizontal flip or vertical flip is selected, and the flip ratio is set to 50%. Step S1041 of the above solution rotates and cuts, and the image of the key construction area is rotated at the selected angle (-25° to +25°). This step enables the model to learn image features at different angles and increase the diversity of samples; cropping a sub-region with a width and height between 10% and 90% of the original image from the original image helps to highlight the key areas of the image and remove unnecessary background interference. Significance achieved: Through rotation and cropping, the generated image retains important construction information and provides samples from different perspectives. Diversity helps to improve the generalization ability of the model, so that it can more effectively identify safety hazards and construction features in practical applications. Step S1042 Brightness, contrast adjustment and blur processing, randomly increase or decrease the brightness value of each pixel (ranging from -30 to +30), can simulate the image effect under different lighting conditions; by setting the contrast coefficient between 1.0 and 2.0, adjust the difference between pixels in the image to make the image more visually distinct; apply Gaussian blur, use a blur kernel of size 5x5, reduce the clarity of image details, and simulate the effect under bad weather or long-distance shooting conditions. Significance achieved: The adjustment of brightness and contrast enables the image to maintain appropriate recognizability under different lighting and visual conditions, improving the model's adaptability to actual working conditions; the introduction of Gaussian blur improves the model's robustness to images in a blurred state, enabling it to process low-quality or blurred images in the real world. Step S1043 Color adjustment and flipping: adjust the hue, saturation, and brightness of the image within a range of + / -10 to make the image's color performance more diverse, enhance details and visual effects; randomly select horizontal flipping or vertical flipping, and the probability of flipping is set to 50%. This operation changes the directionality of the image, allowing the model to recognize entries in different directions.Significance achieved: Color adjustment brings richer visual expression, allowing the model to adapt to color changes in different environments and enhance the comprehensive understanding of construction images; the flipping operation increases the diversity of the data set, further improving the model's adaptability to various directions, ensuring that it can still effectively identify safety hazards in real environments.

[0101] In summary, through the implementation of each of the above steps, the enhancement operation in this embodiment not only enriches the training data set, improves the diversity and complexity, but also enhances the robustness of the model in the face of images under different conditions, different states and different perspectives. The performance of the intelligent recognition model is improved, making it perform better in the actual construction safety hazard identification task.

[0102] Example 5: Figure 5 As shown, based on Example 3, the process after preprocessing provided by the embodiment of the present invention affects the process of dividing the data set, including the following steps:

[0103] S1051: Identify all potential hazard types including minor hazards, major potential hazards and unidentified potential hazards from the preprocessed process impact data set; list all potential hazard types, create a sample list, record the number of samples for each type of potential hazard, and obtain the basic data set;

[0104] For example, for each potential hazard type, perform the following operations:

[0105] Slightly dangerous: randomly select at least 50 images;

[0106] Major hidden dangers: randomly select at least 50 images;

[0107] Unidentified hidden dangers: randomly select at least 50 images;

[0108] If there are less than 50 samples of a certain type of hidden danger, select samples from other hidden danger types to supplement them, ensuring that the total amount of the basic data set is no less than 240 images;

[0109] S1052: Determine the number of samples for each high-risk hidden danger type, set the total sample size to at least 100, and obtain a specific data set;

[0110] For example, the specific selection criteria are as follows:

[0111] Fall risk hazard: Make sure to include 50 images;

[0112] Mechanical injury risk hazards: ensure that 30 images are included;

[0113] High temperature work hazards: Collect 20 images as much as possible;

[0114] By analyzing historical data, we identify the data sources of these hidden danger types and systematically extract target samples to ensure that the total number is at least 100;

[0115] S1053: dividing the data of the remaining process impact data set according to the construction stages of the construction preparation stage, the construction execution stage, and the construction completion feedback stage, and selecting at least 20 images from each stage to obtain a verification set;

[0116] For example:

[0117] Construction preparation stage: randomly select at least 20 images;

[0118] Construction execution phase: randomly select at least 20 images;

[0119] End-of-construction feedback phase: randomly select at least 20 images;

[0120] Combining the samples selected in each stage, the validation set is ensured to contain at least 60 images in total.

[0121] The working principle and beneficial effects of the above technical solution are as follows: This embodiment first identifies all types of hidden dangers including minor dangers, major hidden dangers and unidentified hidden dangers from the preprocessed process impact data set; lists all hidden danger types, creates a sample list, records the number of samples of each type of hidden danger, and obtains a basic data set; secondly, determines the number of samples for each high-risk hidden danger type, sets the total sample size to at least 100, and obtains a specific data set; finally, divides the data of the remaining process impact data set according to the construction stages of the construction preparation stage, the construction execution stage, and the construction completion feedback stage, and selects at least 20 images from each stage to obtain a verification set. The construction of the basic data set in step S105 of the above scheme, through a comprehensive analysis of the preprocessed data set, identifies and classifies all relevant hidden danger types (minor dangers, major hidden dangers, unidentified hidden dangers, etc.), ensures that the model can be exposed to different categories of safety hazards; creates a detailed sample list, accurately records the number of samples of each type of hidden danger, and provides data support for subsequent analysis and data set construction. Significance achieved: Ensure that each type of hidden danger has a sufficient sample size, so that the trained model has a good recognition ability for different hidden danger types and avoids overfitting a specific type; by covering various types of hidden dangers, improve the generalization ability of the model, so that it performs more reliably when processing unseen data. Step S1052 Construction of a specific data set, focus resources and attention on high-risk hidden dangers, ensure that the most critical hidden danger types in the analysis (fall risk, mechanical injury risk and high temperature operation hidden dangers) are fully represented; set a clear sample quantity standard for each high-risk hidden danger to ensure that enough high-risk samples are obtained in the data set. Significance achieved: Through the construction of a specific data set, model training can focus on high-risk hidden danger types, improve its recognition accuracy for these hidden dangers, and make the model more practical and efficient; increase the sensitivity of the model to important hidden dangers, and enhance its ability to quickly identify and respond to high-risk situations in practical applications. Step S1053 Construction of a validation set, divide the remaining data according to the construction stage, ensure that representative samples can be obtained in each stage, so as to test the universality of the model; randomly select images from different construction stages to ensure that the validation set can cover different situations and conditions, and improve the evaluation reliability of the model. Significance achieved: Through sample testing at different construction stages, we can comprehensively evaluate the performance of the model in actual applications, such as how it responds to different construction environments and changes; ensure that the model can maintain high detection accuracy and reliability in a variety of operating scenarios and conditions in actual operation, and provide value for safety management.

[0122] In summary, the data set partitioning scheme of this embodiment effectively builds a multi-level, feature-rich training and verification framework through targeted steps, laying a solid foundation for the development and application of intelligent recognition models. It not only ensures the quality and balance of training data, but also ensures that the model can efficiently and stably identify and respond to various safety hazards in practical applications.

[0123] Example 6: Figure 6 As shown, based on Example 3, the training process of the intelligent recognition model provided by the embodiment of the present invention includes the following steps:

[0124] S1061: Obtain a basic data set, start a randomly disrupted data loading strategy, and input the basic data set after the data loading strategy into the intelligent recognition model for preliminary training;

[0125] S1062: Obtain a specific data set, fine-tune the preliminarily trained intelligent recognition model, adopt an adaptive small batch size strategy, use a small batch (1 / 3 of the specific data set) in the early stage of fine-tuning, and increase the batch (2 / 3 of the specific data set) in the later stage of training;

[0126] S1063: Evaluate the fine-tuned intelligent recognition model using the validation set, and use a variety of evaluation indicators to comprehensively analyze the performance of the intelligent recognition model.

[0127] The working principle and beneficial effects of the above technical solution are as follows: the present embodiment first obtains the basic data set, starts the randomly shuffled data loading strategy, and inputs the basic data set after the data loading strategy into the intelligent recognition model for preliminary training; secondly, obtains the specific data set, fine-tunes the preliminary trained intelligent recognition model, and adopts an adaptive small batch size strategy, using a small batch (1 / 3 of the specific data set) in the early stage of fine-tuning, and increasing the batch (2 / 3 of the specific data set) in the later stage of training; finally, the finely adjusted intelligent recognition model is evaluated using the validation set, and the performance of the intelligent recognition model is comprehensively analyzed using a variety of evaluation indicators. Step S1061 of the above solution obtains the basic data set and starts the randomly shuffled data loading strategy. By randomly shuffling, it ensures that the order of data input is different each time during training, thereby avoiding the model from learning a certain fixed sequence pattern. This helps the model learn sample features more comprehensively; by loading the basic data set in a diversified manner, the model can be exposed to a wider range of features and samples, making it more robust and generalizable. Significance achieved: Avoid overfitting the model to a specific input sequence, thereby enhancing the model's performance on new data; by diversifying the presentation order of training samples, the model can understand different hidden danger features in a more balanced way, thereby improving the accuracy of recognition. Step S1062 obtains a specific data set and makes fine adjustments. In the early stage of training, a smaller batch is used to help the model converge quickly and learn basic features, and in the later stage, the batch is increased to enable the model to stably learn more complex features, making the training process flexible and optimizing the effect; specific data sets usually contain high-risk hidden danger samples. Through fine adjustments, the model can pay more attention to and adapt to these key samples, thereby improving the ability to identify key hidden dangers. Significance achieved: By emphasizing the learning of high-risk hidden danger samples, the model can more efficiently identify potential dangers in practical applications, thereby improving safety; by dynamically adjusting the batch size, the convergence process of the model is optimized, the training speed and performance are improved, and better training results are achieved. Step S1063 uses the validation set for evaluation. By using multiple evaluation indicators such as accuracy, precision, and recall, the performance of the model is comprehensively analyzed to ensure a more three-dimensional understanding of the model. The fine-tuned model is tested using the validation set to ensure that the model not only performs well on the training data, but also has good adaptability to unknown data. Significance achieved: Through a multi-dimensional evaluation method, it is ensured that the model not only performs well in one aspect, but also reaches a high level in overall performance. The evaluation results of the validation set can guide the subsequent model to be further optimized and parameter adjusted, and the model is continuously iterated and improved based on the performance results to adapt to the variability and complexity of practical applications.

[0128] In summary, each step of this embodiment is crucial in the model training process. S1061 ensures the effective use and diversified input of the basic data set, S1062 improves the model's pioneering recognition ability for hidden dangers through flexible batch strategies and key adjustments of high-risk data, and S1063 guides the subsequent optimization of the model through comprehensive verification and evaluation. Overall, an efficient, robust and adaptable intelligent recognition model is constructed, which can play an important role in the field of construction safety.

[0129] Example 7: Figure 7 As shown, based on Example 1, the process of the intelligent detection strategy provided by the embodiment of the present invention includes the following steps:

[0130] S201: Acquire image data through a detection device, the detection device includes a camera, a sensor, and a scanner, etc. The image data is a static photo or a video stream; preprocess the image data, the preprocessing includes operations such as scaling, cropping, rotating, and graying, to obtain preprocessed image data;

[0131] S202: extract key information from the preprocessed image data, including important patterns, structures, and colors, through edge detection; mark the image data set consisting of the key information, and train the model of the intelligent detection strategy to obtain a trained model for testing and inference;

[0132] S203: When a new image is input, the previously learned knowledge is used to identify the object or pattern in the image, and a category label and location information are output.

[0133] The working principle and beneficial effects of the above technical solution are as follows: in this embodiment, image data is first acquired through a detection device, the detection device includes a camera, a sensor and a scanner, etc., and the image data is a static photo or a video stream; the image data is preprocessed, and the preprocessing includes operations such as scaling, cropping, rotation and grayscale to obtain preprocessed image data; secondly, through edge detection, key information in the preprocessed image data is extracted, including important patterns, structures and colors, etc.; the image data set consisting of the key information is marked, and the model of the intelligent detection strategy is trained to obtain a trained model for testing and inference; finally, when a new image is input, the previously learned knowledge is used to identify the object or pattern in the image, and a category label and location information are output. Step S201 of the above scheme obtains image data and performs preprocessing. Image data (static photos or video streams) are obtained through various detection devices (such as cameras, sensors and scanners). The diversity and richness of the data are helpful for subsequent analysis; the collected image data is scaled, cropped, rotated and grayed to ensure the consistency and adaptability of the image data format, and reduce noise and interference in subsequent processing; through appropriate preprocessing, the complexity of image data is reduced, the computational burden is reduced, and the model processing speed is improved. Significance achieved: Through preprocessing operations, the quality and consistency of image data are ensured, thereby improving the accuracy of subsequent model training and detection. High-quality input data can effectively improve the learning effect of the model, making the generated model more accurate and reliable; the prepared image data provides a solid foundation for subsequent feature extraction and model training, ensuring the effectiveness of the model. Step S202: Feature extraction and model training. Through edge detection and other technologies, key information such as important patterns, structures and colors are extracted from the preprocessed image data. Features are the key to distinguishing objects or patterns in images. The image data set composed of the extracted features is marked to provide strong support for subsequent model training. The model of the intelligent detection strategy is trained using the marked data set to obtain a trained model, and the recognition ability of the model is improved by learning key information. Significance achieved: Through feature extraction, complex image information is converted into a form that can be used for identification, which helps the model to more accurately identify the target object; a clear and accurate feature set can help the model reduce the misrecognition rate in actual detection and improve accuracy; the trained model lays a solid foundation for subsequent real-time detection and inference, making the system feasible. Step S203: Model testing and inference. When a new unseen image is input, the trained model can identify the object or pattern in the image based on the previously learned knowledge; the model can provide key outputs such as category labels and location information to convey the detection results to the user or subsequent system.Significance achieved: Through the inference ability of the model, the burden of manual inspection is reduced, process automation is achieved, and efficiency is improved; safety hazard information can be identified and output in a timely manner to ensure a quick response at the construction site and reduce potential dangers; the output category labels and location information provide an important basis for subsequent decision-making and processing, allowing management to take appropriate safety measures.

[0134] In summary, the entire intelligent detection strategy of this embodiment, from the acquisition and preprocessing of image data to the extraction of features and model training, finally realizes real-time detection and inference of images. Each step is closely linked to each other, jointly optimizing the accuracy and efficiency of detection. It not only promotes the intelligent process of construction site safety management, but also improves the overall safety and response speed to a certain extent.

[0135] Example 8: Figure 8 As shown, based on Example 7, the process of obtaining a trained model provided by the embodiment of the present invention includes the following steps:

[0136] S2021: Use edge detection algorithm to process the preprocessed image, calculate the gradient strength and direction of the image, use non-maximum suppression to remove non-edge points, and perform double threshold processing to determine the edge; identify and extract specific patterns through image segmentation, use morphological operations to analyze the structural characteristics of the image, use color histogram to analyze the color distribution in the image, and extract information in the color space;

[0137] S2022: Label the extracted features, establish a labeled data set, input the training set divided by the labeled data into the model for training, and perform training for multiple epochs, where each epoch will be run on the training set;

[0138] S2023: Perform performance evaluation on the validation set, evaluate model performance on the test set, and calculate indicators such as accuracy, recall, and F1-score.

[0139] The working principle and beneficial effects of the above technical solution are as follows: In this embodiment, the edge detection algorithm is first used to process the preprocessed image, the gradient strength and direction of the image are calculated, non-maximum suppression is used to remove non-edge points, and double threshold processing is performed to determine the edge; specific patterns are identified and extracted through image segmentation, the structural features of the image are analyzed using morphological operations, the color distribution in the image is analyzed using a color histogram, and information in the color space is extracted; secondly, the extracted features are marked, a labeled data set is established, and the training set divided by the marked data is input into the model for training, and multiple epochs of training are performed, and each epoch will be run on the training set; finally, performance evaluation is performed on the validation set, and the model performance is evaluated on the test set, and indicators such as accuracy, recall rate and F1-score are calculated. Step S2021 of the above scheme, feature extraction and pattern recognition, can effectively identify the contours and boundaries in the image by using edge detection algorithms (such as the Canny algorithm) to calculate the gradient strength and direction of the image; remove non-edge points, retain the local maximum intensity, effectively focus on the true edge part of the image, and reduce noise interference; determine distinct boundaries and filter out edges to make subsequent processing more accurate; identify important patterns in the image through specific segmentation algorithms (such as threshold segmentation and region growing), thereby decomposing the image into several regions with similar attributes; help separate the region of interest from the background, thereby improving the efficiency and accuracy of subsequent analysis; use morphological operations such as dilation and erosion to analyze the structural features of the image to clearly identify the target shape and improve the expressiveness of the features; enhance the observability of the target in the image, making it easier to detect in subsequent steps; analyze the color distribution of the image and extract effective color features, which can provide important information for object classification; through the extraction of color information, the dependence on shape or texture information in complex scenes can be reduced, providing auxiliary dimensions for classification and recognition. Significance achieved: Through comprehensive and multi-dimensional feature extraction, ensure that the data input to the model can better reflect the specific properties of the image; through effective feature extraction technology, enhance the model's sensitivity to important patterns and structures, and improve the accuracy of subsequent training and inference. Step S2022 Data labeling and model training, comprehensively label the extracted features, establish a labeled data set, and implement a high-quality supervised learning foundation to ensure that the model has sufficient guidance information during the training process; divide the data set into a training set and a validation set; ensure the independence of training and validation to avoid overfitting, thereby improving the generalization ability of the model; epoch training: input training data in multiple epochs and continuously update the model weights; use labeled training data for model training, optimize model parameters, and learn patterns and features in the data.Significance achieved: Through training, the model can identify the relationship between features and tags in the input data, so that it has the ability to predict new data; through a reasonable training and verification process, overfitting is reduced and the performance of the model in practical applications is improved. Step S2023 Performance evaluation and model verification, evaluate the performance of the model on the verification set, using indicators such as accuracy, recall, and F1-score; calculate relevant indicators to provide an intuitive understanding of the performance of the model on unknown data; use an independent test set to further test the generalization ability and stability of the model to ensure its reliability in practical applications. Significance achieved: Through systematic performance evaluation, judge the adaptability of the model to practical problems and determine whether it meets the preset indicators; through the calculated evaluation indicators, help the model development team judge the usability of the model and provide directions and suggestions for model improvement; confidence in the final product, ensure that it can perform stably in various practical application scenarios, and reduce risks in actual use.

[0140] In summary, the entire process of this embodiment, from feature extraction to model training and performance evaluation, is a systematic and organic learning and optimization process. It ensures that the model can learn useful information from the data, achieve high recognition accuracy, and perform well in real-time applications. Through the systematic processing of each step, the foundation is laid for the successful implementation of the intelligent detection system.

[0141] Example 9: Fig. 9 As shown, based on Example 1, the process of automatically recording the test results and storing them in the database by category provided by the embodiment of the present invention includes the following steps:

[0142] S301: Start the detection equipment, obtain the monitoring video stream of the construction site in real time, and ensure the latest dynamics are captured through high-frequency frame acquisition; according to the intelligent detection strategy and the set frame extraction frequency, extract key frames from the video stream; perform basic image preprocessing on the extracted frames;

[0143] S302: loading the previously trained intelligent recognition model in the video stream processing, receiving the image data of the input video stream for analysis; inputting the pre-processed image frame into the intelligent recognition model for real-time detection; the intelligent recognition model identifies the safety hazards therein;

[0144] S303: Automatically record the identified safety hazard information and store the detection results in the database by category; check whether the detected hazards meet the corresponding safety specifications and requirements through the safety hazard standard library; summarize the detected hazards and trigger automatic alarms or remind relevant management personnel according to the severity and urgency.

[0145] The working principle and beneficial effects of the above technical solution are as follows: In this embodiment, the detection equipment is first started to obtain the monitoring video stream of the construction site in real time, and the latest dynamics are captured through high-frequency frame acquisition; according to the intelligent detection strategy, key frames are extracted from the video stream according to the set frame extraction frequency; basic image preprocessing is performed on the extracted frames; secondly, the previously trained intelligent recognition model is loaded in the video stream processing, and the image data of the input video stream is received for analysis; the preprocessed image frames are input into the intelligent recognition model for real-time detection; the intelligent recognition model identifies the safety hazards therein; finally, the identified safety hazard information is automatically recorded, and the detection results are stored in the database by category; the detected hazards are checked through the safety hazard standard library to see if they meet the corresponding safety specifications and requirements; the detected hazards are summarized, and automatic alarms are triggered or relevant management personnel are reminded according to the severity and urgency. The step S301 of the above scheme is to obtain and preprocess the video stream to ensure the normal operation of the equipment, monitor the construction site in real time, and capture key dynamic changes; high-frequency frame acquisition ensures that the details of each moment can be captured, which helps to identify potential safety hazards in a timely manner and reduce the time window for safety accidents; implement frame extraction according to the intelligent detection strategy to reduce the amount of processed data while retaining important information to ensure processing efficiency and accuracy; perform noise reduction, normalization and other operations on the extracted frames to improve image quality and provide clearer and more effective input data for subsequent model analysis. The significance achieved: Through high-frequency frame acquisition and key frame extraction, the computational burden is reduced while ensuring that important information is fully utilized; high-quality preprocessed images provide more representative input for the intelligent recognition model, which helps to improve the accuracy and reliability of model recognition. Step S302 is to load and detect the intelligent recognition model in real time, prepare the trained model so that it can quickly respond to input data, perform real-time analysis, and provide the preprocessed key frames to the model so that it can make effective judgments and classifications; the model performs reasoning based on the input image data to identify possible safety hazards, such as high-risk behaviors and illegal operations. Significance achieved: Use the trained intelligent model to complete real-time detection, reduce manual intervention, and achieve efficient automated safety management; through real-time detection, ensure that hidden dangers can be identified at the first time, create conditions for timely measures, and reduce accident risks. Step S303 records the detection results and stores them. For each safety hazard detected, a record is automatically generated, including information such as hazard category, timestamp, location, and related confidence; the recorded hidden danger information is systematically stored in the database for subsequent query and management; according to the safety hazard standard library, the identified hidden dangers are compared to ensure that they meet safety specifications. If hidden dangers that do not meet the specifications are found, they will be marked by relevant mechanisms; all detected hidden dangers are summarized, and the alarm or reminder system is triggered according to the severity and urgency of the hidden dangers, and the relevant management personnel are notified to handle them accordingly.Significance achieved: By systematically recording hidden danger information, not only the construction safety monitoring capability is improved, but also a basis is provided for subsequent data analysis and safety management decision-making; the automatic summary and alarm mechanism can promote rapid response, ensuring that potential hidden dangers can be quickly noticed by management personnel, so that appropriate measures can be taken to speed up the handling of hidden dangers and improve the safety of the overall construction site.

[0146] In summary, the various steps of this embodiment work together to achieve a complete closed loop from video monitoring to hidden danger identification, recording and response. This not only improves monitoring efficiency and the accuracy of hidden danger identification, but also provides reliable data support and decision-making basis for safety management, thereby achieving more efficient and reliable construction safety management.

[0147] Example 10: Fig.10 As shown, based on Example 9, the process of storing the detection results in the database by category provided by the embodiment of the present invention includes the following steps:

[0148] S3031: Through the automatic recording program, the identified hidden danger information is stored in a structured format; the database table structure is designed according to the hidden danger category, and the information is stored in the database;

[0149] S3032: Summarize the identified hidden danger information and build an intelligent early warning system. By setting thresholds, the automatic alarm system is triggered according to its severity and urgency, and the management personnel are notified in time;

[0150] S3033: Regularly analyze the hidden danger data stored in the database and provide feedback to the intelligent identification model.

[0151] The working principle and beneficial effects of the above technical solution are as follows: first, the present embodiment stores the identified hidden danger information in a structured format through an automatic recording program; designs a database table structure according to the hidden danger category, and stores the information in the database; secondly, the identified hidden danger information is summarized, and an intelligent early warning system is constructed, and the automatic alarm system is triggered according to its severity and urgency by setting a threshold, and the management personnel are notified in time; finally, the hidden danger data stored in the database is regularly analyzed to provide feedback to the intelligent recognition model. Step S3031 of the above scheme automatically records and stores the hidden danger information, develops a program that can efficiently and accurately record the identified hidden danger information, and ensures that the hidden danger information can be quickly archived after detection; uses a structured format to standardize the storage of hidden danger information, which is convenient for subsequent retrieval and analysis; designs a reasonable database table structure according to the hidden danger category, nature, etc., to ensure the classification storage and query efficiency of information. The significance achieved: Systematic recording ensures that each hidden danger is accurately recorded, which is conducive to maintaining the consistency and integrity of the data; structured storage makes the retrieval and management of hidden danger information efficient, which is convenient for subsequent report collation, data analysis and decision support; through a standardized storage structure, management personnel can more easily analyze hidden danger data, generate reports, and formulate safety improvement measures. Step S3032: Hidden danger information summary and intelligent early warning: sort and summarize the hidden danger information stored in the database to form a usable data set for subsequent analysis and decision-making; develop an intelligent early warning system to make judgments based on the set thresholds and the severity and urgency of the hidden dangers; when serious hidden dangers are identified, the system can automatically trigger an alarm to notify relevant managers to ensure timely response. Significance achieved: The intelligent early warning system can respond quickly after the hidden dangers are discovered, enhance the early warning of potential safety risks, and reduce the probability of accidents; through the assessment of the severity and urgency of hidden dangers, managers can allocate resources more scientifically and give priority to the most urgent and serious problems; promptly notify managers to take measures to prevent small problems from evolving into major safety accidents and maintain the safety of the construction site. Step S3033: Data analysis and model feedback: regularly parse and summarize the hidden danger data stored in the database, identify potential trends and laws, so as to carry out targeted safety management; feed back the analysis results to the intelligent identification model to help the model continue to learn and improve, and improve the accuracy and sensitivity of subsequent hidden danger identification. Significance achieved: Through the analysis of historical data, the model can be updated and optimized in real time, thereby improving its recognition rate and accuracy of safety hazards; regular analysis of hazard data allows managers to better understand safety conditions and potential risks, and help formulate more effective safety strategies and measures; through the establishment of a feedback mechanism, a virtuous circle is formed to promote the coordinated development of security monitoring technology and artificial intelligence, and improve overall safety management efficiency.

[0152] In summary, the various steps in the process of storing the detection results in the database by category in this embodiment are interconnected and each plays an important role. Step S3031 ensures the systematic storage of hidden danger information, step S3032 realizes rapid risk warning and response, and step S3033 helps to continuously optimize the detection model through data analysis. It will effectively improve the safety management efficiency of the construction site, reduce the risk of accidents, and improve the overall management level.

[0153] Example 11: Fig.11 As shown, based on Example 9, the embodiment of the present invention provides a method for supervising the safety hazard processing results according to the safety hazard standard library, which includes the following steps:

[0154] The construction contract signed between the construction unit and the owner unit generally contains contract settlement and safety reward and punishment clauses. Therefore, the construction unit must conduct safety training in accordance with relevant laws and regulations during the execution of the construction contract, and frequently carry out and organize safety inspections during the construction process to check for safety hazards. In the past, safety managers went to the construction site to manually check for safety hazards. At present, due to the application of AI technology, AI analysis can be performed through on-site monitoring videos to detect typical safety hazards that often occur, such as not wearing a safety helmet or not tying a safety rope, which improves the efficiency of safety hazard inspection. Whether it is manual inspection of safety hazards or intelligent detection of safety hazards, once discovered, the safety manager will initiate a safety hazard handling process in the system and push the safety hazards to the relevant construction unit for rectification within the prescribed period. When the owner unit settles the contract with the construction unit, the system will automatically detect the handling of the safety hazards of the construction unit. If the construction unit fails to complete the rectification of the safety hazards within the prescribed period, according to the safety reward and punishment clauses agreed in the contract, the fine for not handling the safety hazards in time must be deducted from the settlement amount. In general, in order to avoid being punished during settlement, the construction unit will rectify the safety hazards in a timely manner. In this way, the efficiency of hazard rectification is improved through the control of contract settlement.

[0155] Example 12: Fig.12 As shown, an embodiment of the present invention provides a construction safety intelligent management and control system based on a potential safety hazard database, comprising:

[0156] The model training module is responsible for extracting hidden danger classifications and items from the safety production standardization specifications, and establishing a standard library of hidden dangers for safety hazard classification and hierarchical management; building an intelligent detection and recognition model based on the standard library of hidden dangers, and organizing the process image data of the construction stage, performing data annotation, and using the process image data to train the intelligent recognition model;

[0157] The strategy execution module is responsible for installing detection equipment in the area to be detected on the construction site and recording the installation and control areas of the detection equipment; configuring intelligent detection strategies according to the classification and hierarchical management requirements of the safety hazard standard library, including area division, frame extraction frequency and real-time detection requirements;

[0158] The intelligent detection module is responsible for obtaining the video stream from the detection equipment and calling the intelligent recognition model for detection, automatically recording the detection results and storing them in the database by category; and supervising the safety hazard handling results according to the safety hazard standard library.

[0159] The working principle and beneficial effects of the above technical solution are as follows: the model training module of this embodiment extracts hidden danger classification and items from the safety production standardization specification, establishes a standard library of safety hidden danger classification and hierarchical management; constructs an intelligent detection and recognition model according to the standard library of safety hidden dangers, organizes the process image data of the construction stage, performs data annotation, and uses the process image data to train the intelligent recognition model; the strategy execution module installs the detection equipment in the area to be detected on the construction site, and records the installation and control area of ​​the detection equipment; configures the intelligent detection strategy according to the classification and hierarchical management requirements of the standard library of safety hidden dangers, including regional division, frame extraction frequency and real-time detection requirements; the intelligent detection module obtains the video stream from the detection equipment and calls the intelligent recognition model for detection, obtains the existence of safety hidden dangers, automatically records the detection results and stores them in the database by category; supervises the safety hidden danger processing results according to the standard library of safety hidden dangers. The model training module of the above scheme parses the hidden danger classification and items from the safety production standardization specification, forms a standard library of safety hidden danger classification and hierarchical management, ensures that the hidden danger definition is comprehensive and authoritative; organizes the process image data of the construction stage, annotates and uses these data to train the intelligent recognition model, so that the model can have the automatic recognition ability of potential hidden dangers. Significance achieved: Model construction based on standardized hidden danger library improves the accuracy and versatility of the model so that it can adapt to the identification of hidden dangers in different construction scenarios; through scientific classification and grading standards, it helps on-site managers effectively identify and respond to various safety hazards and improve the overall safety management level. The strategy execution module is responsible for installing suitable detection equipment at the construction site and recording relevant information to ensure the orderly progress of detection work; the safety hazard standard library formulates corresponding intelligent detection strategies, including regional division, frame extraction frequency and real-time detection requirements, to ensure efficient monitoring within a reasonable range. Significance achieved: Through the reasonable configuration of detection areas and equipment, the construction site can be managed in a refined manner, and the coverage and effectiveness of safety monitoring can be improved; resources can be reasonably allocated according to actual needs, detection efficiency can be improved, and more intensive and real-time monitoring can be ensured in high-risk areas to reduce the probability of safety hazards. The intelligent detection module extracts video streams from the detection equipment in real time, calls the intelligent recognition model for detection, and promptly identifies existing safety hazards; automatically records the detection results, and stores them in the database by category to ensure the effective management of hidden danger information. ; Supervise the handling results of safety hazards according to the safety hazard standard library to ensure that the handling of hidden dangers meets the standard requirements. Significance achieved: By identifying and recording hidden dangers in real time, providing a rapid response mechanism, improving the timeliness of accident handling, and minimizing the accident rate; by storing the test results in the database, managers can analyze the data, formulate improvement measures, and promote continuous improvement of safety management; supervise the results of safety hazard handling, ensure that construction companies follow safety regulations, and improve project quality and safety.

[0160] In summary, this embodiment forms a good collaborative relationship between the modules by constructing an intelligent construction safety management and control system based on a potential safety hazard database, aiming to comprehensively improve the safety management and potential hazard handling capabilities of the construction site. The model training module ensures the scientificity and accuracy of the detection model, the strategy execution module is responsible for the reasonable layout and resource allocation of on-site detection, and the intelligent detection module provides real-time monitoring and data support. Through the effective operation of these three modules, the system can achieve dynamic monitoring, timely response and data analysis, ultimately improving the level of construction safety prevention and control, and providing an intelligent solution for the safety management of the construction site.

[0161] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention belong to the scope of equivalent technologies of the present invention, the present invention is also intended to include these modifications and variations.

Claims

1. A construction safety intelligent management and control method based on a potential safety hazard database. It is characterized in that The following steps are involved: Extract hidden danger classifications and items from safety production standardization specifications, and establish a standard library of hidden dangers for classification and hierarchical management of hidden dangers; build an intelligent detection and recognition model based on the standard library of hidden dangers, organize process image data during the construction phase, perform data annotation, and use process image data to train the intelligent recognition model; Install testing equipment in the area to be tested on the construction site, and record the installation and control areas of the testing equipment; configure intelligent testing strategies according to the classification and hierarchical management requirements of the safety hazard standard library; Obtain video streams from detection equipment and call intelligent recognition models for detection, find out potential safety hazards, automatically record detection results and store them in the database by category; Monitor the results of safety hazard handling based on the safety hazard standard library.

2. The construction safety intelligent management and control method based on the potential safety hazard database according to claim 1, It is characterized in that The process of establishing a standard library of safety hazards includes the following steps: Confirm the information types contained in the safety hazard standard library, establish a regulatory document database, and mark the document source and scope of application; Mark key paragraphs of the regulatory document database; establish a classification system based on information types; design the structure of the safety hazard standard library, determine the primary and foreign key relationships, realize the association between data, and establish indexes; Add the user interface of the safety hazard standard library, including the main menu, query interface and information entry interface.

3. The construction safety intelligent management and control method based on the potential safety hazard database according to claim 2, It is characterized in that The information type includes hidden danger project code, hidden danger project name, hidden danger point, basis, hidden danger project classification attribute and hidden danger level.

4. The construction safety intelligent management and control method based on the potential safety hazard database according to claim 1, It is characterized in that The construction process of the intelligent recognition model includes the following steps: Collect process image data at different construction stages under different conditions, perform image scaling operations on all process image data to ensure that each frame of the image has the same resolution and size, use a consistent cropping strategy to highlight key construction areas; and perform image enhancement operations at the same time; Label the process impact data after the enhanced operation, distinguish the types and characteristics of security, and obtain the pre-processed process impact data set; The preprocessed process impact data set is divided into a basic data set, a specific data set and a validation set; the basic data set is used to conduct preliminary training on the intelligent recognition model, the specific data set is used to fine-tune the intelligent recognition model, and the validation set is used to evaluate the performance of the intelligent recognition model and record the accuracy, precision and recall rate indicators of the intelligent recognition model in identifying different types of hidden dangers; A classifier including a feature extraction layer, a context information module, a feature fusion layer, and a classification layer is used as the framework of an intelligent recognition model to identify safety hazards. The basic data set is used to perform preliminary training of the intelligent recognition model, and then the specific data set is used to fine-tune the intelligent recognition model, and the validation set is used to evaluate the performance of the intelligent recognition model.

5. The construction safety intelligent management and control method based on the potential safety hazard database according to claim 4, It is characterized in that The training process of the intelligent recognition model includes the following steps: Obtain the basic data set, start the randomly disrupted data loading strategy, and input the basic data set after the data loading strategy into the intelligent recognition model for preliminary training; Obtain a specific data set, fine-tune the initially trained intelligent recognition model, and adopt an adaptive small batch size strategy, using 1 / 3 of the specific data set in the early stage of fine-tuning, and increasing the batch size to 2 / 3 of the specific data set in the later stage of training; The fine-tuned intelligent recognition model is evaluated using a validation set, and a variety of evaluation indicators are used to comprehensively analyze the performance of the intelligent recognition model.

6. The construction safety intelligent management and control method based on the potential safety hazard database according to claim 1, It is characterized in that Intelligent detection strategies, including area division, frame extraction frequency and real-time detection requirements.

7. The construction safety intelligent management and control method based on the potential safety hazard database according to claim 1, It is characterized in that The process of intelligent detection strategy includes the following steps: Acquire image data through a detection device, the image data being a static photo or a video stream; preprocess the image data to obtain preprocessed image data; Through edge detection, key information is extracted from the preprocessed image data, the image data set consisting of the key information is marked, and the model of the intelligent detection strategy is trained to obtain a trained model for testing and inference; When a new image is input, the previously learned knowledge is used to identify the objects or patterns in the image and output a category label and location information.

8. The construction safety intelligent management and control method based on the potential safety hazard database according to claim 7, It is characterized in that The process of obtaining a trained model includes the following steps: Use edge detection algorithm to process the preprocessed image, calculate the gradient strength and direction of the image, use non-maximum suppression to remove non-edge points, and perform double threshold processing to determine the edge; identify and extract specific patterns through image segmentation, use morphological operations to analyze the structural characteristics of the image, use color histogram to analyze the color distribution in the image, and extract information in the color space; The extracted features are labeled, a labeled data set is created, and the training set divided by the labeled data is input into the model for training. Multiple epochs of training are performed, and each epoch will be run on the training set. The performance is evaluated on the validation set, and the model performance is evaluated on the test set, and the accuracy, recall and F1-score indicators are calculated.

9. The construction safety intelligent management and control method based on the potential safety hazard database according to claim 1, It is characterized in that The process of automatically recording the test results and storing them in the database by category includes the following steps: Start the detection equipment to obtain the monitoring video stream of the construction site in real time, and ensure the latest dynamics are captured through high-frequency frame acquisition; According to the intelligent detection strategy, key frames are extracted from the video stream according to the set frame extraction frequency; basic image preprocessing is performed on the extracted frames; Load the previously trained intelligent recognition model in the video stream processing, and receive the image data of the input video stream for analysis; Input the preprocessed image frames into the intelligent recognition model for real-time detection; Intelligent recognition model identifies potential safety hazards; Automatically record the identified safety hazard information and store the test results into the database by category; Check whether the detected hidden dangers meet the corresponding safety specifications and requirements through the safety hidden danger standard library; summarize the detected hidden dangers, and trigger automatic alarms or remind relevant managers according to the severity and urgency.

10. An intelligent construction safety management and control system based on a potential safety hazard database. It is characterized in that Include: The model training module is responsible for extracting hidden danger classifications and items from the safety production standardization specifications, and establishing a standard library of hidden dangers for safety hazard classification and hierarchical management; building an intelligent detection and recognition model based on the standard library of hidden dangers, and organizing the process image data of the construction stage, performing data annotation, and using the process image data to train the intelligent recognition model; The strategy execution module is responsible for installing detection equipment in the area to be detected on the construction site and recording the installation and control areas of the detection equipment; configuring intelligent detection strategies according to the classification and hierarchical management requirements of the safety hazard standard library; The intelligent detection module is responsible for obtaining the video stream from the detection equipment and calling the intelligent recognition model for detection, automatically recording the detection results and storing them in the database by category; and supervising the safety hazard handling results according to the safety hazard standard library.

Images