Method and system for carrying out edge protection risk intelligent early warning by utilizing Transform

Abstract

The invention relates to the technical field of building construction safety management, in particular to a method and a system for carrying out edge protection risk intelligent early warning by utilizing Transform. With the rapid development of the artificial intelligence technology, the deep learning model obtains significant results in the fields of image recognition, data analysis and the like. As a novel deep learning architecture, Transform is widely applied to multiple fields of natural language processing, computer vision and the like due to the powerful self-attention mechanism and the long sequence data processing capacity of Transform. According to the method, the Transform technology is innovatively introduced into the field of building construction edge protection risk early warning, and the method aims to utilize the unique advantages of the Transform technology to realize deep analysis of construction site personnel behaviors and environment states, so that the intelligent level of edge protection is improved, and safety accidents such as falling are effectively prevented and reduced.

Description

Technical Field

[0001] This invention relates to the field of construction safety management technology, and in particular to a method and system for intelligent early warning of edge protection risks using Transformer, which can realize intelligent and accurate early warning of risks to workers operating near edges. Background Technology

[0002] During construction, working near edges always presents a high level of safety risk. Locations such as the edges of high-rise buildings, the perimeter of deep foundation pits, and the outer edges of scaffolding can lead to serious injuries or fatalities if workers slip and fall. Traditional construction safety management relies primarily on physical protective facilities and manual inspections for edge protection. This approach is not only inefficient but also lacks real-time and accurate risk warnings. With the increasing complexity of construction environments and the ever-increasing demands for construction safety, there is an urgent need for a more efficient and accurate intelligent early warning method and system.

[0003] Existing technologies typically utilize drones, cameras, LiDAR, and sensor data acquisition techniques for data collection. The data is then preprocessed, including denoising, filtering, calibration, and alignment. Image processing, machine learning, and deep learning techniques are then used to extract features from the data. This preprocessed data is then used as training and testing data to generate predictive models for risk forecasting. However, the traditional predictive models employed in these methods suffer from insufficient reliability and low efficiency in risk prediction. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by providing a method and system for intelligent early warning of edge protection risks using Transformer. Transformer technology, a breakthrough in artificial intelligence in recent years, is based on a self-attention mechanism, effectively capturing long-distance dependencies in sequences and possessing powerful parallel computing capabilities and efficient feature extraction. It exhibits excellent performance in multiple fields such as natural language processing and computer vision. Introducing Transformer technology into intelligent early warning of edge protection risks can improve reliability and efficiency, providing more reliable protection for construction safety management.

[0005] In a first aspect, the present invention provides a method for intelligent early warning of edge protection risks using Transformer, comprising: The data acquisition layer is used to collect data from the construction site, read and parse the data, and use timestamp technology to accurately mark the time of the collected data in order to collect data synchronously. The data preprocessing layer is used to clean, reduce noise, interpolate, and normalize the collected data to make it suitable for subsequent model processing. The Transformer model feature extraction layer is used for target feature recognition, spatiotemporal feature fusion, feature optimization and dimensionality reduction. The Transformer model's processing layer utilizes self-attention mechanisms and deep learning capabilities to analyze the extracted features and predict worker behavior and potential risks.

[0006] Furthermore, target feature recognition includes the following steps: S11: Background Modeling. This technique removes static background interference, allowing for a focus on capturing dynamic targets. S12: Target Detection and Tracking. This section employs a Transformer-based target detection algorithm to perform end-to-end target detection, quickly locating the worker's position. A Transformer-based multi-target tracking algorithm is used, employing a self-attention mechanism to model the relationships between different targets. S13: Keypoint recognition, using a Transformer-based pose estimation framework to identify human keypoints.

[0007] Furthermore, spatiotemporal feature fusion includes the following steps: S21: Time series analysis. The Transformer architecture can model both time and space dimensions simultaneously through a self-attention mechanism, learning spatiotemporal features from raw data. S22: Multimodal data integration, which integrates multi-source data including two-dimensional image information, three-dimensional spatial information provided by lidar, and distance measurement results, to further enrich feature representation and correct position errors in two-dimensional images.

[0008] Furthermore, feature optimization and dimensionality reduction include the following steps: S31: Feature selection, a feature selection method based on Transformer, retains the most representative parts by analyzing the self-attention weights, while reducing the feature dimensionality.

[0009] S32: Noise filtering, a Transformer-based denoising model that processes noisy feature sequences affected by environmental factors.

[0010] Furthermore, the Transformer model processing layer includes: The input layer receives multidimensional time-series data from the feature extraction layer. The encoding layer uses one or more Transformer layers to build a deep neural network, with each layer containing multiple multi-head attention mechanisms and feedforward neural networks; The decoding layer is used to map the features output by the Transformer layer to specific prediction results; The output layer includes a classifier or a regressor, whereby the classifier is used to determine whether dangerous behavior will occur, and the regressor is used to predict specific location coordinates or other continuous variables.

[0011] Furthermore, the intelligent early warning method for edge protection risks using Transformer also includes an early warning response layer for risk classification assessment and multi-channel early warning.

[0012] Secondly, the present invention provides an intelligent early warning system for edge protection risks using Transformer, which employs the method described in the present invention, including: A data acquisition module is used to collect data from the construction site, read and parse the data, and use timestamp technology to accurately mark the time of the collected data to synchronize the data collection. The module is a combination of at least two data acquisition modules or devices.

[0013] The data preprocessing module is used to clean, reduce noise, interpolate, and normalize the collected data to make it suitable for subsequent model processing. The Transformer model feature extraction module is used for target feature recognition, spatiotemporal feature fusion, feature optimization and dimensionality reduction. The Transformer model processing module uses self-attention mechanism and deep learning capabilities to analyze the data after feature extraction and predict worker behavior and potential risks.

[0014] Furthermore, the intelligent early warning system for edge protection risks using Transformer also includes an early warning response module for risk classification assessment and multi-channel early warning.

[0015] Thirdly, embodiments of this application provide an electronic device, including: a memory storing executable program code, and a processor coupled to the memory, wherein the processor calls the executable program code stored in the memory to execute the method of embodiments of this application.

[0016] Fourthly, embodiments of this application provide a computer storage medium storing a computer program, which is executed by a processor to perform the methods described in the embodiments of this application. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 These are the Transformer model training steps of this invention.

[0019] Figure 2 This is a schematic diagram of the intelligent early warning method for edge protection risks of the present invention.

[0020] Figure 3 This is a schematic diagram of the feature extraction layer of the Transformer model of the present invention.

[0021] Figure 4 This is a schematic diagram of the Transformer model processing layer of the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0023] This invention, based on existing technology, utilizes the output of the Transformer model, combined with preset risk assessment indicators and grading standards, to evaluate the risks of workers operating near edges. When a high-risk situation is detected, early warning signals are promptly issued through multiple channels to notify relevant personnel to take measures to prevent accidents.

[0024] like Figure 1 The Transformer model training process shown includes the following steps: S01. Data Preparation: We collected a large amount of video data covering various operational scenarios and manually annotated it, marking keyframes and behavior labels. We ensured the dataset covered different weather conditions, lighting environments, and construction stages to enhance the model's generalization ability.

[0025] S02, Training Strategy: The transformer model is trained using a labeled dataset, and the weight parameters are adjusted and the loss function is optimized using the backpropagation algorithm. To prevent overfitting, techniques such as Dropout regularization and layer normalization can be introduced to ensure the model's stability and generalization ability.

[0026] S03, Hyperparameter Tuning: Choosing appropriate combinations of hyperparameters, including learning rate, batch size, number of transformer layers and heads per layer, and hidden layer dimensions, can help find the optimal configuration. This reduces wasted computational resources and tuning time.

[0027] Considering the dynamic changes at construction sites, the model needs to have rapid response capabilities. To this end, the model can be compressed and optimized before deployment, using pruning to remove unimportant connections and parameters, and model quantization techniques to reduce computational load and memory consumption, thereby improving inference speed. In addition to accuracy, metrics such as recall and F1 score should also be monitored to ensure no potential risks are missed. The model should be regularly updated and retrained, continuously improving performance using newly collected data. Simultaneously, an online learning mechanism should be introduced to enable the model to adapt in real-time to the constantly changing environment and worker behavior patterns at the construction site.

[0028] The trained transformer model is integrated into the entire early warning system, interacting with other modules through a message queue mechanism. This ensures the model can run stably in real-world environments and seamlessly integrate with existing safety management systems, such as monitoring cameras, sensor networks, and safety management software at construction sites, enabling real-time data acquisition, analysis, and early warning capabilities. The following are the specific design and implementation details of the model: like Figure 2 The present invention provides a method for intelligent early warning of edge protection risks using Transformer, as shown in the figure. Figure 1 The diagram includes: a data acquisition layer 101, a data preprocessing layer 102, a Transformer model feature extraction layer 103, and a Transformer model processing layer 104.

[0029] The data acquisition layer 101 is used to collect various types of data from the construction site, read and parse the data, and use timestamp technology to accurately mark the time of the collected data to ensure data consistency through synchronous data collection. The data includes multi-source data such as video image data, 3D point cloud maps, and sensor data collected by high-definition cameras, lidar sensors, or wearable devices such as accelerometers and gyroscopes, and this invention is not limited to these sources.

[0030] The data preprocessing layer 102 is used to clean, denoise, interpolate, and normalize the collected data to make it suitable for subsequent model processing.

[0031] Specifically, for video image data, the raw data is cleaned to remove video noise, improve the contrast and clarity of the image, and make the details in the image clearer, providing a better data foundation for subsequent image-based feature extraction and target recognition tasks.

[0032] For sensor data, preprocessing mainly relies on filtering algorithms, including mean filtering and median filtering.

[0033] The mean filtering method replaces the current pixel value by calculating the average value of neighboring pixels, thereby achieving the purpose of smoothing the image and removing noise. The median filtering method selects the median of neighboring pixels as the current pixel value, demonstrating excellent performance in removing impulse noise such as salt-and-pepper noise. Both filtering algorithms effectively remove noise interference from sensor data, improving data reliability.

[0034] To address the issue of missing data values, interpolation methods are used for filling in the missing values, including linear interpolation and spline interpolation.

[0035] The linear interpolation method estimates missing values ​​based on adjacent known data points using a linear function. It is simple, efficient, and has a good filling effect when the data changes relatively steadily. The spline interpolation method uses spline functions to fit known data points to obtain a smoother curve, and then calculates the missing values, which is more advantageous when dealing with complex data variations.

[0036] After processing different types of data, it is necessary to normalize these data. The purpose of normalization is to ensure that different types of data have the same dimensions and value range, avoiding the model's oversensitivity or neglect of certain features during training due to differences in data dimensions and value ranges. This facilitates subsequent model processing, effectively improves the model's training efficiency and accuracy, allows the model to better learn the features and patterns in the data, and enhances the model's performance.

[0037] like Figure 3 The Transformer model feature extraction layer 103 shown is mainly used for target feature recognition, spatiotemporal feature fusion, feature optimization and dimensionality reduction.

[0038] Furthermore, target feature recognition includes the following steps: S11, Background Modeling: First, static background interference is removed using background modeling techniques, allowing us to focus on capturing dynamic targets. The Transformer architecture, combined with a self-attention mechanism, can effectively analyze spatial information in videos, separating foreground targets from the background and enabling accurate identification even in complex scenes.

[0039] S12, Target Detection and Tracking: Employing Transformer-based target detection algorithms (such as DETR), this approach breaks away from the limitations of traditional target detection methods that rely on manually designed anchor boxes, enabling direct end-to-end target detection and rapid worker location. Furthermore, a Transformer-based multi-target tracking algorithm utilizes a self-attention mechanism to model the relationships between different targets, continuously tracking their trajectories and maintaining tracking continuity even after brief occlusion.

[0040] S13, Key Point Recognition: By utilizing Transformer-based pose estimation frameworks, such as self-attention-based model structures, key points on the human body (e.g., head, shoulders, elbows, wrists) can be identified. This method can fully explore the long-distance dependencies between human joints, providing richer behavioral descriptions and helping to determine whether a worker is in a dangerous posture or movement.

[0041] Furthermore, spatiotemporal feature fusion includes the following steps: S21. Time Series Analysis: Considering that worker movement is a continuous process, the Transformer architecture can model both time and space dimensions simultaneously through a self-attention mechanism. Unlike traditional methods, it does not require calculating parameters such as displacement, velocity, and acceleration separately to construct time series features. Instead, it learns spatiotemporal features directly from the raw data to prepare for subsequent model inputs.

[0042] S22, Multimodal Data Integration: If the system integrates LiDAR or other depth sensing devices, Transformer can leverage its powerful feature fusion capabilities to fuse two-dimensional image information with three-dimensional spatial information and distance measurement results provided by LiDAR, further enriching the feature representation and correcting positional errors in the two-dimensional image.

[0043] Further feature optimization and dimensionality reduction include the following steps: S31. Feature Selection: During training, the Transformer model's self-attention mechanism automatically focuses on important features. However, to further improve efficiency, it can be combined with Transformer-based feature selection methods. By analyzing self-attention weights and other techniques, the most representative features can be retained while reducing feature dimensionality, thus improving model training efficiency.

[0044] S32, Noise Filtering: Transformer-based denoising models can leverage their ability to understand data context to process noisy feature sequences influenced by environmental factors. By capturing and removing noise patterns from feature sequences through a self-attention mechanism, the accuracy of subsequent analysis can be improved.

[0045] like Figure 4 The Transformer model processing layer 104, as shown, utilizes self-attention and deep learning capabilities to analyze the data after feature extraction, predicting worker behavior and potential risks. Specifically, it includes: Input layer 201: It receives multidimensional time-series data from the feature extraction layer as input. This data includes, but is not limited to, the worker's position coordinates, speed, orientation, and key point information of the human body.

[0046] Coding layer 202: A deep neural network is constructed using one or more Transformer layers. Transformer layers utilize a self-attention mechanism, which captures dependencies between different positions in the input sequence, eliminating the need for sequential processing like recurrent neural networks (RNNs). This allows for more efficient handling of long sequence data and complex spatiotemporal relationships. Each layer contains multiple multi-head attention mechanisms and a feed-forward network; stacking multiple Transformer layers further enhances the model's expressive power.

[0047] Decoding layer 203: Located after the encoding layer, it is used to map the features output by the Transformer layer to specific prediction results, such as the position coordinates at the next moment or the probability value of whether there is a risk of falling.

[0048] Output layer 204: Depending on the specific application scenario, the output layer can be designed as a classifier or a regressor. The classifier is used to determine whether a dangerous behavior will occur (e.g., 0 - no risk, 1 - risky), while the regressor predicts specific location coordinates or other continuous variables.

[0049] Furthermore, this intelligent early warning method for border protection risks also includes an early warning response layer for risk classification assessment and multi-channel early warning.

[0050] For risk grading and assessment, based on the prediction results of the Transformer model, the early warning response module first categorizes each detected behavior into three levels: low, medium, and high. For low-risk behaviors, the system only records the information without issuing an alarm; while for medium- and high-risk behaviors, the corresponding early warning measures are triggered immediately.

[0051] By combining parameters such as the worker's specific location, movement speed, and direction, the risk assessment criteria are further refined. Considering the worker's location information, such as distance from the edge, the risk level increases when the worker's distance from the edge is less than a certain threshold. Analyzing the worker's motion state, including speed, acceleration, and direction, the risk level increases if the worker approaches the edge at a high speed. Attention is also paid to the worker's posture information, such as body tilt angle and center of gravity position; when the worker's posture is unstable and they are close to the edge, it is considered high risk. By comprehensively considering these factors, the risk level is divided into three levels: low, medium, and high. When a worker approaches the edge at a speed exceeding a certain threshold, even if the distance is still considerable, it should be considered high risk. Low risk indicates that the worker's behavior is in a relatively safe state, and the system only records data; medium risk means that the worker's behavior has a certain potential risk, and the system issues a warning; high risk indicates that the worker is on the verge of danger and may fall at any time, and the system immediately triggers emergency warning measures.

[0052] Multi-channel early warning systems can include measures such as sound alarms, visual cues via SMS / app notifications, and emergency braking. Considering the potential differences in work habits and personal preferences among workers, personalized early warning settings are provided, along with trend analysis based on historical data. This allows for the anticipation of potential new risks and the proactive deployment of corresponding preventative measures.

[0053] Based on the same general inventive concept, this invention also protects a system for intelligent early warning of edge protection risks using Transformer, which can employ the above-mentioned method for intelligent early warning of edge protection risks using Transformer, specifically including: The data acquisition module is used to collect various types of data from the construction site, read and parse the data, and use timestamp technology to accurately mark the time of the collected data to ensure data consistency through synchronous data collection. The module is a combination of at least two types of modules or devices.

[0054] The data preprocessing module is used to clean, reduce noise, interpolate, and normalize the collected data to make it suitable for subsequent model processing.

[0055] The Transformer model feature extraction module is used for target feature recognition, spatiotemporal feature fusion, feature optimization and dimensionality reduction.

[0056] The Transformer model processing module is used to analyze the data after feature extraction using self-attention mechanism and deep learning capabilities to predict worker behavior and potential risks.

[0057] Furthermore, the intelligent early warning system for border protection risks also includes an early warning response module for risk classification assessment and multi-channel early warning.

[0058] The present invention discloses an electronic device, comprising: a memory storing executable program code; a processor coupled to the memory; the processor calling the executable program code stored in the memory to execute a method for intelligent early warning of edge protection risks using Transformer.

[0059] This invention also protects a computer storage medium storing a computer program that is executed by a processor to provide an intelligent early warning method for edge protection risks using Transformer.

[0060] The computer storage medium of this invention can be any combination of one or more computer-readable media. A computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. For example, a computer-readable storage medium can be, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0061] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, capable of sending, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device.

[0062] The program code contained on a computer-readable medium may be transmitted using any suitable medium, including, but not limited to, wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0063] Note that the above description is merely a preferred embodiment of the present invention and the technical principles employed. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions can be made without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in detail through the above embodiments, the present invention is not limited to the above embodiments, and may include many other equivalent embodiments without departing from the concept of the present invention, all of which fall within the scope of protection of the present invention.

Claims

1. A method for intelligent early warning of edge protection risks using Transformer, characterized in that, include: The data acquisition layer is used to collect data from the construction site, read and parse the data, and use timestamp technology to accurately mark the time of the collected data in order to collect data synchronously. The data preprocessing layer is used to clean, reduce noise, interpolate, and normalize the collected data to make it suitable for subsequent model processing. The Transformer model feature extraction layer is used for target feature recognition, spatiotemporal feature fusion, feature optimization and dimensionality reduction. The Transformer model processing layer is used to analyze the data after feature extraction using self-attention mechanism and deep learning capabilities, and to predict worker behavior and potential risks.

2. The method for intelligent early warning of edge protection risks using Transformer according to claim 1, characterized in that, The target feature identification includes the following steps: S11: Background Modeling. This technique removes static background interference, allowing for a focus on capturing dynamic targets. S12: Target Detection and Tracking. This section employs a Transformer-based target detection algorithm to perform end-to-end target detection, quickly locating the worker's position. A Transformer-based multi-target tracking algorithm is used, employing a self-attention mechanism to model the relationships between different targets. S13: Keypoint recognition, using a Transformer-based pose estimation framework to identify human keypoints.

3. The method for intelligent early warning of edge protection risks using Transformer according to claim 1 or 2, characterized in that, The spatiotemporal feature fusion includes the following steps: S21: Time series analysis. The Transformer architecture can model both time and space dimensions simultaneously through a self-attention mechanism, learning spatiotemporal features from raw data. S22: Multimodal data integration, which integrates multi-source data including two-dimensional image information, three-dimensional spatial information provided by lidar, and distance measurement results, to further enrich feature representation and correct position errors in two-dimensional images.

4. The method for intelligent early warning of edge protection risks using Transformer according to any one of claims 1-3, characterized in that, The feature optimization and dimensionality reduction include the following steps: S31: Feature selection, a feature selection method based on Transformer, retains the most representative parts by analyzing the self-attention weights, while reducing the feature dimensionality. S32: Noise filtering, a Transformer-based denoising model that processes noisy feature sequences affected by environmental factors.

5. The method for intelligent early warning of edge protection risks using Transformer according to any one of claims 1-4, characterized in that, The Transformer model processing layer includes: The input layer receives multidimensional time-series data from the feature extraction layer. The encoding layer uses one or more Transformer layers to build a deep neural network, with each layer containing multiple multi-head attention mechanisms and feedforward neural networks; The decoding layer is used to map the features output by the Transformer layer to specific prediction results; The output layer includes a classifier or a regressor, whereby the classifier is used to determine whether dangerous behavior will occur, and the regressor is used to predict specific location coordinates or other continuous variables.

6. The method for intelligent early warning of edge protection risks using Transformer according to any one of claims 1-5, characterized in that, It also includes an early warning and response layer, used for risk classification and assessment, and multi-channel early warning.

7. A system for intelligent early warning of edge protection risks using Transformer, comprising the method for intelligent early warning of edge protection risks using Transformer as described in any one of claims 1 to 6, characterized in that, include: A data acquisition module is used to collect data from the construction site, read and parse the data, and use timestamp technology to accurately mark the time of the collected data to synchronize the data collection. The module is a combination of at least two data acquisition modules or devices. The data preprocessing module is used to clean, reduce noise, interpolate, and normalize the collected data to make it suitable for subsequent model processing. The Transformer model feature extraction module is used for target feature recognition, spatiotemporal feature fusion, feature optimization and dimensionality reduction. The Transformer model processing module is used to analyze the data after feature extraction using self-attention mechanism and deep learning capabilities to predict worker behavior and potential risks.

8. The intelligent early warning system for edge protection risks using Transformer according to claim 7, characterized in that, It also includes an early warning response module for risk classification assessment and multi-channel early warning.

9. An electronic device, comprising: A memory storing executable program code, a processor coupled to the memory, the processor calling the executable program code stored in the memory, characterized in that it executes the method for intelligent early warning of edge protection risks using Transformer as described in any one of claims 1 to 6.

10. A computer storage medium storing a computer program, which, when executed by a processor, performs the method for intelligent early warning of edge protection risks using Transformer as described in any one of claims 1 to 6.

Images