Engineering safety intelligent management method and system based on AI hidden danger recognition

Abstract

The invention discloses an engineering safety intelligent management method and system based on AI hidden danger recognition, and relates to the technical field of intelligent detection management. Real-time image data Raw are obtained through a camera, preprocessed image data Ima are obtained through preprocessing, an image feature set FTS is extracted through an AI model, and objective and automatic data collection of the field state is achieved. And based on the image feature set FTS and a preset standard, the material stacking stability Smp and the scaffold structure defect degree Dsf are calculated, standardized quantitative evaluation is provided, and subjective judgment is eliminated. Comprehensive risk indexes of the material stacking area and the scaffold area are calculated, and macroscopic study and judgment of the area safety situation are formed. And calculating a comprehensive safety state evaluation value Vto and generating alarm information Air, thereby realizing automatic risk decision and instant early warning. And rectification is carried out according to the alarm information Air, iterative optimization is carried out based on the feedback information Fd, and a continuous learning and self-optimization closed-loop management system is constructed.

Description

Technical Field

[0001] This invention relates to the field of intelligent detection and management technology, specifically to an intelligent management method and system for engineering safety based on AI-based hazard identification. Background Technology

[0002] The engineering construction field, due to its large-scale projects and dynamic, ever-changing environments, places extremely high demands on management technology. Safety management is central to ensuring the smooth progress of projects, and monitoring the status of large amounts of material storage and temporary scaffolding on-site is of paramount importance. In typical application scenarios such as large-scale buildings or infrastructure, how to efficiently and accurately monitor these constantly changing physical entities has become a major management challenge. Therefore, the industry urgently needs an innovative technological solution: a method and system for intelligent engineering safety management. Its core is based on advanced AI technology, endowing machines with vision perception and analysis capabilities to automatically identify and assess various safety hazards.

[0003] The traditional management model, which relies primarily on manual inspections, has revealed several shortcomings in the management of material storage and scaffolding facilities in the aforementioned scenarios. First, inspection results are influenced by the safety officer's personal experience and subjective judgment, making it difficult to standardize and prone to assessment bias. Second, limited by manpower and time costs, inspections cannot achieve continuous, 24 / 7 monitoring, allowing risks to arise in unchecked periods and areas. Third, inspection data is mostly paper-based or recorded offline, resulting in significant information lag. This not only hinders real-time early warnings but also prevents managers from effectively summarizing and deeply analyzing the data, making it difficult to discern systemic risk trends. Finally, for slow-developing, easily overlooked hidden hazards such as changes in internal stress within material piles or the accumulation of minute deformations in scaffolding, effective and continuous monitoring by the human eye is extremely challenging. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an intelligent management method and system for engineering safety based on AI-based hazard identification, which solves the problems mentioned in the background section.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an intelligent management method for engineering safety based on AI-based hazard identification, comprising the following steps:

[0006] S1. Acquire real-time image data Raw through cameras installed at the engineering site, preprocess the real-time image data Raw to obtain preprocessed image data Ima, and use an AI model to segment the target in the preprocessed image data Ima and extract the image feature set FTS.

[0007] S2. Based on the image feature set FTS and combined with the preset engineering safety standard CR, calculate the material stacking stability Smp and the scaffolding structure defect degree Dsf.

[0008] S3. Based on the material stacking stability Smp and the scaffolding structural defect degree Dsf, calculate the comprehensive risk index Rms of the material stacking area and the comprehensive risk index Rsf of the scaffolding, respectively.

[0009] S4. Based on the comprehensive risk index algorithm Rms for material storage area and the comprehensive risk index Rsf for scaffolding, combined with the historical database HIS, calculate the comprehensive safety status assessment value Vto and compare it with the graded comprehensive safety status assessment threshold Fth. Based on the comparison result, generate alarm information Air.

[0010] S5. Safety management personnel rectify the site according to the alarm information Air and record the feedback information Fd. They then upload the alarm information Air and the feedback information Fd and update the historical database HIS. Based on the feedback information Fd and the historical database HIS, they perform iterative optimization.

[0011] Preferably, S1 includes S11 and S12;

[0012] S11. By controlling the cameras deployed at the engineering site to acquire images, real-time image data Raw is obtained. The Brown-Conrady model is used to perform preliminary image correction on the real-time image data Raw. Adaptive histogram equalization is used to optimize local contrast and enhance image edges. The enhanced image parameters are adjusted to the preset ModelInputReq format to obtain preprocessed image data Ima. Specifically, the preset ModelInputReq format has a fixed image size of 640*640, 3 image channels, and the image pixel values ​​are normalized to [0,1] and the pixel value data type is converted to 32-bit floating-point type.

[0013] The Mask R-CNN model is used to segment the preprocessed image data Ima, identify and divide the material pile area MP and the scaffolding unit area SU, and generate the instance segmentation mask Mask and the target bounding box coordinates Box = {Xmin, Ymin, Xmax, Ymax}, where Xmin represents the minimum value of the target X coordinate, Ymin represents the minimum value of the target Y coordinate, Xmax represents the maximum value of the target X coordinate, and Ymax represents the maximum value of the target Y coordinate.

[0014] Preferably, in step S12, for the material pile area MP, the estimated actual height Hgt_i of the material pile MP_i is estimated based on the instance segmentation mask Mask(MP_i) of the material pile MP_i, combined with the monocular depth estimation algorithm. The bottom area Ara_i of the material pile MP_i is estimated based on the projection of the instance segmentation mask Mask(MP_i) of the material pile MP_i onto the ground plane. The estimated volume Vol_i of the material pile MP_i is estimated using 3D reconstruction technology based on the instance segmentation mask Mask(MP_i) and the estimated actual height Hgt_i of the material pile MP_i. The 3D minimum bounding box volume Vbox_i of the material pile MP_i is calculated based on the target bounding box coordinates Box(MP_i) of the material pile MP_i. The density factor Cpt_i of the material pile MP_i is obtained based on the ratio of the estimated volume Vol_i of the material pile MP_i to the 3D minimum bounding box volume Vbox_i, and the material pile feature set FTS_MP is generated.

[0015] For the scaffolding unit area SU, based on the instance segmentation mask Mask(SU_j) of the j-th scaffolding unit SU_j in the scaffolding unit area SU, the Mask R-CNN model is used to segment and calculate the number of vertical load-bearing members Nvs_j and the number of horizontal connecting members Nhl_j. The straightness of the vertical load-bearing members and the connection point position of the horizontal connecting members of the scaffolding unit SU_j are segmented by the Mask R-CNN model and compared with the preset ideal model and the preset maximum allowable structural deformation model to evaluate the structural deformation degree Def_j of the scaffolding unit SU_j. The value range of the structural deformation degree Def_j of the scaffolding unit SU_j is normalized to [0,1], where 0 indicates no obvious deformation and 1 indicates severe deformation. The scaffolding unit feature set FTS_SU is generated.

[0016] The material pile feature set FTS_MP and the scaffolding unit feature set FTS_SU are integrated to generate the image feature set FTS.

[0017] Preferably, S2 includes S21;

[0018] S21. Based on the material pile feature set FTS_MP in the image feature set FTS, and combined with the preset engineering safety standard CR, construct a stability index algorithm to calculate the material stacking stability Smp_i of the material pile MP_i.

[0019] The stability index algorithm expression is as follows:

[0020]

[0021] In the formula, α represents the sensitivity adjustment index of the bottom area Ara_i of the material pile MP_i from the preset engineering safety standard CR, β represents the sensitivity adjustment index of the estimated actual height Hgt_i of the material pile MP_i from the preset engineering safety standard CR, and ε represents the configuration parameter, specifically 1*10. -4 ;

[0022] Based on the scaffolding unit feature set FTS_SU in the image feature set FTS, and combined with the preset engineering safety standard CR, a scaffolding structure defect index algorithm is constructed to calculate the scaffolding structure defect degree Dsf_j of scaffolding unit SU_j.

[0023] The algorithm expression for the scaffolding structural defect index is as follows:

[0024]

[0025] In the formula, σ represents the influence factor of the structural deformation degree Def_j of scaffolding unit SU_j from the preset engineering safety standard CR, k represents the theoretical proportionality factor of the number of vertical load-bearing members Nvs_j and the number of horizontal connecting members Nhl_j of scaffolding unit SU_j from the preset engineering safety standard CR, exp represents the natural exponential function, and |……| represents the absolute value symbol.

[0026] Preferably, S3 includes S31 and S32;

[0027] S31. Based on the material stacking stability Smp_i of material stack MP_i and the estimated actual height Hgt_i of material stack MP_i in the material stack feature set FTS_MP, and combined with the preset engineering safety standard CR, construct the comprehensive risk index algorithm for the material stacking area, and calculate the comprehensive risk index Rms of the material stacking area.

[0028] The algorithm expression for the comprehensive risk index of the material storage area is as follows:

[0029]

[0030] In the formula, Nmp represents the total number of material piles in the material storage area MP, Ts_max represents the upper limit reference value of material pile stability from the preset engineering safety standard CR, Hsite represents the maximum allowable height of material stacking from the preset engineering safety standard CR, δ1 represents the stability risk contribution adjustment index from the preset engineering safety standard CR, δ2 represents the ultra-high risk contribution adjustment index from the preset engineering safety standard CR, and max(a, b) represents taking the maximum value between a and b.

[0031] Preferably, in step S32, based on the scaffolding unit SU_j, the human entities within the preprocessed image data Ima are identified and segmented using the Mask R-CNN model. By comparing the number of people within the scaffolding unit SU_j in the image with the maximum number of people that the scaffolding unit SU_j can accommodate in the preset engineering safety standard CR, the personnel occupancy factor Loc_j of the scaffolding unit SU_j in the preprocessed image data Ima is obtained. Combined with the preset engineering safety standard CR, a comprehensive scaffolding risk index algorithm is constructed, and the comprehensive scaffolding risk index Rsf is calculated.

[0032] The algorithm expression for the comprehensive risk index of scaffolding is as follows:

[0033]

[0034] In the formula, Nsu represents the total number of scaffold units in the scaffold unit area SU, w1 represents the structural defect contribution adjustment index from the preset engineering safety standard CR, and w2 represents the structural deformation contribution adjustment index from the preset engineering safety standard CR.

[0035] Preferably, S4 includes S41 and S42;

[0036] S41. Based on the comprehensive risk index algorithm Rms for material storage areas and the comprehensive risk index Rsf for scaffolding, calculate the basic sum of standardized risk components SC. According to the historical database HIS, obtain the average value Savg of the basic sum of standardized risk components Sc in the historical database HIS. Combining the comprehensive risk index algorithm Rms for material storage areas, the comprehensive risk index Rsf for scaffolding, the basic sum of standardized risk components SC, and the average value Savg of the basic sum of standardized risk components Sc in the historical database HIS, construct a comprehensive safety status assessment algorithm and calculate the comprehensive safety status assessment value Vto.

[0037] The expression for the comprehensive security status assessment algorithm is as follows:

[0038]

[0039] In the formula, Rms_safe represents the upper limit of the comprehensive risk index of the material storage area from the preset engineering safety standard CR, Rsf_safe represents the upper limit of the comprehensive risk index of the scaffolding from the preset engineering safety standard CR, ζ1 represents the risk contribution adjustment index of the material storage area from the preset engineering safety standard CR, and ζ2 represents the risk contribution adjustment index of the scaffolding unit area from the preset engineering safety standard CR.

[0040] Preferably, in step S42, the comprehensive security status assessment value Vto is compared with the graded comprehensive security status assessment threshold Fth;

[0041] If the comprehensive safety status assessment value Vto is less than the low comprehensive safety status assessment threshold Fth_low, then the current status is determined to be safe, the risk level Alo is set to 0, and no alarm prompt is required.

[0042] If the low comprehensive security status assessment threshold Fth_low ≤ comprehensive security status assessment value Vto < high comprehensive security status assessment threshold Fth_high, then the current status is determined to be low risk, the risk level Alo = 1 is set, a low-level alarm is issued through the APP and a prompt is made to assign security personnel to rectify the situation within 30 minutes.

[0043] If the comprehensive security status assessment value Vto is greater than or equal to the high comprehensive security status assessment threshold Fth_high, then the current status is determined to be high risk, the risk level Alo is set to 2, a high-level alarm is issued through the APP and a prompt is made to assign security personnel to carry out rectification.

[0044] Based on the risk level Alo, the corresponding real-time image data Raw, pre-processed image data Ima, comprehensive risk index Rms of the material storage area, and comprehensive risk index Rsf of the scaffolding are collected to generate alarm information Air.

[0045] Preferably, S5 includes S51;

[0046] S51. Safety personnel go to the site to rectify the situation based on the alarm information Air and record the feedback information Fd. They then upload the alarm information Air and the recorded feedback information Fd to the historical database His. The feedback information Fd includes the actual inspection result Er, the false alarm rate Rpr, and the false alarm rate Rur.

[0047] If the false positive rate Rpr and the false negative rate Rur are less than the preset error detection rate threshold Rth, then no iterative optimization is required.

[0048] If the false alarm rate Rpr and the false alarm rate Rur are greater than or equal to the preset error detection rate threshold Rth, then the data analyst will adjust the engineering safety standard CR and the graded comprehensive safety status assessment threshold Fth.

[0049] Based on the adjusted engineering safety standard CR and the graded comprehensive safety status assessment threshold Fth, the accuracy of the adjusted engineering safety standard CR and the graded comprehensive safety status assessment threshold Fth was analyzed using preprocessed image data Ima from the historical database HIS.

[0050] An AI-based intelligent management system for engineering safety identification includes an image data acquisition and processing module, a compliance assessment module, a risk assessment module, a comprehensive risk judgment module, and an iterative optimization module.

[0051] The image data acquisition and processing module acquires real-time image data Raw through cameras installed at the engineering site, preprocesses the real-time image data Raw to obtain preprocessed image data Ima, and uses an AI model to perform target segmentation on the preprocessed image data Ima and extract the image feature set FTS.

[0052] The compliance assessment module calculates the material stacking stability Smp and the scaffolding structure defect degree Dsf based on the image feature set FTS and the preset engineering safety standard CR.

[0053] The risk assessment module calculates the comprehensive risk index Rms of the material storage area and the comprehensive risk index Rsf of the scaffolding based on the material stacking stability Smp and the scaffolding structural defect degree Dsf, respectively.

[0054] The comprehensive risk assessment module calculates the comprehensive safety status assessment value Vto based on the comprehensive risk index algorithm Rms for material storage areas and the comprehensive risk index Rsf for scaffolding, combined with the historical database HIS, and compares it with the graded comprehensive safety status assessment threshold Fth. Based on the comparison results, alarm information Air is generated.

[0055] The iterative optimization module allows safety management personnel to rectify on-site issues based on alarm information Air and record feedback information Fd. The alarm information Air and feedback information Fd are then uploaded and updated in the historical database HIS. Iterative optimization is performed based on the feedback information Fd and the historical database HIS.

[0056] This invention provides an intelligent management method and system for engineering safety based on AI-based hazard identification, which has the following beneficial effects:

[0057] (1) By utilizing AI models to objectively and meticulously extract features from on-site images, the system achieves automated and high-precision quantification of the physical entity's state. Subsequently, a standardized mathematical model is used to quantitatively assess the compliance of the extracted features, ensuring the objectivity and consistency of the assessment process. Following this, by comprehensively analyzing multi-source risk information and comparing it with the hierarchical comprehensive safety status assessment threshold Fth, automated risk decision-making and immediate early warning are achieved, effectively replacing the lagging judgments relying on human experience. Finally, by introducing feedback and iteration mechanisms, the entire system possesses the ability to learn and optimize itself. This series of interconnected technical steps achieves a fundamental shift from passive, qualitative, and discrete manual management to proactive, quantitative, and continuous intelligent management, significantly improving the accuracy of identifying safety hazards at construction sites and management efficiency, thereby effectively reducing accident risks and ensuring project safety.

[0058] (2) The project site is monitored around the clock with high-precision image data acquisition, processing, and feature extraction via cameras and AI models. This process not only achieves objective and accurate acquisition of features that are difficult to quantify manually, such as material stack height (Hgt), density (Cpt), and scaffold structure deformation (Def), but also solves the problems of incomplete coverage and easy neglect of dynamic and subtle changes during manual inspections. Subsequently, based on these objective and rich feature data, combined with the preset engineering safety standard (CR), the material stack stability (Smp) and scaffold structure defect degree (Dsf) are calculated using a standardized mathematical model. This transforms the compliance judgment, which previously relied on the subjective experience of safety officers, into a unified and reproducible quantitative assessment, completely eliminating individual differences and subjectivity in the assessment results and providing a solid and reliable data foundation for subsequent risk assessments.

[0059] (3) By aggregating the assessment indicators of individual objects into regional material storage area comprehensive risk index Rms and scaffolding comprehensive risk index Rsf, managers can grasp the overall safety situation of different areas from a macro perspective, making up for the shortcomings of manual inspections in forming a global risk view. Subsequently, through a complex algorithm that comprehensively considers synergistic effects and historical data trends, the final comprehensive safety status assessment value Vto is calculated and automatically generated as an alarm message Air by comparing it with the graded comprehensive safety status assessment threshold Fth. This realizes an automated closed loop from data analysis to risk decision-making and early warning, greatly shortening the response time and transforming safety management from post-event processing to pre-event early warning. Finally, an iterative optimization mechanism based on feedback information Fd is introduced. The system can dynamically adjust and optimize the preset engineering safety standard CR and graded comprehensive safety status assessment threshold Fth according to the actual inspection results Er, as well as the false alarm rate Rpr and the false alarm rate Rur. This learning and evolution capability ensures that the system can continuously improve its accuracy and effectiveness in long-term operation, forming a continuously improving intelligent management ecosystem. Attached Figure Description

[0060] Figure 1 This is a schematic diagram illustrating the steps of an AI-based intelligent management method for engineering safety based on hazard identification, as described in this invention.

[0061] Figure 2 This is a schematic block diagram of an intelligent engineering safety management system based on AI hazard identification according to the present invention.

[0062] Figure 3 A data flow diagram for calculating the comprehensive safety status assessment value Vto. Detailed Implementation

[0063] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0064] Example 1.

[0065] This invention provides an intelligent engineering safety management method based on AI-based hazard identification. Please refer to [link / reference]. Figure 1 and Figure 3 This includes the following steps:

[0066] S1. Acquire real-time image data Raw through cameras installed at the engineering site, preprocess the real-time image data Raw to obtain preprocessed image data Ima, and use an AI model to segment the target in the preprocessed image data Ima and extract the image feature set FTS.

[0067] S2. Based on the image feature set FTS and combined with the preset engineering safety standard CR, calculate the material stacking stability Smp and the scaffolding structure defect degree Dsf.

[0068] S3. Based on the material stacking stability Smp and the scaffolding structural defect degree Dsf, calculate the comprehensive risk index Rms of the material stacking area and the comprehensive risk index Rsf of the scaffolding, respectively.

[0069] S4. Based on the comprehensive risk index algorithm Rms for material storage area and the comprehensive risk index Rsf for scaffolding, combined with the historical database HIS, calculate the comprehensive safety status assessment value Vto and compare it with the graded comprehensive safety status assessment threshold Fth. Based on the comparison result, generate alarm information Air.

[0070] S5. Safety management personnel rectify the site according to the alarm information Air and record the feedback information Fd. They then upload the alarm information Air and the feedback information Fd and update the historical database HIS. Based on the feedback information Fd and the historical database HIS, they perform iterative optimization.

[0071] In this embodiment, the image feature set FTS, which is difficult to grasp manually using traditional methods, is first extracted, such as the key density factor Cpt and the structural deformation degree Def. This objective and detailed data input provides a solid foundation for standardized compliance calculations. Based on this, the system calculates quantitative material stacking stability Smp and scaffolding structural defect degree Dsf using a fixed mathematical model. This transformation from objective data to quantitative indicators completely replaces subjective judgments relying on personal experience, fundamentally solving the problems of assessment bias and inconsistent standards caused by differences in inspection personnel in the background technology. After obtaining individual compliance indicators, this invention further realizes the macro-level and real-time nature of risk assessment. The system aggregates the discrete material stacking stability Smp and scaffolding structural defect degree Dsf calculated in the previous step to form a comprehensive risk index Rms for the material stacking area and a comprehensive risk index Rsf for the scaffolding, reflecting the overall safety level of the area. This analysis from "point" to "surface" allows managers to gain a comprehensive understanding of the overall situation. Following this, the system calculates the final comprehensive safety status assessment value Vto based on these two macro-risk indices and immediately generates an alarm message Air. This automated analysis and decision-making process effectively overcomes the drawbacks of data lag and slow response in traditional management, elevating safety monitoring from periodic spot checks to continuous and proactive early warning. More importantly, this invention constructs a closed-loop management system with self-evolving capabilities. The feedback information Fd submitted by on-site personnel after rectifying issues based on the alarm message Air, especially the key performance indicators containing the false alarm rate Rpr and the false alarm rate Rur, provides valuable real-world verification data. Based on this feedback, the engineering safety standard CR and the graded comprehensive safety status assessment threshold Fth are dynamically adjusted and iteratively optimized. This learning mechanism ensures that the assessment model and judgment criteria can continuously adapt to changes on-site, constantly improving their accuracy and reliability—something that traditional static management methods simply cannot achieve.

[0072] Example 2.

[0073] This embodiment is an explanation based on Embodiment 1. Please refer to the following: Figure 1 and Figure 3 Specifically: S1 includes S11 and S12;

[0074] S11. By controlling the cameras deployed at the engineering site to acquire images, real-time image data Raw is obtained. The Brown-Conrady model is used to perform preliminary image correction on the real-time image data Raw. Adaptive histogram equalization is used to optimize local contrast and enhance image edges. The enhanced image parameters are adjusted to the preset ModelInputReq format to obtain preprocessed image data Ima. Specifically, the preset ModelInputReq format has a fixed image size of 640*640, 3 image channels, and the image pixel values ​​are normalized to [0,1] and the pixel value data type is converted to 32-bit floating-point type.

[0075] The Mask R-CNN model is used to segment the preprocessed image data Ima, identify and divide the material pile area MP and the scaffolding unit area SU, and generate the instance segmentation mask Mask and the target bounding box coordinates Box = {Xmin, Ymin, Xmax, Ymax}, where Xmin represents the minimum value of the target X coordinate, Ymin represents the minimum value of the target Y coordinate, Xmax represents the maximum value of the target X coordinate, and Ymax represents the maximum value of the target Y coordinate.

[0076] S12. For the material pile area MP, based on the instance segmentation mask Mask(MP_i) of the material pile MP_i, and combined with the monocular depth estimation algorithm, estimate the estimated actual height Hgt_i of the material pile MP_i. Based on the projection of the instance segmentation mask Mask(MP_i) of the material pile MP_i onto the ground plane, estimate the bottom area Ara_i of the material pile MP_i. Based on the instance segmentation mask Mask(MP_i) and the estimated actual height Hgt_i of the material pile MP_i, estimate the estimated volume Vol_i of the material pile MP_i through 3D reconstruction technology. Based on the target bounding box coordinates Box(MP_i) of the material pile MP_i, calculate the 3D minimum bounding box volume Vbox_i of the material pile i. Based on the ratio of the estimated volume Vol_i of the material pile MP_i to the 3D minimum bounding box volume Vbox_i, obtain the compactness factor Cpt_i of the material pile MP_i, and generate the material pile feature set FTS_MP.

[0077] For the scaffolding unit area SU, based on the instance segmentation mask Mask(SU_j) of the j-th scaffolding unit SU_j in the scaffolding unit area SU, the Mask R-CNN model is used to segment and calculate the number of vertical load-bearing members Nvs_j and the number of horizontal connecting members Nhl_j. The straightness of the vertical load-bearing members and the connection point position of the horizontal connecting members of the scaffolding unit SU_j are segmented by the Mask R-CNN model and compared with the preset ideal model and the preset maximum allowable structural deformation model to evaluate the structural deformation degree Def_j of the scaffolding unit SU_j. The value range of the structural deformation degree Def_j of the scaffolding unit SU_j is normalized to [0,1], where 0 indicates no obvious deformation and 1 indicates severe deformation. The scaffolding unit feature set FTS_SU is generated.

[0078] The material pile feature set FTS_MP and the scaffolding unit feature set FTS_SU are integrated to generate the image feature set FTS.

[0079] S2 includes S21;

[0080] S21. Based on the material pile feature set FTS_MP in the image feature set FTS, and combined with the preset engineering safety standard CR, construct a stability index algorithm to calculate the material stacking stability Smp_i of the material pile MP_i.

[0081] The stability index algorithm expression is as follows:

[0082]

[0083] In the formula, α represents the sensitivity adjustment index of the bottom area Ara_i of the material pile MP_i from the preset engineering safety standard CR, β represents the sensitivity adjustment index of the estimated actual height Hgt_i of the material pile MP_i from the preset engineering safety standard CR, and ε represents the configuration parameter, specifically 1*10. -4 .

[0084] Based on the scaffolding unit feature set FTS_SU in the image feature set FTS, and combined with the preset engineering safety standard CR, a scaffolding structure defect index algorithm is constructed to calculate the scaffolding structure defect degree Dsf_j of scaffolding unit SU_j.

[0085] The algorithm expression for the scaffolding structural defect index is as follows:

[0086]

[0087] In the formula, σ represents the influence factor of the structural deformation degree Def_j of scaffolding unit SU_j from the preset engineering safety standard CR, k represents the theoretical proportionality factor of the number of vertical load-bearing members Nvs_j and the number of horizontal connecting members Nhl_j of scaffolding unit SU_j from the preset engineering safety standard CR, exp represents the natural exponential function, and |……| represents the absolute value symbol.

[0088] In this embodiment, a closed-loop technology chain from data acquisition to risk assessment is constructed through multi-level image processing and dynamic parameterized modeling. Based on the Brown-Conrady model, image correction and adaptive histogram equalization are used to generate preprocessed image data Ima with a resolution of 640×640, 3 channels, and pixels normalized to [0,1] format. This effectively eliminates distortion and noise interference, providing high-quality input for the Mask R-CNN model to accurately segment the material pile area MP and the scaffolding unit area SU. On this basis, through monocular depth estimation and 3D reconstruction technology, the estimated actual height Hgt_i, bottom area Ara_i, and density factor Cpt_i of the material pile MP_i are dynamically quantified. Combined with the number of vertical load-bearing members Nvs_j, the number of horizontal connecting members Nhl_j, and the degree of structural deformation Def_j of the scaffolding unit SU_j, a refined feature description of the physical entity is formed. Further, the sensitivity parameters and stability index algorithms and scaffolding structural defect index algorithms from the Engineering Safety Standard CR are introduced. Physical characteristics are transformed into quantitative indices for the material stacking stability Smp_i of material pile MP_i and the scaffolding structural defect degree Dsf_j of scaffolding unit SU_j. By dynamically adjusting the sensitivity adjustment index α of the bottom area Ara_i of material pile MP_i and the sensitivity adjustment index β of the estimated actual height Hgt_i of material pile MP_i in the Engineering Safety Standard CR on the weight of height and area, as well as the influence factor σ of the structural deformation degree Def_j of scaffolding unit SU_j and the theoretical proportional factor k of the number of vertical load-bearing members Nvs_j and the number of horizontal connecting members Nhl_j of scaffolding unit SU_j on the sensitivity of deformation and member ratio, a high degree of objectivity and scenario adaptability of the evaluation process is achieved. The close integration of this technology chain not only solves the problem of insufficient perception of hidden volume changes and minute structural deformations in traditional manual inspections, but also avoids the risk of misjudgment caused by static thresholds through the flexible configuration of parameterized models, ultimately forming an intelligent evaluation system that combines accuracy, robustness and scalability.

[0089] Example 3.

[0090] This embodiment is an explanation based on Embodiment 2. Please refer to [link / reference]. Figure 1 and Figure 3 Specifically: S3 includes S31 and S32.

[0091] S31. Based on the material stacking stability Smp_i of material stack MP_i and the estimated actual height Hgt_i of material stack MP_i in the material stack feature set FTS_MP, and combined with the preset engineering safety standard CR, construct the comprehensive risk index algorithm for the material stacking area, and calculate the comprehensive risk index Rms of the material stacking area.

[0092] The algorithm expression for the comprehensive risk index of the material storage area is as follows:

[0093]

[0094] In the formula, Nmp represents the total number of material piles in the material storage area MP, Ts_max represents the upper limit reference value of material pile stability from the preset engineering safety standard CR, Hsite represents the maximum allowable height of material stacking from the preset engineering safety standard CR, δ1 represents the stability risk contribution adjustment index from the preset engineering safety standard CR, δ2 represents the ultra-high risk contribution adjustment index from the preset engineering safety standard CR, and max(a, b) represents taking the maximum value of a and b.

[0095] S32. Based on the scaffolding unit SU_j, the Mask R-CNN model is used to identify and segment human entities within the preprocessed image data Ima. By comparing the number of people within the scaffolding unit SU_j in the image with the maximum number of people that the scaffolding unit SU_j can accommodate in the preset engineering safety standard CR, the personnel occupancy factor Loc_j of the scaffolding unit SU_j in the preprocessed image data Ima is obtained. Combined with the preset engineering safety standard CR, a comprehensive scaffolding risk index algorithm is constructed, and the comprehensive scaffolding risk index Rsf is calculated.

[0096] The algorithm expression for the comprehensive risk index of scaffolding is as follows:

[0097]

[0098] In the formula, Nsu represents the total number of scaffold units in the scaffold unit area SU, w1 represents the structural defect contribution adjustment index from the preset engineering safety standard CR, and w2 represents the structural deformation contribution adjustment index from the preset engineering safety standard CR.

[0099] In this embodiment, an advanced risk aggregation algorithm elevates the assessment to a new dimension with a macroscopic perspective and dynamic scenario awareness. In the evaluation of the material storage area, a comprehensive risk index algorithm for the material storage area is used to calculate the comprehensive risk index Rms. When aggregating the risks of each material pile, this algorithm not only considers the material storage stability Smp_i of the material pile MP_i itself, but also amplifies the stability risk constituted by the difference between it and the upper limit reference value Ts_max of the material pile stability by introducing a stability risk contribution adjustment index δ1. Simultaneously, it emphasizes the violation risk caused by the estimated actual height Hgt_i of the material pile MP_i exceeding the maximum allowable height Hsite of the material pile through an ultra-high risk contribution adjustment index δ2, thereby achieving a comprehensive quantification of regional risks. In the evaluation of the scaffolding area, a comprehensive risk index algorithm for scaffolding is used to calculate the comprehensive risk index Rsf. This algorithm quantifies the direct risk of the scaffolding structural defect degree Dsf_j of scaffolding unit SU_j through a structural defect contribution adjustment index w1, and innovatively introduces an interactive risk term composed of the personnel occupancy factor Loc_j of scaffolding unit SU_j and the structural deformation degree Def_j. This interactive risk is adjusted by a structural deformation contribution adjustment index w2. Only when the personnel occupancy factor Loc_j of scaffolding unit SU_j, representing the presence of personnel, is not zero, will the deformation of the scaffolding pose a direct threat to people, thus amplifying its risk weight. This design transforms risk assessment from a static calculation detached from actual working conditions into a dynamic and humanistic consideration of the real-time safety exposure of personnel, greatly improving the pertinence of risk identification and the effectiveness of early warning.

[0100] Example 4.

[0101] This embodiment is an explanation based on embodiment 3. Please refer to the following: Figure 1 and Figure 3 Specifically: S4 includes S41 and S42.

[0102] S41. Based on the comprehensive risk index algorithm Rms for material storage areas and the comprehensive risk index Rsf for scaffolding, calculate the basic sum of standardized risk components SC. According to the historical database HIS, obtain the average value Savg of the basic sum of standardized risk components Sc in the historical database HIS. Combining the comprehensive risk index algorithm Rms for material storage areas, the comprehensive risk index Rsf for scaffolding, the basic sum of standardized risk components SC, and the average value Savg of the basic sum of standardized risk components Sc in the historical database HIS, construct a comprehensive safety status assessment algorithm and calculate the comprehensive safety status assessment value Vto.

[0103] The expression for the comprehensive security status assessment algorithm is as follows:

[0104]

[0105] In the formula, Rms_safe represents the upper limit of the comprehensive risk index of the material storage area from the preset engineering safety standard CR, Rsf_safe represents the upper limit of the comprehensive risk index of the scaffolding from the preset engineering safety standard CR, ζ1 represents the risk contribution adjustment index of the material storage area from the preset engineering safety standard CR, and ζ2 represents the risk contribution adjustment index of the scaffolding unit area from the preset engineering safety standard CR.

[0106] Specific example of comprehensive security status assessment:

[0107] The comprehensive risk index of the material storage area, Rms, is 0.8; the comprehensive risk index of the scaffolding, Rsf, is 1.5.

[0108] The upper limit of the comprehensive risk index for the material storage area is Rms_safe = 0.5;

[0109] The upper limit of the comprehensive risk index for scaffolding is Rsf_safe = 0.6;

[0110] The risk contribution adjustment index ζ1 for the material storage area is 1.5;

[0111] The risk contribution adjustment index ζ2 for the scaffolding unit area is 2.0;

[0112] The moderating coefficient for the synergistic effect is χ = 0.3;

[0113] The sensitivity adjustment coefficient τ for the risk trend term is 0.5;

[0114] The average value of the basic sum of standardized risk components (Sc) in the historical HIS database is Savg = 5.0;

[0115] Based on the expression of the comprehensive security status assessment algorithm:

[0116]

[0117] S42. Compare the comprehensive safety status assessment value Vto with the graded comprehensive safety status assessment threshold Fth;

[0118] If the comprehensive safety status assessment value Vto is less than the low comprehensive safety status assessment threshold Fth_low, then the current status is determined to be safe, the risk level Alo is set to 0, and no alarm prompt is required.

[0119] If the low comprehensive security status assessment threshold Fth_low ≤ comprehensive security status assessment value Vto < high comprehensive security status assessment threshold Fth_high, then the current status is determined to be low risk, the risk level Alo = 1 is set, a low-level alarm is issued through the APP and a prompt is made to assign security personnel to rectify the situation within 30 minutes.

[0120] If the comprehensive security status assessment value Vto is greater than or equal to the high comprehensive security status assessment threshold Fth_high, then the current status is determined to be high risk, the risk level Alo is set to 2, a high-level alarm is issued through the APP and a prompt is made to assign security personnel to carry out rectification.

[0121] Based on the risk level Alo, the corresponding real-time image data Raw, pre-processed image data Ima, comprehensive risk index Rms of the material storage area, and comprehensive risk index Rsf of the scaffolding are collected to generate alarm information Air.

[0122] S5 includes S51:

[0123] S51. Safety personnel go to the site to rectify the situation based on the alarm information Air and record the feedback information Fd. They then upload the alarm information Air and the recorded feedback information Fd to the historical database His. The feedback information Fd includes the actual inspection result Er, the false alarm rate Rpr, and the false alarm rate Rur.

[0124] If the false positive rate Rpr and the false negative rate Rur are less than the preset error detection rate threshold Rth, then no iterative optimization is required.

[0125] If the false alarm rate Rpr and the false alarm rate Rur are greater than or equal to the preset error detection rate threshold Rth, then the data analyst will adjust the engineering safety standard CR and the graded comprehensive safety status assessment threshold Fth.

[0126] Based on the adjusted engineering safety standard CR and the graded comprehensive safety status assessment threshold Fth, the accuracy of the adjusted engineering safety standard CR and the graded comprehensive safety status assessment threshold Fth was analyzed using preprocessed image data Ima from the historical database HIS.

[0127] In this embodiment, a comprehensive safety status assessment algorithm is used to calculate the comprehensive safety status assessment value Vto. The core technology of this algorithm lies in the introduction of a dynamic trend adjustment factor. This factor quantifies the degree and speed at which the current standardized risk component base sum Sc deviates from historical norms by comparing it with the average value (Savg) of the standardized risk component base sum Sc from the historical database HIS. When the standardized risk component base sum Sc is significantly higher than the average value (Savg) of the standardized risk component base sum Sc in the historical database HIS, this adjustment factor amplifies the final assessment result, thereby achieving accurate capture of dynamic risk trends. Based on this accurate assessment, the system compares the comprehensive safety status assessment value Vto with the low comprehensive safety status assessment threshold Fth_low and the high comprehensive safety status assessment threshold Fth_high to determine different risk levels Alo. This grading mechanism ensures that the highest level risk (Alo 2) triggers immediate rectification instructions, while providing a reasonable handling window for lower level risks (Alo 1), achieving optimized allocation of response resources and making the delivery of alarm information Air highly timely and operable. Furthermore, by analyzing the feedback information Fd after rectification, quantitative monitoring of the false alarm rate Rpr and the false negative rate Rur is introduced. Once these indicators exceed the preset error detection rate threshold Rth, the system will trigger an adjustment process for the core parameter library—namely, the engineering safety standard CR and the graded comprehensive safety status assessment threshold Fth. This adaptive optimization capability based on actual performance feedback ensures the continuous optimization of the system's risk model and decision thresholds.

[0128] Example 5.

[0129] A smart engineering safety management system based on AI-driven hazard identification; please refer to [reference needed]. Figure 2 Specifically, it includes an image data acquisition and processing module, a compliance assessment module, a risk assessment module, a comprehensive risk judgment module, and an iterative optimization module.

[0130] The image data acquisition and processing module acquires real-time image data (Raw) through cameras installed at the engineering site, preprocesses the real-time image data (Raw) to obtain preprocessed image data (Ima), and uses an AI model to perform target segmentation on the preprocessed image data (Ima) and extract the image feature set (FTS).

[0131] The compliance assessment module calculates the material stacking stability (Smp) and scaffolding structural defect degree (Dsf) based on the image feature set (FTS) and the preset engineering safety standard (CR).

[0132] The risk assessment module calculates the comprehensive risk index Rms of the material storage area and the comprehensive risk index Rsf of the scaffolding based on the material stacking stability Smp and the scaffolding structural defect degree Dsf, respectively.

[0133] The comprehensive risk assessment module calculates the comprehensive safety status assessment value Vto based on the comprehensive risk index algorithm Rms for material storage areas and the comprehensive risk index Rsf for scaffolding, combined with the historical database HIS, and compares it with the graded comprehensive safety status assessment threshold Fth. Based on the comparison results, an alarm message Air is generated.

[0134] The iterative optimization module allows safety management personnel to rectify on-site issues based on alarm information Air and record feedback information Fd. The alarm information Air and feedback information Fd are then uploaded and updated in the historical database HIS. Iterative optimization is performed based on the feedback information Fd and the historical database HIS.

[0135] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An intelligent management method for engineering safety based on AI-based hazard identification, characterized in that: Includes the following steps: S1. Acquire real-time image data Raw through cameras installed at the engineering site, preprocess the real-time image data Raw to obtain preprocessed image data Ima, and use an AI model to segment the target in the preprocessed image data Ima and extract the image feature set FTS. S2. Based on the image feature set FTS and combined with the preset engineering safety standard CR, calculate the material stacking stability Smp and the scaffolding structure defect degree Dsf. S3. Based on the material stacking stability Smp and the scaffolding structural defect degree Dsf, calculate the comprehensive risk index Rms of the material stacking area and the comprehensive risk index Rsf of the scaffolding, respectively. S4. Based on the comprehensive risk index algorithm Rms for material storage area and the comprehensive risk index Rsf for scaffolding, combined with the historical database HIS, calculate the comprehensive safety status assessment value Vto and compare it with the graded comprehensive safety status assessment threshold Fth. Based on the comparison result, generate alarm information Air. S5. Safety management personnel rectify the site according to the alarm information Air and record the feedback information Fd. They then upload the alarm information Air and the feedback information Fd and update the historical database HIS. Based on the feedback information Fd and the historical database HIS, they perform iterative optimization.

2. The intelligent engineering safety management method based on AI hazard identification according to claim 1, characterized in that: S1 includes S11: S11. By controlling the cameras deployed at the engineering site to acquire images, real-time image data Raw is obtained. The Brown-Conrady model is used to perform preliminary image correction on the real-time image data Raw. Local contrast is optimized and image edges are enhanced through adaptive histogram equalization. The enhanced image parameters are adjusted to the preset ModelInputReq format to obtain preprocessed image data Ima. The Mask R-CNN model is used to segment the preprocessed image data Ima, identify and divide the material pile area MP and the scaffolding unit area SU, and generate the instance segmentation mask Mask and the target bounding box coordinates Box = {Xmin, Ymin, Xmax, Ymax}, where Xmin represents the minimum value of the target X coordinate, Ymin represents the minimum value of the target Y coordinate, Xmax represents the maximum value of the target X coordinate, and Ymax represents the maximum value of the target Y coordinate.

3. The intelligent engineering safety management method based on AI hazard identification according to claim 2, characterized in that: S1 also includes S12: S12. For material pile area MP, based on the instance segmentation mask Mask(MP_i) of the i-th material pile MP_i in material pile area MP, and combined with the monocular depth estimation algorithm, estimate the estimated actual height Hgt_i of material pile MP_i. Based on the projection of the instance segmentation mask Mask(MP_i) of material pile MP_i on the ground plane, estimate the bottom area Ara_i of material pile MP_i. Based on the instance segmentation mask Mask(MP_i) and the estimated actual height Hgt_i of material pile MP_i, estimate the estimated volume Vol_i of material pile MP_i through 3D reconstruction technology. Based on the target bounding box coordinates Box(MP_i) of material pile MP_i, calculate the 3D minimum bounding box volume Vbox_i of material pile i. Based on the ratio of the estimated volume Vol_i of material pile MP_i to the 3D minimum bounding box volume Vbox_i, obtain the compactness factor Cpt_i of material pile MP_i, and generate the material pile feature set FTS_MP. For the scaffolding unit area SU, based on the instance segmentation mask Mask(SU_j) of the j-th scaffolding unit SU_j in the scaffolding unit area SU, the Mask R-CNN model is used to segment and calculate the number of vertical load-bearing members Nvs_j and the number of horizontal connecting members Nhl_j. The straightness of the vertical load-bearing members and the connection point position of the horizontal connecting members of the scaffolding unit SU_j are segmented by the Mask R-CNN model and compared with the preset ideal model and the preset maximum allowable structural deformation model to evaluate the degree of structural deformation Def_j of the scaffolding unit SU_j. The material pile feature set FTS_MP and the scaffolding unit feature set FTS_SU are integrated to generate the image feature set FTS.

4. The intelligent engineering safety management method based on AI hazard identification according to claim 3, characterized in that: S2 includes S21: S21. Based on the material pile feature set FTS_MP in the image feature set FTS, and combined with the preset engineering safety standard CR, construct a stability index algorithm to calculate the material stacking stability Smp_i of the material pile MP_i. The stability index algorithm expression is as follows: In the formula, α represents the sensitivity adjustment index of the bottom area Ara_i of the material pile MP_i from the preset engineering safety standard CR, β represents the sensitivity adjustment index of the estimated actual height Hgt_i of the material pile MP_i from the preset engineering safety standard CR, and ε represents the configuration parameter. Based on the scaffolding unit feature set FTS_SU in the image feature set FTS, and combined with the preset engineering safety standard CR, a scaffolding structure defect index algorithm is constructed to calculate the scaffolding structure defect degree Dsf_j of scaffolding unit SU_j. The algorithm expression for the scaffolding structural defect index is as follows: In the formula, σ represents the influence factor of the structural deformation degree Def_j of scaffolding unit SU_j from the preset engineering safety standard CR, k represents the theoretical proportionality factor of the number of vertical load-bearing members Nvs_j and the number of horizontal connecting members Nhl_j of scaffolding unit SU_j from the preset engineering safety standard CR, and exp represents the natural exponential function.

5. The intelligent management method for engineering safety based on AI hazard identification according to claim 4, characterized in that: S3 includes S31; S31. Based on the material stacking stability Smp_i of material stack MP_i and the estimated actual height Hgt_i of material stack MP_i in the material stack feature set FTS_MP, and combined with the preset engineering safety standard CR, construct the comprehensive risk index algorithm for the material stacking area, and calculate the comprehensive risk index Rms of the material stacking area. The algorithm expression for the comprehensive risk index of the material storage area is as follows: In the formula, Nmp represents the total number of material piles in the material storage area MP, Ts_max represents the upper limit reference value of material pile stability from the preset engineering safety standard CR, Hsite represents the maximum allowable height of material stacking from the preset engineering safety standard CR, δ1 represents the stability risk contribution adjustment index from the preset engineering safety standard CR, δ2 represents the ultra-high risk contribution adjustment index from the preset engineering safety standard CR, and max(a, b) represents taking the maximum value of a and b.

6. The intelligent management method for engineering safety based on AI hazard identification according to claim 5, characterized in that: S3 also includes S32: S32. Based on the scaffolding unit SU_j, the Mask R-CNN model is used to identify and segment human entities within the preprocessed image data Ima. By comparing the number of people within the scaffolding unit SU_j in the image with the maximum number of people that the scaffolding unit SU_j can accommodate in the preset engineering safety standard CR, the personnel occupancy factor Loc_j of the scaffolding unit SU_j in the preprocessed image data Ima is obtained. Combined with the preset engineering safety standard CR, a comprehensive scaffolding risk index algorithm is constructed, and the comprehensive scaffolding risk index Rsf is calculated. The algorithm expression for the comprehensive risk index of scaffolding is as follows: In the formula, Nsu represents the total number of scaffold units in the scaffold unit area SU, w1 represents the structural defect contribution adjustment index from the preset engineering safety standard CR, and w2 represents the structural deformation contribution adjustment index from the preset engineering safety standard CR.

7. The intelligent engineering safety management method based on AI hazard identification according to claim 6, characterized in that: S4 includes S41; S41. Based on the comprehensive risk index algorithm Rms for material storage areas and the comprehensive risk index Rsf for scaffolding, calculate the basic sum of standardized risk components SC. According to the historical database HIS, obtain the average value Savg of the basic sum of standardized risk components Sc in the historical database HIS. Combining the comprehensive risk index algorithm Rms for material storage areas, the comprehensive risk index Rsf for scaffolding, the basic sum of standardized risk components SC, and the average value Savg of the basic sum of standardized risk components Sc in the historical database HIS, construct a comprehensive safety status assessment algorithm and calculate the comprehensive safety status assessment value Vto. The expression for the comprehensive security status assessment algorithm is as follows: In the formula, Rms_safe represents the upper limit of the comprehensive risk index of the material storage area from the preset engineering safety standard CR, Rsf_safe represents the upper limit of the comprehensive risk index of the scaffolding from the preset engineering safety standard CR, ζ1 represents the risk contribution adjustment index of the material storage area from the preset engineering safety standard CR, and ζ2 represents the risk contribution adjustment index of the scaffolding unit area from the preset engineering safety standard CR.

8. The intelligent management method for engineering safety based on AI hazard identification according to claim 7, characterized in that: S4 also includes S42: S42. Compare the comprehensive safety status assessment value Vto with the graded comprehensive safety status assessment threshold Fth; If the comprehensive safety status assessment value Vto is less than the low comprehensive safety status assessment threshold Fth_low, then the current status is determined to be safe, the risk level Alo is set to 0, and no alarm prompt is required. If the low comprehensive security status assessment threshold Fth_low ≤ comprehensive security status assessment value Vto < high comprehensive security status assessment threshold Fth_high, then the current status is determined to be low risk, the risk level Alo = 1 is set, a low-level alarm is issued through the APP, and security personnel are assigned to carry out rectification. If the comprehensive security status assessment value Vto is greater than or equal to the high comprehensive security status assessment threshold Fth_high, then the current status is determined to be high risk, the risk level Alo is set to 2, a high-level alarm is issued through the APP and a prompt is made to assign security personnel to carry out rectification. Based on the risk level Alo, the corresponding real-time image data Raw, pre-processed image data Ima, comprehensive risk index Rms of the material storage area, and comprehensive risk index Rsf of the scaffolding are collected to generate alarm information Air.

9. The intelligent management method for engineering safety based on AI hazard identification according to claim 8, characterized in that: S5 includes S51: S51. Safety personnel go to the site to rectify the situation based on the alarm information Air and record the feedback information Fd. They then upload the alarm information Air and the recorded feedback information Fd to the historical database His. The feedback information Fd includes the actual inspection result Er, the false alarm rate Rpr, and the false alarm rate Rur. If the false positive rate Rpr and the false negative rate Rur are less than the preset error detection rate threshold Rth, then no iterative optimization is required. If the false alarm rate Rpr and the false alarm rate Rur are greater than or equal to the preset error detection rate threshold Rth, then the data analyst will adjust the engineering safety standard CR and the graded comprehensive safety status assessment threshold Fth. Based on the adjusted engineering safety standard CR and the graded comprehensive safety status assessment threshold Fth, the accuracy of the adjusted engineering safety standard CR and the graded comprehensive safety status assessment threshold Fth was analyzed using preprocessed image data Ima from the historical database HIS.

10. An intelligent engineering safety management system based on AI hazard identification, applied to the intelligent engineering safety management method based on AI hazard identification as described in any one of claims 1-9, characterized in that: It includes an image data acquisition and processing module, a compliance assessment module, a risk assessment module, a comprehensive risk judgment module, and an iterative optimization module; The image data acquisition and processing module acquires real-time image data Raw through cameras installed at the engineering site, preprocesses the real-time image data Raw to obtain preprocessed image data Ima, and uses an AI model to perform target segmentation on the preprocessed image data Ima and extract the image feature set FTS. The compliance assessment module calculates the material stacking stability Smp and the scaffolding structure defect degree Dsf based on the image feature set FTS and the preset engineering safety standard CR. The risk assessment module calculates the comprehensive risk index Rms of the material storage area and the comprehensive risk index Rsf of the scaffolding based on the material stacking stability Smp and the scaffolding structural defect degree Dsf, respectively. The comprehensive risk assessment module calculates the comprehensive safety status assessment value Vto based on the comprehensive risk index algorithm Rms for material storage areas and the comprehensive risk index Rsf for scaffolding, combined with the historical database HIS, and compares it with the graded comprehensive safety status assessment threshold Fth. Based on the comparison results, alarm information Air is generated. The iterative optimization module allows safety management personnel to rectify on-site issues based on alarm information Air and record feedback information Fd. The alarm information Air and feedback information Fd are then uploaded and updated in the historical database HIS. Iterative optimization is performed based on the feedback information Fd and the historical database HIS.

Images