Road construction protection method, system and device

Abstract

The invention relates to the technical field of road construction safety protection, and particularly discloses a road construction protection method, system and device. The method comprises the following steps: acquiring multi-modal sensor data and performing multi-source data fusion processing to obtain an environment data sequence of a construction site; extracting feature information from the environment data sequence, and performing risk identification based on a lightweight deep learning model in the edge calculation unit to obtain a potential risk identification sequence; uploading the identification sequence to a cloud for model optimization, and generating an updated identification model parameter; and finally, the optimized model parameters are applied to an edge calculation unit, dynamic risk identification of real-time construction environment data is carried out, and a real-time early warning signal is generated. According to the invention, through fusion of the multi-modal data and real-time optimization of the deep learning model, accurate safety risk identification can be realized, false alarm or missing alarm can be effectively avoided, the safety of a construction site is improved, and potential accidents are reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of road construction safety protection, and in particular to a road construction protection method, system and device. Background Art

[0002] With the continuous advancement of urban construction and infrastructure projects, the safety management of road construction sites has received more and more attention. Traditional road construction safety protection methods mostly rely on manual inspections or simple sensor monitoring, which is difficult to cope with the complex and changing construction environment, especially in dynamically changing construction sites. There are often problems such as failure to detect potential risks in time, false alarms or missed alarms. This not only threatens the lives of construction workers, but may also cause traffic accidents or other serious consequences.

[0003] At present, with the development of artificial intelligence and Internet of Things technology, sensor-based data collection and the application of deep learning algorithms have become research directions for improving the accuracy and real-time performance of construction safety management. However, existing technologies still have certain limitations in terms of the complexity and dynamic changes of the construction environment. Especially in the process of multi-source data fusion and real-time risk identification, how to avoid false alarms or missed alarms is still a technical problem. Existing solutions often have insufficient accuracy in identifying risk factors, or cannot adapt quickly when the construction environment changes significantly, and cannot provide truly accurate early warnings.

[0004] Therefore, how to identify risks in real time based on the fusion of multimodal sensor data and combined with deep learning models has become a core technical challenge for improving road construction safety protection capabilities. The present invention aims to solve the above technical problems by accurately identifying potential safety risks in real time and effectively avoiding false alarms or missed alarms, thereby providing a safer working environment for construction workers and pedestrians. Summary of the invention

[0005] The present invention provides a road construction protection method, system and device to solve the problem of how to accurately identify potential safety risks in real time in a complex and dynamically changing road construction environment, and effectively avoid false alarms or missed alarms to ensure the safety of construction workers and pedestrians.

[0006] In order to solve the above technical problems, the present invention provides a road construction protection method, comprising: Acquire multimodal sensor data, perform multi-source data fusion processing, and obtain the environmental data sequence of the construction site; Extracting feature information from the environmental data sequence, and performing risk identification on the feature information based on a lightweight deep learning model in an edge computing unit to obtain a potential risk identification sequence; The potential risk identification sequence is sent to the cloud dynamic model optimization module to adaptively adjust and optimize the identification model to generate updated identification model parameters; The optimized identification model parameters are applied to the risk identification module of the edge computing unit to perform dynamic risk identification processing on the real-time construction environment data and generate real-time warning signals.

[0007] Furthermore, the step of acquiring multimodal sensor data specifically includes: The distance data, image data and temperature data of the construction site are collected in real time through lidar, camera and infrared sensor, and the data are synchronously marked with timestamps.

[0008] Furthermore, the step of multi-source data fusion processing specifically includes: Standardize the formats of distance data, image data and temperature data, unify the units and formats of each sensor data, and obtain a standardized data sequence; Through the weighted fusion algorithm, the data weight coefficients of each sensor are combined to generate a unified environmental data sequence.

[0009] Furthermore, the step of applying the optimized identification model parameters to the risk identification module of the edge computing unit specifically includes: The updated recognition model parameters are loaded into the edge computing unit, and the model is used to process the construction environment data collected in real time, identify potential risk factors at the construction site and generate real-time warning signals.

[0010] Furthermore, the step of extracting feature information specifically includes: Extract image features, distance change features, and temperature fluctuation features to form a preliminary feature data set; Based on the feature data set, risk identification processing is performed through a deep learning model in the edge computing unit to identify potential risk factors at the construction site.

[0011] Furthermore, the step of multi-source data fusion processing specifically includes: The distance data, image data and temperature data are formatted in a standardized manner to unify the data format, and a weighted fusion algorithm is used to combine the weight coefficients of each data source to generate an environmental data sequence of the construction site.

[0012] Furthermore, the potential risk identification sequence includes a plurality of risk categories and their corresponding occurrence times and occurrence probabilities.

[0013] Furthermore, the cloud-based dynamic model optimization module optimizes the model parameters by calculating the error function, specifically including: Through the back propagation algorithm, the model parameters are gradient updated according to the error function to generate updated recognition model parameters.

[0014] Furthermore, a road construction protection system comprises: A data acquisition module, used to collect multi-dimensional data from multi-modal sensors at the construction site; A data processing module, used for format standardization and data fusion processing of the collected data; A risk identification module, extracting characteristic information from the environmental data sequence and performing risk identification; A model optimization module, used for receiving the risk identification sequence and performing model optimization and parameter update in the cloud; The early warning module performs dynamic risk identification and processing on real-time construction environment data to generate real-time early warning signals.

[0015] Furthermore, a road construction protection device comprises: A memory, used to store instructions, model parameters and environmental data for executing the road construction protection method; A processor, used to load and execute instructions in the memory to complete data processing, risk identification and model optimization; The communication interface is used to communicate with external devices, upload the risk identification sequence of the construction site to the cloud server in real time, receive the optimization model parameters from the cloud, and trigger the early warning signal.

[0016] The invention combines the fusion of multimodal sensor data with the dynamic optimization of deep learning models to achieve real-time and accurate identification of potential safety risks in complex and dynamically changing road construction environments, and effectively avoids the problems of false alarms and missed alarms. Compared with traditional road construction safety protection methods, the invention significantly improves the safety and management efficiency of construction sites through the following innovative methods: First, the introduction of multimodal data fusion technology enables sensor data from different sources (such as lidar, camera, temperature sensor) to be processed uniformly, greatly improving the accuracy and consistency of environmental data. Compared with the traditional single data source monitoring method, this invention can more comprehensively and accurately reflect the actual situation of the construction site.

[0017] Secondly, the application of lightweight deep learning models can perform real-time data analysis and risk identification in edge computing units. The innovation of this model lies in its ability to dynamically adapt to changes in the construction environment and adjust the identification strategy in real time, thereby achieving efficient safety risk identification and avoiding the problem of traditional methods that cannot quickly respond to complex environmental changes.

[0018] Finally, by uploading the potential risk identification sequence to the cloud-based dynamic model optimization module for adaptive adjustment, the model can be kept up to date in the ever-changing construction environment, thereby improving the accuracy of risk identification and the real-time nature of early warning. The combination of cloud and edge computing makes the model optimization process more efficient and flexible, further improving the overall performance of the system.

[0019] Therefore, the beneficial effect of the present invention is that it significantly reduces potential safety hazards at construction sites through accurate real-time early warning, reduces errors and delays in human operations, and improves the safety and intelligence level of road construction. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 A schematic diagram of a road construction protection method provided in an embodiment of the present application; Figure 2 A structural block diagram of a road construction protection system provided in an embodiment of the present application; Figure 3 A structural block diagram of a road construction protection computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by technicians in the technical field of the present application; the terms used in the specification of the application herein are only for the purpose of describing specific embodiments and are not intended to limit the present application; the terms "including" and "having" and any variations thereof in the specification and claims of the present application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first" and "second" in the specification and claims of the present application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0022] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0023] Example 1: Reference Figure 1 , is a schematic flow chart of a road construction protection method provided by an embodiment of the present invention, and the process may at least include steps S100-S400: S100, acquiring multimodal sensor data, performing multi-source data fusion processing, and obtaining an environmental data sequence of the construction site; S200, extracting feature information from the environmental data sequence, and performing risk identification on the feature information based on a lightweight deep learning model in an edge computing unit to obtain a potential risk identification sequence; S300, sending the potential risk identification sequence to a cloud-based dynamic model optimization module, adaptively adjusting and optimizing the identification model, and generating updated identification model parameters; S400: Apply the optimized identification model parameters to the risk identification module of the edge computing unit, perform dynamic risk identification processing on the real-time construction environment data, and generate a real-time warning signal.

[0024] Step S100 at least includes steps S110-S130: S110: Through the multimodal sensors installed at the construction site, the distance, temperature and image data of the construction environment are collected in real time.

[0025] Specifically, the distance data is obtained through the acquisition unit of each sensor. , Temperature data and image data , where i represents the identification index of each sensor. The collected input data sources include the distance information (in meters) obtained by the lidar, the image pixel matrix of the camera, and the temperature parameters (in degrees Celsius) of the infrared sensor. The data of all sensors are synchronized with the timestamp t to ensure the time consistency of multimodal data. The data is temporarily stored in the local cache unit for standardized operations in subsequent processing links.

[0026] S120: Perform format standardization processing on data acquired from different sensors to unify the data formats.

[0027] Furthermore, the multimodal data in the cache (t), (t), (t) Enter the standardization processing unit in turn. (t), first convert the units and convert the values ​​collected by different sensors into the standard unit meter (m). (t) The unit is also Celsius (°C). Image data The pixel format of (t) is converted to a 256-level resolution matrix of grayscale levels. The standardized data set can be expressed as:

[0028] Where n represents the number of sensors. After standardization, data from different sources are guaranteed to be compatible, which is convenient for subsequent fusion processing.

[0029] S130: Perform multi-source data fusion processing on the formatted data, combine the advantages and characteristics of each sensor, and generate an environmental data sequence of the construction site for further analysis.

[0030] After completing the format standardization, the standardized data D(t), T(t), and I(t) are sent to the multi-source data fusion processing unit. Specifically, through the weighted fusion algorithm, various types of data are comprehensively processed to generate a unified environmental data sequence. The weight coefficients α, β, and γ are set to correspond to the weights of distance, temperature, and image data respectively. The environmental data sequence can be expressed as:

[0031] in, (t) (t) (t) represents the mean processing results of the distance, temperature and image data at time t, respectively. The specific calculation method is as follows:

[0032] The environmental data sequence E(t) after data fusion will be used as input for subsequent feature extraction and risk identification for further real-time analysis.

[0033] Step S200 at least includes steps S210-S230: S210: Perform a preliminary analysis on the environmental data sequence, extract image features, distance change features, and temperature anomaly information, and form a preliminary feature data set.

[0034] Based on the environmental data sequence E(t) generated by the S130 module, data analysis is performed to further obtain key feature information of the construction site. Specifically, image features are extracted from the environmental data sequence. (t), distance change characteristics (t) and temperature anomaly characteristics (t), respectively represent the gray matrix features of the image, the rate of change based on the sensor distance, and the degree of temperature fluctuation. The calculation formula for feature extraction is as follows:

[0035] in, is the image grayscale weight coefficient, (t) represents the gray value of each image pixel, (t) is the reading of each distance sensor, (t) represents the data of the temperature sensor. Through the above formula, the preliminary feature data set { (t), (t), (t)} will be used as input for the deep learning model.

[0036] S220: Input the extracted feature data into the lightweight deep learning model in the edge computing unit, perform risk identification processing on the data, and identify potential risk factors at the construction site.

[0037] The preliminary feature dataset { (t), (t), (t)} is input into the lightweight deep learning model M in the edge computing unit, which uses a multi-layer neural network structure to perform risk identification processing on the feature data. Specifically, the model will use forward propagation to calculate the output of each layer and identify potential construction environment risks based on the training parameters. The output of model M is defined as the potential risk probability vector P(t)={ , ,… }, where k represents different risk types, is the probability value of the i-th risk and satisfies:

[0038] Based on the probability value, the model determines the specific risk factors that may exist at the scene (such as approaching vehicles and pedestrians) and generates preliminary risk identification results as input for the next step.

[0039] S230: Generate a potential risk identification sequence based on the identified potential risk factors, and record the identification results locally for use by subsequent processing modules.

[0040] According to the potential risk probability vector P(t) output by the lightweight deep learning model M, a potential risk identification sequence R(t) is further generated. Specifically, by setting the threshold Determine whether the risk probability in P(t) exceeds the safety standard. ≥ , the category is determined to be a potential risk, and the risk category i and timestamp t are recorded to form a risk sequence R(t)={ , ,… },in The specific risk record at each moment includes the risk category and trigger time. The potential risk identification sequence R(t) will be stored locally for subsequent model optimization and real-time warning module processing.

[0041] Step S300 at least includes steps S310-S330: S310: Upload the locally generated potential risk identification sequence to the cloud server for further analysis by the cloud dynamic model optimization module.

[0042] The potential risk identification sequence R(t) generated by the S230 module is uploaded to the cloud server through the network interface. Specifically, the environmental data sequence E(t) and the identified risk category data R(t)={ … } as input data, and with the corresponding timestamp t for data synchronization. The key variables uploaded in this sequence include the identification results of various risks and its probability of occurrence (t), where j represents different risk types and satisfies j∈{1,2,…,m}.

[0043] S320: Perform data analysis on the received potential risk identification sequence in the cloud, and adaptively adjust and optimize the parameters of the deep learning model based on changes in the construction site environment and identification requirements.

[0044] The cloud-based dynamic model optimization module performs further analysis based on the uploaded risk identification sequence R(t) and environmental data E(t). First, the error of the identification model is measured by calculating the error function L. Specifically, the error function is defined as:

[0045] in, is the actual risk probability, (t) is the risk probability predicted by the model. Based on this error function, the parameters of the deep learning model are adjusted through the back propagation algorithm. Perform gradient update, and the updated parameters are defined as:

[0046] in is the learning rate, is the updated parameter. The model will adaptively adjust the parameters to improve recognition accuracy according to the dynamic changes of environmental data.

[0047] S330: Send the optimized identification model parameters back to the edge computing unit to locally update the deep learning model in the risk identification module.

[0048] Optimized model parameters The data is sent to the edge computing unit through the transmission module and the old model parameters are replaced locally. After loading new parameters into the edge computing unit, the deep learning model is then in the latest optimized state, enabling more accurate identification of potential risks at the construction site.

[0049] Step S400 at least includes steps S410-S430: S410: Load the cloud-optimized identification model parameters into the risk identification module in the edge computing unit to ensure that the model is in the latest optimized state.

[0050] Optimized recognition model parameters received from the cloud Transmitted to the edge computing unit through a secure data channel. Specifically, during the loading process, the integrity and consistency of the parameters are verified to prevent data corruption or loss. After the loading is completed, the risk identification module will use the latest parameters for subsequent data processing to ensure that the model can cope with changes in the real-time construction environment.

[0051] S420: Based on the updated identification model, dynamic risk identification is performed on the construction environment data collected in real time to identify new potential risk factors.

[0052] In the edge computing unit, the real-time collected construction environment data sequence E(t) is input into the updated recognition model. This data includes the latest information from multimodal sensors, such as distance changes, temperature anomalies, and image data. Specifically, the model will perform risk assessment based on the formula:

[0053] in, (t) represents the new potential risk identification result, and f is the function mapping of the deep learning model. The model analyzes real-time data to identify new potential risk factors, such as pedestrians approaching and vehicle movement, and generates a list of risk factors.

[0054] S430: Trigger a real-time warning signal based on the identified risk factors, and remind construction personnel to pay attention to site safety through sound and light alarms or remote notifications.

[0055] Once the newly identified risk factors (t) When the preset threshold condition is reached, the system immediately triggers a real-time warning signal. Specifically, the generation of the warning signal follows the following logic:

[0056] When Alert=1, the system sends an alarm to the construction personnel through the sound and light alarm and the remote notification system so that they can quickly pay attention and take necessary safety measures.

[0057] The invention combines the fusion of multimodal sensor data with the dynamic optimization of deep learning models to achieve real-time and accurate identification of potential safety risks in complex and dynamically changing road construction environments, and effectively avoids the problems of false alarms and missed alarms. Compared with traditional road construction safety protection methods, the invention significantly improves the safety and management efficiency of construction sites through the following innovative methods: First, the introduction of multimodal data fusion technology enables sensor data from different sources (such as lidar, camera, temperature sensor) to be processed uniformly, greatly improving the accuracy and consistency of environmental data. Compared with the traditional single data source monitoring method, this invention can more comprehensively and accurately reflect the actual situation of the construction site.

[0058] Secondly, the application of lightweight deep learning models can perform real-time data analysis and risk identification in edge computing units. The innovation of this model lies in its ability to dynamically adapt to changes in the construction environment and adjust the identification strategy in real time, thereby achieving efficient safety risk identification and avoiding the problem of traditional methods that cannot quickly respond to complex environmental changes.

[0059] Finally, by uploading the potential risk identification sequence to the cloud-based dynamic model optimization module for adaptive adjustment, the model can be kept up to date in the ever-changing construction environment, thereby improving the accuracy of risk identification and the real-time nature of early warning. The combination of cloud and edge computing makes the model optimization process more efficient and flexible, further improving the overall performance of the system.

[0060] Therefore, the beneficial effect of the present invention is that it significantly reduces potential safety hazards at construction sites through accurate real-time early warning, reduces errors and delays in human operations, and improves the safety and intelligence level of road construction.

[0061] Embodiment 2: Figure 2 : is a structural block diagram of a road construction protection system provided by an embodiment of the present invention. Figure 2 As shown, the system may include: The data acquisition module 10 is used to collect multi-dimensional data from multi-modal sensors at the construction site, including distance data, temperature data and image data. Through laser radar, camera and infrared sensor, key parameters in the construction environment are collected in real time to ensure the accuracy and consistency of the data. All collected data are synchronized with timestamps for subsequent standardization and fusion processing.

[0062] The data processing module 20 is used to standardize the format and perform data fusion processing on the collected data. Specifically, the data processing module unifies the format of multi-source data and filters noise to make it compatible with data from different sources, and integrates it into an environmental data sequence through a weighted fusion algorithm. The standardized and fused data is used as the input of subsequent modules to ensure the accuracy of feature extraction and risk identification.

[0063] The risk identification module 30 extracts feature information from the environmental data sequence and identifies risks based on the lightweight deep learning model in the edge computing unit. This module extracts key features such as images, distance changes, and temperature anomalies, and uses the deep learning model to identify potential risk factors, such as pedestrians or vehicles approaching. The identified potential risks are recorded locally and a risk identification sequence is generated for subsequent processing.

[0064] The model optimization module 40 is used to receive the risk identification sequence and perform model optimization and parameter update in the cloud. Specifically, after the risk identification sequence is uploaded to the cloud, the model optimization module will adaptively adjust the parameters of the deep learning model according to changes in the construction environment. The optimized model parameters are then transmitted back to the edge computing unit to ensure the real-time and accuracy of the model.

[0065] The early warning module 50 applies the model parameters optimized in the cloud to the risk identification module of the edge computing unit to perform dynamic risk identification processing of real-time construction environment data. Once a new risk factor is identified, the system immediately triggers a real-time early warning signal to remind construction personnel to pay attention to safety through sound and light alarms or remote notifications.

[0066] Beneficial effects: The present invention realizes accurate early warning in complex construction environments through multimodal data collection and fusion processing, combined with real-time optimization and dynamic recognition of deep learning models. The system can quickly respond to changes in the construction environment, reduce the risk of false alarms and missed alarms, and improve the safety and intelligent management level of the construction site.

[0067] Embodiment three: Figure 3 FIG. 2 shows a structural block diagram of a road construction protection computer device according to an embodiment of the present invention. Figure 3 As shown, the device comprises: Memory 310: used to store instructions, model parameters and environmental data for executing the road construction protection method of the present invention. Memory 310 contains multimodal data acquisition instructions, data fusion processing instructions, parameter sets of risk identification models, and cloud optimization and update instruction content. After receiving multimodal sensor data from the construction site, processor 320 can load corresponding instructions and parameters from memory 310 to perform risk identification and early warning.

[0068] Processor 320: used to load and execute instructions stored in memory 310 to complete data processing, risk identification and model optimization. Specifically, processor 320 performs the following tasks: Perform format standardization and fusion operations on multi-source data to generate a unified data sequence for the construction environment.

[0069] A lightweight deep learning model is used to analyze the fused data to identify potential risk factors at the construction site.

[0070] Through dynamic interaction between edge computing and cloud servers, optimized updates of model parameters are received and applied to keep the risk identification model up to date.

[0071] Communication interface 330: used to communicate with external devices (e.g., cloud servers or other safety management systems). Through communication interface 330, the device can upload the risk identification sequence of the construction site to the cloud server in real time, receive the optimization model parameters from the cloud, and trigger the early warning signal of the construction site when necessary. Communication interface 330 supports wireless network communication and can also be connected through a wired network to ensure the stability and security of data transmission.

[0072] If the memory 310, the processor 320 and the communication interface 330 are implemented independently, they can be interconnected via an industrial standard bus, such as a PCI bus or an EISA bus, to achieve internal data exchange. Figure 3 Only one thick line is used to represent the bus structure. Optionally, in a specific implementation, if the memory 310, the processor 320 and the communication interface 330 are integrated on a chip, the modules can communicate with each other through the internal interface to enhance the integration and data transmission efficiency.

[0073] The design of the device enables it to have the ability to process multi-dimensional data in complex construction environments. Through a risk identification model that collaborates with the cloud, it can dynamically and in real time identify and warn of potential safety risks, thereby ensuring the safety of the construction site and intelligent operations.

[0074] Obviously, the embodiments described above are only some embodiments of the present application, rather than all embodiments. The preferred embodiments of the present application are given in the accompanying drawings, but they do not limit the patent scope of the present application. The present application can be implemented in many different forms. On the contrary, the purpose of providing the embodiments is to make the understanding of the disclosure of the present application more thorough and comprehensive. Although the present application is described in detail with reference to the aforementioned embodiments, for those skilled in the art, it is still possible to modify the technical solutions recorded in the aforementioned specific implementation methods, or to effectively replace some of the technical features therein. Any effective structure made using the contents of the specification and drawings of this application, directly or indirectly used in other related technical fields, is also within the scope of patent protection of this application.

Claims

1. A road construction protection method, characterized in that: The following steps are involved: Acquire multimodal sensor data, perform multi-source data fusion processing, and obtain the environmental data sequence of the construction site; Extracting feature information from the environmental data sequence, and performing risk identification on the feature information based on a lightweight deep learning model in an edge computing unit to obtain a potential risk identification sequence; The potential risk identification sequence is sent to the cloud dynamic model optimization module to adaptively adjust and optimize the identification model to generate updated identification model parameters; The optimized identification model parameters are applied to the risk identification module of the edge computing unit to perform dynamic risk identification processing on the real-time construction environment data and generate real-time warning signals.

2. The road construction protection method according to claim 1, characterized in that: The step of acquiring multimodal sensor data specifically includes: The distance data, image data and temperature data of the construction site are collected in real time through lidar, camera and infrared sensor, and the data are synchronously marked with timestamps.

3. The road construction protection method according to claim 1, characterized in that: The multi-source data fusion processing step specifically includes: Standardize the formats of distance data, image data and temperature data, unify the units and formats of each sensor data, and obtain a standardized data sequence; Through the weighted fusion algorithm, the data weight coefficients of each sensor are combined to generate a unified environmental data sequence.

4. The road construction protection method according to claim 1, characterized in that: The step of applying the optimized identification model parameters to the risk identification module of the edge computing unit specifically includes: The updated recognition model parameters are loaded into the edge computing unit, and the model is used to process the construction environment data collected in real time, identify potential risk factors at the construction site and generate real-time warning signals.

5. The road construction protection method according to claim 1, characterized in that: The step of extracting feature information specifically includes: Extract image features, distance change features, and temperature fluctuation features to form a preliminary feature data set; Based on the feature data set, risk identification processing is performed through a deep learning model in the edge computing unit to identify potential risk factors at the construction site.

6. The road construction protection method according to claim 1, characterized in that: The multi-source data fusion processing step specifically includes: The distance data, image data and temperature data are formatted in a standardized manner to unify the data format, and a weighted fusion algorithm is used to combine the weight coefficients of each data source to generate an environmental data sequence of the construction site.

7. The road construction protection method according to claim 1, characterized in that: The potential risk identification sequence includes multiple risk categories and their corresponding occurrence times and occurrence probabilities.

8. The road construction protection method according to claim 1, characterized in that: The cloud-based dynamic model optimization module optimizes the model parameters by calculating the error function, specifically including: Through the back propagation algorithm, the model parameters are gradient updated according to the error function to generate updated recognition model parameters.

9. A road construction protection system, characterized in that: include: A data acquisition module, used to collect multi-dimensional data from multi-modal sensors at the construction site; A data processing module, used for format standardization and data fusion processing of the collected data; A risk identification module, extracting characteristic information from the environmental data sequence and performing risk identification; A model optimization module, used for receiving the risk identification sequence and performing model optimization and parameter update in the cloud; The early warning module performs dynamic risk identification and processing on real-time construction environment data to generate real-time early warning signals.

10. A road construction protection device, characterized in that: include: A memory, used to store instructions, model parameters and environmental data for executing the road construction protection method; A processor, used to load and execute instructions in the memory to complete data processing, risk identification and model optimization; The communication interface is used to communicate with external devices, upload the risk identification sequence of the construction site to the cloud server in real time, receive the optimization model parameters from the cloud, and trigger the early warning signal.

Images