Construction site potential safety hazard checking method and device

Abstract

The invention relates to a construction site potential safety hazard checking method comprising the following steps: carrying out video capture on a construction site to obtain a construction video, and carrying out target identification on the construction video through a deep learning algorithm to obtain an identification target; paying attention to the recognition target through a preset attention mechanism to obtain an attention target object, and performing deep analysis on the attention target object to obtain an analysis result; performing environment monitoring on the construction site to obtain environment monitoring data, and performing environment trend prediction on the environment monitoring data through the trained prediction model to obtain a prediction result; performing comprehensive evaluation on the analysis result and the prediction result to obtain a comprehensive evaluation result, performing potential safety hazard judgment on the construction site based on the comprehensive evaluation result, and if no potential safety hazard exists, stopping potential safety hazard checking on the construction site; the potential safety hazard can be found and early warned more comprehensively, accurately and timely, and the probability of occurrence of safety accidents is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of potential safety hazards, and in particular to a method, device, equipment and storage medium for troubleshooting potential safety hazards at a construction site. Background Art

[0002] In the current construction industry, due to the complex and changeable construction environment, a large number of operators, and complicated and frequently operated machinery and equipment, the safety risk issues at the construction site are particularly prominent. Traditional methods of troubleshooting safety hazards often rely on regular manual inspections and empirical judgments, which are not only inefficient and prone to omissions, but also insufficient in predicting potential risk trends and unable to achieve real-time monitoring and early warning. In addition, with the development of modern technology, the amount of data and information in the construction process is increasing. How to effectively use these big data for intelligent analysis and decision-making to improve the level of construction safety management has become a key issue that needs to be solved urgently. Summary of the invention

[0003] The main purpose of the present invention is to provide a method, device, equipment and storage medium for checking safety hazards at construction sites, which solve the technical problems of easy omissions and insufficient prediction of potential risk trends.

[0004] To achieve the above purpose, the present invention provides a method for troubleshooting safety hazards at a construction site, comprising the following steps: Capturing a video of the construction site to obtain a construction video, and performing target recognition on the construction video through a deep learning algorithm to obtain a recognized target; wherein the recognized target includes construction personnel and mechanical equipment; Attention is paid to the recognition target through a preset attention mechanism to obtain an attention target object, and an in-depth analysis is performed on the attention target object to obtain an analysis result; Conduct environmental monitoring on the construction site to obtain environmental monitoring data, and use the trained prediction model to predict the environmental trend of the environmental monitoring data to obtain prediction results; Comprehensively evaluate the analysis results and the prediction results to obtain a comprehensive evaluation result, and judge the safety hazards of the construction site based on the comprehensive evaluation result. If there are no safety hazards, stop checking the safety hazards of the construction site.

[0005] As a further solution of the present invention, the environmental monitoring data includes temperature data, humidity data, noise data and harmful gas data.

[0006] As a further solution of the present invention, target recognition is performed on the construction video by a deep learning algorithm to obtain the recognized target, including: Decomposing the construction video by frames through a preset deep learning algorithm to obtain video sequence frames; Preprocessing the video sequence frames to obtain preprocessed images; Input the preprocessed image into a deep learning algorithm for target recognition; wherein the deep learning algorithm includes a C3Ghost module, a YOLOv5n module, a lightweight convolution module and a PAGCP module; wherein the lightweight convolution module includes a GSConv module and a VOV-GSCSP module; Extracting target features from the preprocessed image using the C3Ghost module to obtain extracted target features; Performing target recognition on the preprocessed image through the YOLOv5n module to obtain target recognition features; Performing feature dimensionality reduction on the target recognition feature through the GSConv module in the lightweight convolution module to obtain a reduced-dimensional feature; The dimensionality reduction feature is enhanced by the VOV-GSCSP module in the lightweight convolution module to obtain an enhanced feature; The enhanced features are globally channel pruned through the PAGCP module to obtain a recognition target.

[0007] As a further solution of the present invention, the recognition target is paid attention to through a preset attention mechanism to obtain a focused target object, and the focused target object is deeply analyzed to obtain an analysis result, including: assigning coordinates to the identified target within the preprocessed image; Segmenting the identified target from the plurality of pre-processed images to obtain the contours of each target object and the position information of the target object; wherein the identified target includes a plurality of target objects; Setting a target object to be focused on, and matching and focusing on contours of each target object based on the set target object to be focused on through a preset attention mechanism to obtain the focused target object; Determine the trajectory of the corresponding target object and the dynamics of the target object based on the target object of interest and the position of the target object of interest; The trajectory of the target object and the dynamic state of the target object are input into a preset risk probability model for calculation to obtain a risk probability value corresponding to the target object; Based on the risk probability value, an in-depth analysis is performed on the corresponding target object to obtain an analysis result.

[0008] As a further solution of the present invention, the trajectory of the target object and the dynamics of the target object are input into a preset risk probability model for calculation to obtain a risk probability value corresponding to the target object, including: Performing spatiotemporal feature vectorization on the trajectory and dynamics of the target object to obtain the spatiotemporal feature vector of the target object; Performing time-varying convolution processing on the spatiotemporal feature vector of the target object to obtain a multi-scale prediction structure; Inputting the multi-scale prediction structure into a preset nonlinear chaotic mapping model to perform disturbance inference of dynamic behavior, and obtaining a probability distribution of a potential risk field; Deeply integrate the probability distribution of the potential risk field to obtain a dynamic risk assessment map of the target object; Inputting the dynamic risk assessment graph into a hierarchical Laplace transfer operator model to derive risk transfer in a high-dimensional space, and obtaining transfer trends under different risk situations; A risk-sensitive adaptive calculation is performed on the transfer trends under the different risk situations to obtain a risk probability value of the corresponding target object.

[0009] As a further solution of the present invention, the training process of the prediction model includes: Acquire environmental monitoring training data, and input the environmental monitoring training data into a preliminary prediction model, wherein the environmental monitoring training data includes temperature training data, humidity training data, noise training data, and harmful gas training data; the preliminary prediction model includes a preliminary temperature data model, a preliminary humidity data model, a preliminary noise data model, and a preliminary harmful gas model; Obtaining actual labels pre-calibrated in the environmental monitoring training data; wherein the actual labels include a first prediction result corresponding to the temperature training data, a second prediction result corresponding to the humidity training data, a third prediction result corresponding to the noise training data, and a fourth prediction result corresponding to the harmful gas training data; Inputting the temperature training data into the preliminary temperature data model to obtain a first temperature prediction result; Input the humidity training data into the preliminary humidity data model to obtain a first humidity prediction result; Inputting the noise training data into the preliminary noise data model to obtain a first noise prediction result; Inputting the harmful gas training data into the preliminary harmful gas model to obtain a first harmful gas prediction result; Compare and calculate the first temperature prediction result with the first prediction result to obtain a first loss value; Compare and calculate the first humidity prediction result and the second humidity prediction result to obtain a second loss value; Compare and calculate the first noise prediction result with the third prediction result to obtain a third loss value; Compare and calculate the first harmful gas prediction result with the fourth prediction result to obtain a fourth loss value; The model parameters of the preliminary prediction model are iteratively adjusted, and the first loss value, the second loss value, the third loss value, and the fourth loss value are minimized through loss-based optimization to obtain the trained prediction model.

[0010] As a further solution of the present invention, based on the comprehensive assessment result, the safety hazard of the construction site is judged, and if there is no safety hazard, the safety hazard investigation of the construction site is stopped, including: Based on the comprehensive assessment results, a safety hazard analysis is performed on the construction site to obtain the type of safety hazard; Assign a preset weight value to each type of safety hazard; Scoring the potential safety hazard using a preset scoring model to obtain a scoring value of the potential safety hazard; Calculate the overall safety factor of the construction site based on the weight value and score value corresponding to the safety hazard type; Compare the overall safety factor with the preset safety level threshold to determine the safety level of the construction site; If the security level is lower than the preset security level, there is no potential safety hazard; If there are no safety hazards, stop checking the safety hazards at the construction site.

[0011] The present invention also provides a construction site safety hazard detection device, comprising: A capture module is used to capture the video of the construction site, obtain the construction video, and perform target recognition on the construction video through a deep learning algorithm; wherein the targets include construction personnel and mechanical equipment; An analysis module, used to pay attention to the image corresponding to the target through a preset attention mechanism to obtain an attention image, and perform in-depth analysis on the attention image to obtain an analysis result; A prediction module is used to perform environmental monitoring on the construction site, obtain environmental monitoring data, and perform environmental trend prediction on the environmental monitoring data through a preset prediction model to obtain a prediction result; The evaluation module is used to comprehensively evaluate the analysis results and the prediction results to obtain a comprehensive evaluation result, and to judge the safety hazards of the construction site based on the comprehensive evaluation result. If there are no safety hazards, the safety hazard inspection of the construction site is stopped.

[0012] The present invention also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the above methods when executing the computer program.

[0013] The present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of any of the above-mentioned methods are implemented.

[0014] The method, device, equipment and storage medium for troubleshooting safety hazards at construction sites provided by the present invention include the following steps: capturing video of the construction site to obtain a construction video, and identifying a target in the construction video through a deep learning algorithm to obtain an identified target; paying attention to the identified target through a preset attention mechanism to obtain a focused target object, and performing in-depth analysis on the focused target object to obtain an analysis result; performing environmental monitoring on the construction site to obtain environmental monitoring data, and performing environmental trend prediction on the environmental monitoring data through a trained prediction model to obtain a prediction result; performing comprehensive evaluation on the analysis result and the prediction result to obtain a comprehensive evaluation result, and judging the safety hazards at the construction site based on the comprehensive evaluation result. If no safety hazards exist, stopping the inspection of safety hazards at the construction site, thereby solving the technical problems of easy omissions and insufficient prediction of potential risk trends, and achieving more comprehensive, accurate and timely discovery and early warning of safety hazards, thereby reducing the probability of safety accidents. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a schematic diagram of the steps of a method for troubleshooting safety hazards at a construction site in one embodiment of the present invention; Figure 2 It is a structural block diagram of a device for troubleshooting safety hazards at a construction site in one embodiment of the present invention; Figure 3 It is a schematic block diagram of the structure of a computer device according to an embodiment of the present invention.

[0016] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings in conjunction with the embodiments. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0018] like Figure 1 As shown, Figure 1 It is a schematic diagram of the steps of a method for troubleshooting safety hazards at a construction site in one embodiment of the present invention; In one embodiment of the present invention, a method for troubleshooting safety hazards at a construction site is provided, comprising the following steps: Step S1, capturing video of the construction site to obtain a construction video, and performing target recognition on the construction video through a deep learning algorithm to obtain a recognized target; wherein the recognized target includes construction personnel and mechanical equipment.

[0019] Specifically, video capture deployment: First, set up high-definition cameras in key areas of the construction site to ensure that most of the construction activities can be covered. The cameras should have the function of continuous recording or triggered recording, and can capture video images of the site uninterruptedly around the clock or under specific conditions. The video signals captured by the camera are transmitted to the data processing center or local server in real time through wired or wireless networks to ensure timely acquisition of video data. The received raw video data is preprocessed as necessary, such as noise reduction, contrast enhancement, format conversion, etc., so that subsequent algorithms can better analyze it. Apply deep learning algorithms, such as convolutional neural networks (CNN) or improved YOLO (You Only Look Once), Faster R-CNN and other target detection models, to analyze the video frames frame by frame. After training, the model can identify various objects on the construction site, especially construction personnel and mechanical equipment, and output their location coordinates, action categories, quantity and other information. The information of the identified construction personnel and mechanical equipment will be recorded and displayed by the system in categories, which is convenient for management personnel to view and trace in real time, and can also provide a basis for further data analysis.

[0020] The following technical effects can be achieved through the above steps: Through real-time video capture and target recognition, all-weather and all-round monitoring of the construction site can be achieved, ensuring the continuity and timeliness of safety management work. The deep learning algorithm has a high degree of recognition accuracy and can effectively distinguish and mark construction personnel and various mechanical equipment, thereby avoiding false alarms and missed reports and improving management efficiency. When the system identifies unsafe behaviors, such as workers not wearing safety helmets and operating mechanical equipment in violation of regulations, it can immediately issue an alarm, which helps to take corrective measures quickly and prevent accidents. The large amount of structured data collected can be used for statistical analysis to understand the distribution of personnel and the use of machinery during the construction process, providing a basis for optimizing resource allocation and formulating safety regulations. Replacing manual inspections with automated target recognition can greatly save human resources and improve work efficiency. At the same time, it can reduce safety omissions caused by human factors and promote the development of construction safety management in the direction of intelligence.

[0021] Step S2, focusing on the recognition target through a preset attention mechanism to obtain a focused target object, and performing in-depth analysis on the focused target object to obtain an analysis result.

[0022] Specifically, an attention mechanism is embedded in the deep learning model, and the attention mechanism can assign weights according to the importance of the identified target, dynamic change characteristics, or other predefined criteria. For the identified targets such as construction personnel and mechanical equipment, the attention score of each target is calculated, which is usually based on factors such as the target's characteristics, behavior, and interaction with other targets. The attention score is applied to the target, so that the deep learning model focuses more on those target objects with higher scores (i.e., more worthy of attention). According to the above attention score sorting or screening, the target objects of focus are selected, such as construction personnel with frequent high-risk behaviors, abnormally operating mechanical equipment, etc. Further deep learning analysis is performed on the selected target objects of focus, which may involve trajectory prediction, behavior pattern analysis, abnormal behavior detection, ergonomic evaluation, etc. Use advanced sequence models, spatiotemporal models, or causal models to understand the behavioral laws and potential risks of the target.

[0023] The following technical effects can be achieved through the above steps: By focusing on the most critical targets through the attention mechanism, computing resources can be reasonably allocated, important events can be analyzed preferentially, and the response speed and efficiency of the system can be improved. In-depth analysis of the focus targets can discover potential risks, such as dangerous operations, equipment failures, etc., and provide early warnings and targeted intervention suggestions. In-depth mining of the behavioral characteristics of key focus targets will help to form refined construction management strategies, improve work processes, and ensure construction safety and quality. By paying attention to and analyzing the special behaviors of construction personnel, it is possible to accurately assess whether they comply with safety regulations, and provide data support for training improvements and policy formulation. In summary, through the preset attention mechanism and in-depth analysis, accurate focus and in-depth interpretation of key elements in the construction site are achieved, the level of intelligent construction safety management is improved, and it is conducive to risk prevention and control and efficiency improvement.

[0024] Step S3, performing environmental monitoring on the construction site to obtain environmental monitoring data, and performing environmental trend prediction on the environmental monitoring data using a trained prediction model to obtain a prediction result.

[0025] Specifically, a variety of environmental monitoring equipment is arranged in the construction site, such as temperature and humidity sensors, wind speed and direction meters, noise monitors, air quality detectors, dust concentration detectors, etc., to collect various indicators of the construction site environment in real time. Each environmental monitoring device sends the monitoring data to the data processing center regularly or in real time to form a continuous environmental monitoring database. The collected original environmental monitoring data is cleaned, invalid or erroneous data is removed, and integrated into a standardized data set that can be used by the model. Using historical environmental monitoring data, the prediction model is trained through machine learning algorithms (such as time series analysis, ARIMA, LSTM, Prophet, etc.) to enable it to understand and capture the changing trends of environmental variables. The real-time environmental monitoring data is input into the trained prediction model. The prediction model predicts the trend of environmental parameters (such as temperature, humidity, wind speed, etc.) within a certain period of time in the future based on the established mathematical model and the learned rules.

[0026] The following technical effects can be achieved through the above steps: Through environmental trend prediction, possible extreme climate conditions, excessive pollution and other environmental problems can be predicted in advance, and the construction party can be notified in time to take protective measures to avoid safety accidents caused by environmental changes. Predicting future environmental conditions helps to reasonably arrange construction procedures and the use of mechanical equipment. For example, after predicting rainfall, the earth excavation, concrete pouring and other operation plans can be adjusted to reduce the impact of adverse environmental factors on project quality and progress. Adjust the configuration of construction materials, manpower and other resources based on environmental prediction to reduce waste caused by environmental factors, thereby achieving cost savings. Real-time monitoring and prediction of environmental parameters can help construction companies actively comply with environmental laws and regulations, reduce environmental pollution and maintain a good social image. Therefore, by implementing environmental monitoring and trend prediction of monitoring data, construction sites can conduct more sophisticated and efficient environmental risk management based on scientific prediction results, and effectively improve the safety and sustainability of the construction process.

[0027] Step S4, comprehensively evaluate the analysis results and the prediction results to obtain a comprehensive evaluation result, and judge the safety hazards of the construction site based on the comprehensive evaluation result. If there are no safety hazards, stop checking the safety hazards of the construction site.

[0028] Specifically, first, the analysis results of construction personnel and mechanical equipment obtained by deep learning in step S3 are organically integrated with the prediction results of environmental trends. The safety risk assessment indicators of different dimensions can be integrated by weighted summation, hierarchical analysis method, fuzzy comprehensive evaluation and other methods. Design and establish a comprehensive assessment model, which can combine the above two parts of the results, consider the correlation and impact degree between various indicators, and generate an overall score or grade that fully reflects the safety status of the construction site. According to the comprehensive assessment results calculated by the comprehensive assessment model, set the corresponding safety threshold or risk level classification standard as the basis for judging whether there are safety hazards in the construction site. If the comprehensive assessment results show that the safety risk of the construction site is within an acceptable range, it means that there are no obvious safety hazards at present. At this time, the high-frequency and large-scale safety hazard investigation actions on the construction site can be temporarily stopped, and regular monitoring and spot checks can be maintained instead.

[0029] The following technical effects can be achieved through the above steps: By comprehensively evaluating data from multiple sources, the safety status of the construction site can be reviewed from an overall perspective, overcoming the limitations of a single data source and improving the comprehensiveness and accuracy of safety hazard identification. The safety hazards originally described qualitatively are converted into quantitative indicators, making the judgment of safety hazards more objective and scientific, which is conducive to construction managers making more reasonable decisions. The focus and frequency of safety management work are dynamically adjusted according to the comprehensive evaluation results to avoid excessive investigation or ignoring potential risks, and to achieve effective use of resources and optimization of safety investment. Through the comprehensive evaluation system, problems can be discovered at the early stage of safety hazards, and timely response measures can be taken to effectively prevent the occurrence of safety accidents and improve the overall safety management level of construction projects. In short, this step achieves accurate judgment of safety hazards at the construction site through comprehensive evaluation analysis results and prediction results, promotes the intelligent upgrade of the safety management system, and helps the industry achieve a higher level of safety production.

[0030] In a specific embodiment, the environmental monitoring data includes temperature data, humidity data, noise data and harmful gas data.

[0031] In a specific embodiment, the construction video is subjected to target recognition through a deep learning algorithm to obtain a recognized target, including: Decomposing the construction video by frames through a preset deep learning algorithm to obtain video sequence frames; Preprocessing the video sequence frames to obtain preprocessed images; Input the preprocessed image into a deep learning algorithm for target recognition; wherein the deep learning algorithm includes a C3Ghost module, a YOLOv5n module, a lightweight convolution module and a PAGCP module; wherein the lightweight convolution module includes a GSConv module and a VOV-GSCSP module; Extracting target features from the preprocessed image using the C3Ghost module to obtain extracted target features; Performing target recognition on the preprocessed image through the YOLOv5n module to obtain target recognition features; Performing feature dimensionality reduction on the target recognition feature through the GSConv module in the lightweight convolution module to obtain a reduced-dimensional feature; The dimensionality reduction feature is enhanced by the VOV-GSCSP module in the lightweight convolution module to obtain an enhanced feature; The enhanced features are globally channel pruned through the PAGCP module to obtain a recognition target.

[0032] Specifically, the YOLOv5n module is used as the basic framework of the deep learning algorithm. On the basis of the network of the YOLOv5n module, the lightweight C3Ghost module is combined with the original C3 module to compress the calculation amount, model weight and size in the convolution process; the lightweight convolution module is integrated in the neck network of the YOLOv5n module, and the lightweight convolution module includes the GSConv module and the VOV-GSCSP module. These modules reduce the model parameters while enhancing the feature extraction capability; the PAGCP module is introduced in the YOLOv5n module, which can perform global channel pruning, compress the parameter amount and reduce the training overhead; video frame decomposition: first, in the process of target recognition of construction videos, accurate target recognition results can be achieved through the application of deep learning algorithms. The realization of the whole process depends on the preset deep learning algorithms, which not only include advanced model structures, but also combine multiple optimization modules to ensure the balance between computing resources and accuracy. First of all, frame decomposition of construction videos is the basic step of target recognition. By processing the video frame by frame through the preset deep learning algorithm, a video sequence frame can be obtained. These video sequence frames need to go through a preprocessing step before entering the next step of processing. The purpose of preprocessing is to clean up the noise in the image and adjust the image's brightness, contrast and other parameters so that the subsequent target recognition algorithm can process on a more consistent and standardized image basis. The preprocessed image is the core input of the entire target recognition process, and its quality directly affects the target recognition effect. When the preprocessed image is input into the deep learning algorithm for target recognition, the entire recognition process is a collaborative work of multiple modules. First, the C3Ghost module is responsible for target feature extraction from the preprocessed image. The C3Ghost module is an important innovation in lightweight convolution. Its design goal is to reduce the amount of computation and model size without sacrificing feature extraction capabilities. The C3Ghost module uses a specific convolution operation to optimize the computational process by combining the original C3 module, so that large-scale data can be processed more efficiently when extracting target features. These extracted target features are the key to the entire target recognition algorithm because they contain core information such as the shape, texture, and edges of the target. Next, after receiving these extracted target features, the YOLOv5n module performs higher-level target recognition. As the basic framework of deep learning algorithms, YOLOv5n integrates the lightweight C3Ghost module, further optimizes the calculation process, and compresses the computational complexity of the convolutional network and the weight size of the model. This allows YOLOv5n to not only maintain high-precision target recognition capabilities, but also operate effectively in resource-constrained environments. The target recognition features obtained by the YOLOv5n module contain specific target classification information and positioning information, which helps identify specific targets in the video.Afterwards, these target recognition features are passed to the lightweight convolution module, where the feature dimension reduction is first performed through the GSConv module. The GSConv module is part of the lightweight convolution, and its purpose is to reduce the dimension of the features while maintaining the main feature information, thereby reducing the computational complexity of subsequent processing. Feature dimension reduction not only reduces the computational burden of the model, but also improves processing efficiency, especially in real-time target recognition tasks, where the reduced features can be processed faster. After feature dimension reduction, the VOV-GSCSP module further enhances these reduced features. The VOV-GSCSP module combines multiple convolution operations, especially in extracting and enhancing detail features. Through this module, the reduced features are re-amplified and optimized, making the details clearer and the recognition accuracy improved. The enhanced feature is a result of multiple optimizations and processing, which represents the final feature information of the target and is ready to enter the final pruning and optimization steps. Finally, the enhanced feature enters the PAGCP module for global channel pruning. The function of the PAGCP module is to prune the channels of the convolutional network globally, remove unnecessary calculation paths and redundant parameters, thereby further compressing the size of the model and reducing the computational overhead of training and inference. Through this module, the final recognition target is formed, and the efficiency and availability of the model are guaranteed. While maintaining a high recognition rate, the pruned model can better adapt to various constraints in practical applications, such as insufficient computing resources or high real-time processing requirements. Through the close coordination and progressive steps of these steps, target recognition in the construction video is successfully achieved. The entire process fully utilizes the advantages of deep learning algorithms to ensure efficient and accurate target recognition results. Each module plays an indispensable role in the entire process and jointly promotes the final realization of the recognition target.

[0033] In a specific embodiment, the recognition target is paid attention to through a preset attention mechanism to obtain a focused target object, and the focused target object is deeply analyzed to obtain an analysis result, including: assigning coordinates to the identified target within the preprocessed image; Segmenting the identified target from the plurality of pre-processed images to obtain the contours of each target object and the position information of the target object; wherein the identified target includes a plurality of target objects; Setting a target object to be focused on, and matching and focusing on contours of each target object based on the set target object to be focused on through a preset attention mechanism to obtain the focused target object; Determine the trajectory of the corresponding target object and the dynamics of the target object based on the target object of interest and the position of the target object of interest; The trajectory of the target object and the dynamic state of the target object are input into a preset risk probability model for calculation to obtain a risk probability value corresponding to the target object; Based on the risk probability value, an in-depth analysis is performed on the corresponding target object to obtain an analysis result.

[0034] Specifically, in the preprocessed image, pixel coordinates are assigned to each identified target object, and the position index of the target on the two-dimensional image plane is established. Image segmentation technology (such as Mask R-CNN, etc.) is used to separate the contours of each target object from a series of preprocessed images, and the precise position information of each target object in each frame of the image is recorded to construct a record of the position change of the target object in the time series. The type of target object that needs special attention is set according to actual needs, and then the preset attention mechanism, such as Softmax Attention or other forms of attention modules, is used to match the contours of all target objects, and the target objects of interest that meet the set conditions are highlighted. According to the position information of the target object of interest in different preprocessed images, the position points at each moment are connected to infer the motion trajectory of the target object of interest in the video. The dynamic features of the target object of interest, such as speed, direction, posture change, etc., are analyzed. The above information can be obtained from the target trajectory. The trajectory and dynamic features of the target object of interest are input into the preset risk probability model for calculation. The model can judge the behavioral risk level of the target according to historical data and rules, and output the risk probability value corresponding to each target object of interest. Based on the calculated risk probability values, in-depth behavioral analysis and safety assessment are conducted on the target objects of concern, such as predicting possible dangerous events in the future and evaluating potential safety hazards during the construction process, ultimately obtaining detailed analysis results for each target object.

[0035] The following technical effects can be achieved through the above steps: Through the attention mechanism, the system can accurately monitor specific important targets in a targeted manner, effectively reducing the possibility of false alarms and missed alarms. Through in-depth analysis of high-risk targets, the system can give early warning signals before problems occur, providing on-site managers with opportunities for timely intervention and reducing the probability of safety accidents. It provides accurate trajectory and dynamic behavior analysis of target objects, which is conducive to optimizing resource allocation, improving work processes, and improving the overall management level of the construction process. Combined with the calculation results of the risk probability model, the construction environment can be continuously evaluated and optimized, and production safety conditions can be continuously improved to ensure the smooth progress of engineering projects and the safety of personnel.

[0036] In a specific embodiment, the trajectory of the target object and the dynamics of the target object are input into a preset risk probability model for calculation to obtain a risk probability value corresponding to the target object, including: Performing spatiotemporal feature vectorization on the trajectory and dynamics of the target object to obtain the spatiotemporal feature vector of the target object; Performing time-varying convolution processing on the spatiotemporal feature vector of the target object to obtain a multi-scale prediction structure; Inputting the multi-scale prediction structure into a preset nonlinear chaotic mapping model to perform disturbance inference of dynamic behavior, and obtaining a probability distribution of a potential risk field; Deeply integrate the probability distribution of the potential risk field to obtain a dynamic risk assessment map of the target object; Inputting the dynamic risk assessment graph into a hierarchical Laplace transfer operator model to derive risk transfer in a high-dimensional space, and obtaining transfer trends under different risk situations; A risk-sensitive adaptive calculation is performed on the transfer trends under the different risk situations to obtain a risk probability value of the corresponding target object.

[0037] Specifically, the multi-scale prediction structure organically combines information from different time and space scales to comprehensively and accurately predict the dynamic behavior of the target object; in this scheme, in order to calculate the risk probability value of the target object, the trajectory and dynamic information of the target object need to be first input into the preset risk probability model for processing. This processing process involves a series of complex steps, each of which is precisely calculated for different analysis dimensions, and the ultimate goal is to obtain the risk probability value of the target object. First, the entire calculation process starts with the spatiotemporal feature vectorization of the trajectory and dynamic data of the target object. The trajectory and dynamics of the target object are continuous data that change with time and space, and these data include the position information, speed, acceleration, etc. of the target object at different time points. In order to enable these complex spatiotemporal data to be effectively calculated in subsequent models, they must first be vectorized, that is, these dynamically changing spatiotemporal data are converted into a set of feature vectors that can be processed in mathematical models. These spatiotemporal feature vectors contain key information of the target object at different time and space positions, laying the foundation for the subsequent convolution processing of the time-varying features of the target object. Next, the spatiotemporal feature vectors obtained after vectorization will be input into the time-varying convolution processing. At this stage, the model uses a time-varying convolution kernel to perform convolution operations on these spatiotemporal feature vectors. The main function of the time-varying convolution kernel is to capture the changing laws of the target object at different time and spatial scales, so that a multi-scale prediction structure can be generated. This multi-scale prediction structure predicts the future behavior of the target object. It can simultaneously take into account the rapid changes of the target object in a short period of time and the long-term trend, thereby providing more comprehensive prediction information for subsequent risk assessment. Then, we input this multi-scale prediction structure into the preset nonlinear chaotic mapping model for perturbation inference of dynamic behavior. The nonlinear chaotic mapping model is designed to handle small perturbations in complex systems and predict the potential impact of these perturbations on the future behavior of the system. In this model, the multi-scale prediction structure is used to infer the possible future behavior path of the target object and generate a potential risk field probability distribution. This probability distribution reflects the various risk states that the target object may experience in the future, thereby providing a preliminary probabilistic basis for more in-depth risk analysis. In order to further improve the accuracy of the prediction, the potential risk field probability distribution needs to be deeply fused. The main purpose of this process is to combine different risk states to generate a more comprehensive dynamic risk assessment map. The dynamic risk assessment graph can be considered as the result of a comprehensive analysis of the possible future risk status of the target object. It integrates risk information from different time and space scales and provides an overall assessment of the future risk of the target object. Next, the dynamic risk assessment graph will be input into the hierarchical Laplace transfer operator model to derive risk transfer in high-dimensional space.The hierarchical Laplace transfer operator model is designed to process complex high-dimensional data, and on this basis, derive the transfer trend under different risk situations. This process can be understood as evaluating the possibility and path of the target object's transfer from one risk state to another in a high-dimensional space, so as to derive the evolution trend of different risk situations. Finally, the transfer trend under different risk situations is calculated by risk-sensitive adaptive calculation. The main purpose of risk-sensitive adaptive calculation is to dynamically adjust the model parameters so that the prediction results can more sensitively reflect the changes in the risk situation, thereby obtaining a more accurate risk probability value. This process combines the results of all the above steps and finally outputs a risk probability value for the target object to provide data support for relevant decisions. Through the above steps, we can see that the entire calculation process is closely linked in each link, and each step provides the necessary information and basis for the next step of calculation. Finally, through these complex calculations and models, we can obtain the risk probability value of the target object, providing strong support for risk management and decision-making.

[0038] In a specific embodiment, the training process of the prediction model includes: Acquire environmental monitoring training data, and input the environmental monitoring training data into a preliminary prediction model, wherein the environmental monitoring training data includes temperature training data, humidity training data, noise training data, and harmful gas training data; the preliminary prediction model includes a preliminary temperature data model, a preliminary humidity data model, a preliminary noise data model, and a preliminary harmful gas model; Obtaining actual labels pre-calibrated in the environmental monitoring training data; wherein the actual labels include a first prediction result corresponding to the temperature training data, a second prediction result corresponding to the humidity training data, a third prediction result corresponding to the noise training data, and a fourth prediction result corresponding to the harmful gas training data; Inputting the temperature training data into the preliminary temperature data model to obtain a first temperature prediction result; Input the humidity training data into the preliminary humidity data model to obtain a first humidity prediction result; Inputting the noise training data into the preliminary noise data model to obtain a first noise prediction result; Inputting the harmful gas training data into the preliminary harmful gas model to obtain a first harmful gas prediction result; Compare and calculate the first temperature prediction result with the first prediction result to obtain a first loss value; Compare and calculate the first humidity prediction result and the second humidity prediction result to obtain a second loss value; Compare and calculate the first noise prediction result with the third prediction result to obtain a third loss value; Compare and calculate the first harmful gas prediction result with the fourth prediction result to obtain a fourth loss value; The model parameters of the preliminary prediction model are iteratively adjusted, and the first loss value, the second loss value, the third loss value, and the fourth loss value are minimized through loss-based optimization to obtain the trained prediction model.

[0039] Specifically, environmental monitoring training data are collected, which cover historical data records of four monitoring types: temperature, humidity, noise and harmful gases. These training data are respectively input into preliminary prediction models, namely preliminary temperature data model, preliminary humidity data model, preliminary noise data model and preliminary harmful gas model. The temperature training data is input into the preliminary temperature data model to obtain a first temperature prediction result. Similarly, the humidity, noise and harmful gas training data are respectively input into the corresponding models to obtain a first humidity prediction result, a first noise prediction result and a first harmful gas prediction result. The actual labels pre-calibrated in the environmental monitoring training data are obtained, and these labels represent the real observation results of the corresponding data. The first temperature prediction result is compared with the first prediction result corresponding to the temperature training data to calculate the first loss value. The first humidity prediction result is compared with the second prediction result corresponding to the humidity training data to obtain the second loss value. Similarly, the third loss value and the fourth loss value of noise and harmful gases are calculated. Gradient descent or other optimization algorithms are used to iteratively adjust the model parameters of the preliminary prediction model according to the calculated loss values. The goal is to optimize so that the first loss value, second loss value, third loss value, and fourth loss value generated by the four prediction models when processing training data are as small as possible.

[0040] The following technical effects can be achieved through the above steps: Through repeated iterative training, the prediction results of the prediction model will be closer and closer to the actual observed values, improving the accuracy of the prediction. This method can predict multiple environmental factors at the same time, which helps to fully understand and analyze the construction environment status. The trained prediction model can predict the environmental parameters of the construction site in real time. Once abnormal changes are found, early warning can be given to reduce construction risks. According to the prediction results, construction managers can allocate resources more scientifically and reasonably, such as adjusting the construction plan according to the predicted weather conditions, and strengthening ventilation and risk elimination according to the prediction results of harmful gases. Through automated prediction and analysis, the manual processing process of environmental monitoring is simplified, and the intelligent and refined management of the construction environment is realized.

[0041] In a specific embodiment, based on the comprehensive assessment result, the safety hazard of the construction site is judged, and if there is no safety hazard, the safety hazard investigation of the construction site is stopped, including: Based on the comprehensive assessment results, a safety hazard analysis is performed on the construction site to obtain the type of safety hazard; Assign a preset weight value to each type of safety hazard; Scoring the potential safety hazard using a preset scoring model to obtain a scoring value of the potential safety hazard; Calculate the overall safety factor of the construction site based on the weight value and score value corresponding to the safety hazard type; Compare the overall safety factor with the preset safety level threshold to determine the safety level of the construction site; If the security level is lower than the preset security level, there is no potential safety hazard; If there are no safety hazards, stop checking the safety hazards at the construction site.

[0042] Specifically, a comprehensive and detailed safety hazard analysis is conducted on the construction site based on the comprehensive assessment results, and the types of safety hazards that may exist are clearly pointed out, such as improper operation of personnel, equipment failure, and poor environmental conditions. For each type of safety hazard identified, a preset weight value is assigned to it according to its importance to construction safety, the size of potential risks, and the severity of the consequences that may result. A preset scoring model is used to quantitatively score each type of safety hazard. The scoring model may consider factors such as the severity, frequency, and difficulty of rectification of the hazard, so as to obtain the score value of each safety hazard. The score values ​​of each type of safety hazard are multiplied by their corresponding weight values ​​and then summed to obtain the overall safety factor of the construction site. The overall safety factor reflects the overall safety status of the construction site. The calculated overall safety factor is compared with the preset safety level threshold to determine the safety level of the construction site. If the overall safety factor is lower than a certain safety level threshold, the construction site is considered to be at a lower safety level. If the construction site is determined to have reached the preset safety level, that is, there are no major safety hazards, then the high-intensity, large-scale safety hazard investigation work can be suspended and turned to regular daily inspections and monitoring to maintain the safety management level.

[0043] The following technical effects can be achieved through the above steps: This step realizes the scientific sorting of various safety hazards through quantitative analysis and weighting, and clarifies the focus and priority of safety management. Intensive safety hazard inspection is only carried out when there are safety hazards at the construction site, avoiding unnecessary duplication of work and waste of resources, so that safety management resources can be used efficiently. By calculating the overall safety factor and determining the safety level in real time, safety precautions can be adjusted dynamically to ensure that the construction site is always in a controllable and safe state. Through this rigorous hazard inspection and rating method, safety hazards can be discovered and eliminated earlier, truly preventing them before they happen, and effectively reducing the probability of safety accidents.

[0044] The above describes the method for troubleshooting safety hazards at a construction site in an embodiment of the present invention. The following describes the device for troubleshooting safety hazards at a construction site in an embodiment of the present invention. Figure 2 In one embodiment of the present invention, a device for checking hidden dangers of safety at a construction site comprises: The capture module 21 is used to capture the video of the construction site, obtain the construction video, and perform target recognition on the construction video through a deep learning algorithm; wherein the targets include construction personnel and mechanical equipment; An analysis module 22 is used to pay attention to the image corresponding to the target through a preset attention mechanism to obtain an attention image, and perform in-depth analysis on the attention image to obtain an analysis result; The prediction module 23 is used to perform environmental monitoring on the construction site, obtain environmental monitoring data, and perform environmental trend prediction on the environmental monitoring data through a preset prediction model to obtain a prediction result; The evaluation module 24 is used to comprehensively evaluate the analysis results and the prediction results to obtain a comprehensive evaluation result, and to judge the safety hazards of the construction site based on the comprehensive evaluation result. If there are no safety hazards, the safety hazard investigation of the construction site is stopped.

[0045] In this embodiment, for the specific implementation of each unit in the above device embodiment, please refer to the above method embodiment, which will not be repeated here.

[0046] Reference Figure 3 The present invention also provides a computer device in an embodiment, wherein the internal structure of the computer device can be as follows: Figure 3 As shown. The computer device includes a processor, a memory, a display screen, an input device system, a network interface and a database connected through a system bus. Among them, the processor designed by the computer is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the above method is implemented.

[0047] Those skilled in the art will understand that Figure 3 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0048] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the above method is implemented. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0049] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided by the present invention and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM.

[0050] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0051] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for checking safety hazards at construction sites. Features: The following steps are involved: Capturing a video of the construction site to obtain a construction video, and performing target recognition on the construction video through a deep learning algorithm to obtain a recognized target; wherein the recognized target includes construction personnel and mechanical equipment; Attention is paid to the recognition target through a preset attention mechanism to obtain an attention target object, and an in-depth analysis is performed on the attention target object to obtain an analysis result; Conduct environmental monitoring on the construction site to obtain environmental monitoring data, and use the trained prediction model to predict the environmental trend of the environmental monitoring data to obtain prediction results; Comprehensively evaluate the analysis results and the prediction results to obtain a comprehensive evaluation result, and judge the safety hazards of the construction site based on the comprehensive evaluation result. If there are no safety hazards, stop checking the safety hazards of the construction site.

2. The method for troubleshooting safety hazards at construction sites according to claim 1, Features: The environmental monitoring data includes temperature data, humidity data, noise data and harmful gas data.

3. The method for troubleshooting safety hazards at construction sites according to claim 1, Features: The construction video is subjected to target recognition through a deep learning algorithm to obtain the recognized target, including: Decomposing the construction video by frames through a preset deep learning algorithm to obtain video sequence frames; Preprocessing the video sequence frames to obtain preprocessed images; Input the preprocessed image into a deep learning algorithm for target recognition; wherein the deep learning algorithm includes a C3Ghost module, a YOLOv5n module, a lightweight convolution module and a PAGCP module; wherein the lightweight convolution module includes a GSConv module and a VOV-GSCSP module; Extracting target features from the preprocessed image using the C3Ghost module to obtain extracted target features; Performing target recognition on the preprocessed image through the YOLOv5n module to obtain target recognition features; Performing feature dimensionality reduction on the target recognition feature through the GSConv module in the lightweight convolution module to obtain a reduced-dimensional feature; The dimensionality reduction feature is enhanced by the VOV-GSCSP module in the lightweight convolution module to obtain an enhanced feature; The enhanced features are globally channel pruned through the PAGCP module to obtain a recognition target.

4. The method for troubleshooting safety hazards at construction sites according to claim 1, Features: The recognition target is paid attention to through a preset attention mechanism to obtain a focused target object, and the focused target object is deeply analyzed to obtain an analysis result, including: assigning coordinates to the identified target within the preprocessed image; Segmenting the identified target from the plurality of pre-processed images to obtain the contours of each target object and the position information of the target object; wherein the identified target includes a plurality of target objects; Setting a target object to be focused on, and matching and focusing on contours of each target object based on the set target object to be focused on through a preset attention mechanism to obtain the focused target object; Determine the trajectory of the corresponding target object and the dynamics of the target object based on the target object of interest and the position of the target object of interest; The trajectory of the target object and the dynamic state of the target object are input into a preset risk probability model for calculation to obtain a risk probability value corresponding to the target object; Based on the risk probability value, an in-depth analysis is performed on the corresponding target object to obtain an analysis result.

5. The method for troubleshooting safety hazards at construction sites according to claim 4, Features: The trajectory of the target object and the dynamic state of the target object are input into a preset risk probability model for calculation to obtain a risk probability value corresponding to the target object, including: Performing spatiotemporal feature vectorization on the trajectory and dynamics of the target object to obtain the spatiotemporal feature vector of the target object; Performing time-varying convolution processing on the spatiotemporal feature vector of the target object to obtain a multi-scale prediction structure; Inputting the multi-scale prediction structure into a preset nonlinear chaotic mapping model to perform disturbance inference of dynamic behavior, and obtaining a probability distribution of a potential risk field; Deeply integrate the probability distribution of the potential risk field to obtain a dynamic risk assessment map of the target object; Inputting the dynamic risk assessment graph into a hierarchical Laplace transfer operator model to derive risk transfer in a high-dimensional space, and obtaining transfer trends under different risk situations; A risk-sensitive adaptive calculation is performed on the transfer trends under the different risk situations to obtain a risk probability value of the corresponding target object.

6. The method for troubleshooting safety hazards at construction sites according to claim 1, Features: The training process of the prediction model includes: Acquire environmental monitoring training data, and input the environmental monitoring training data into a preliminary prediction model, wherein the environmental monitoring training data includes temperature training data, humidity training data, noise training data, and harmful gas training data; the preliminary prediction model includes a preliminary temperature data model, a preliminary humidity data model, a preliminary noise data model, and a preliminary harmful gas model; Obtaining actual labels pre-calibrated in the environmental monitoring training data; wherein the actual labels include a first prediction result corresponding to the temperature training data, a second prediction result corresponding to the humidity training data, a third prediction result corresponding to the noise training data, and a fourth prediction result corresponding to the harmful gas training data; Inputting the temperature training data into the preliminary temperature data model to obtain a first temperature prediction result; Input the humidity training data into the preliminary humidity data model to obtain a first humidity prediction result; Inputting the noise training data into the preliminary noise data model to obtain a first noise prediction result; Inputting the harmful gas training data into the preliminary harmful gas model to obtain a first harmful gas prediction result; Compare and calculate the first temperature prediction result with the first prediction result to obtain a first loss value; Compare and calculate the first humidity prediction result and the second humidity prediction result to obtain a second loss value; Compare and calculate the first noise prediction result with the third prediction result to obtain a third loss value; Compare and calculate the first harmful gas prediction result with the fourth prediction result to obtain a fourth loss value; The model parameters of the preliminary prediction model are iteratively adjusted, and the first loss value, the second loss value, the third loss value, and the fourth loss value are minimized through loss-based optimization to obtain the trained prediction model.

7. The method for troubleshooting safety hazards at construction sites according to claim 1, Features: Based on the comprehensive assessment results, the safety hazard of the construction site is judged. If there is no safety hazard, the safety hazard investigation of the construction site is stopped, including: Based on the comprehensive assessment results, a safety hazard analysis is performed on the construction site to obtain the type of safety hazard; Assign a preset weight value to each type of safety hazard; Scoring the potential safety hazard using a preset scoring model to obtain a scoring value of the potential safety hazard; Calculate the overall safety factor of the construction site based on the weight value and score value corresponding to the safety hazard type; Compare the overall safety factor with the preset safety level threshold to determine the safety level of the construction site; If the security level is lower than the preset security level, there is no potential safety hazard; If there are no safety hazards, stop checking the safety hazards at the construction site.

8. A device for checking safety hazards at construction sites. Features: include: A capture module is used to capture the video of the construction site, obtain the construction video, and perform target recognition on the construction video through a deep learning algorithm; wherein the targets include construction personnel and mechanical equipment; An analysis module, used to pay attention to the image corresponding to the target through a preset attention mechanism to obtain an attention image, and perform in-depth analysis on the attention image to obtain an analysis result; A prediction module is used to perform environmental monitoring on the construction site, obtain environmental monitoring data, and perform environmental trend prediction on the environmental monitoring data through a preset prediction model to obtain a prediction result; The evaluation module is used to comprehensively evaluate the analysis results and the prediction results to obtain a comprehensive evaluation result, and to judge the safety hazards of the construction site based on the comprehensive evaluation result. If there are no safety hazards, the safety hazard inspection of the construction site is stopped.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program. It is characterized in that When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, It is characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images