Power construction monitoring method, system, equipment and medium

Abstract

The invention relates to a power construction monitoring method, system and device and a medium, and the method comprises the steps: carrying out the space-time calibration and anti-interference processing through collected construction images, temperature and dust data, and carrying out the fusion to generate a multi-source fusion data set containing environment information; performing feature extraction on the multi-source fusion data set based on a neural network model, synchronously identifying an action track of a constructor and equipment operation parameters, and constructing a feature map set associated with behaviors and equipment states; performing action and equipment anomaly conjoint analysis on the feature map set, and generating a construction risk event set by analyzing human body joints and temperature feature detection; performing spatial distance calculation and time monitoring on the risk event, and quantifying the coupling risk level; and based on the dynamic risk threshold, a grading early warning instruction is triggered, and precise response of field personnel warning, equipment power failure and evacuation guidance is realized. According to the method, the technical problems of low construction risk identification precision, equipment anomaly detection lagging and risk assessment deficiency in a complex environment are solved.

Description

Technical Field

[0001] The present invention relates to the field of construction safety technology, and in particular to a method, system, equipment and medium for monitoring electric power construction. Background Art

[0002] Safety management and control of power construction is crucial for ensuring the stable operation of power systems. This is especially true in scenarios like substation construction and transmission line installation, where complex construction sites and frequent high-risk operations make it difficult to achieve all-weather, all-scenario risk monitoring using traditional manual inspection methods. With the advancement of computer vision and the Internet of Things (IoT) technologies, intelligent construction supervision methods based on video surveillance are gaining adoption. By analyzing construction worker behavior and equipment status in real time, they can effectively prevent safety incidents such as falls from climbing and improper operation of live equipment.

[0003] However, existing power construction monitoring technology still has significant flaws: First, conventional video surveillance systems lack robustness to environmental interference such as lighting changes and equipment occlusion, resulting in distorted feature extraction and frequent misjudgments and missed detections. Second, the identification of high-risk actions by construction workers relies on simple trajectory tracking algorithms, making it difficult to fine-tune modeling of complex behaviors such as climbing postures and tool compliance. Furthermore, the detection sensitivity of electrical equipment anomalies is insufficient, and subtle features are easily obscured by noise, making it difficult to trigger early warnings. More critically, existing methods lack the ability to analyze the spatiotemporal correlation between human behavior and equipment anomalies. Risk events are often treated in isolation, making it impossible to accurately assess the comprehensive risk level caused by the coupling of multiple factors, resulting in a lag in the triggering of emergency response commands. Summary of the Invention

[0004] Based on this, the purpose of the present invention is to provide a power construction monitoring method, system, equipment and medium that can adapt to complex construction environments, accurately identify high-risk behaviors and equipment anomalies, and support dynamic risk level assessment.

[0005] The purpose of the present invention is achieved by the following scheme:

[0006] In a first aspect, the present invention provides a method for monitoring electric power construction, comprising the following steps:

[0007] S1: Perform time-space position calibration and anti-interference processing on the construction image data, temperature data, and dust concentration data collected by multi-source sensors at the construction site to eliminate sensor acquisition delays, compensate for spatial perspective differences, and suppress noise, generating a pre-processed multi-source fusion dataset.

[0008] S2: Perform multi-level feature extraction on the multi-source fusion dataset based on a neural network model to identify the movement trajectories of construction workers and the operating parameters of equipment, and generate a feature graph set including behavioral feature graphs and equipment status feature graphs;

[0009] S3: Identify high-risk actions and equipment anomalies based on the feature graph collection, analyze human joint movement patterns and tool usage compliance, and detect abnormal equipment temperature changes to generate a construction risk collection containing dangerous behavior events and equipment failure events.

[0010] S4: Perform spatiotemporal correlation analysis on the construction risk set. By calculating the event distance in three-dimensional space and combining it with continuous monitoring in the time dimension, a risk level assessment result is generated. The risk level assessment result is used to quantify the comprehensive accident probability level within the construction area.

[0011] S5: Perform graded warning processing based on the risk level assessment results, trigger different levels of emergency response instructions through preset risk thresholds, generate and transmit comprehensive warning notifications to the construction unit's management terminal, and the emergency response instructions include one of the following: on-site personnel warning information push signal, equipment power cut-off instruction and evacuation route guidance signal.

[0012] In a second aspect, the present invention provides a power construction monitoring system, which is configured with the following modules:

[0013] The data acquisition preprocessing module is used to perform time-space position calibration and anti-interference processing on the construction image data, temperature data, and dust concentration data collected by multi-source sensors at the construction site, eliminate sensor acquisition delays, compensate for spatial perspective differences, and suppress noise, thereby generating a preprocessed multi-source fusion data set;

[0014] The multi-source feature extraction module is used to perform multi-level feature extraction processing on the multi-source fusion data set based on the neural network model, identify the movement trajectory of construction workers and the operating parameters of equipment, and generate a feature map set including behavioral feature maps and equipment status feature maps;

[0015] The abnormal behavior recognition module is used to identify high-risk actions and equipment anomalies in feature graphs, analyze the movement patterns of human joints and the compliance of tool use, and generate a construction risk set containing dangerous behavior events and equipment failure events by detecting abnormal temperature changes and electrical spark characteristics of equipment;

[0016] The comprehensive risk assessment module is used to perform spatiotemporal correlation analysis on the construction risk set. By calculating the event distance in three-dimensional space and combining it with continuous monitoring in the time dimension, it generates risk level assessment results. The risk level assessment results are used to quantify the comprehensive accident probability level within the construction area.

[0017] The emergency response and early warning module is used to perform graded early warning processing based on the risk level assessment results, trigger different levels of emergency response instructions through preset risk thresholds, generate and transmit comprehensive early warning notifications to the construction unit's management terminal. Emergency response instructions include one of the following: on-site personnel warning information push signals, equipment power cut-off instructions, and evacuation route guidance signals.

[0018] In a third aspect, the present application provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements any of the above-mentioned power construction monitoring methods when executing the computer program.

[0019] In a fourth aspect, the present application provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, any one of the above-mentioned power construction monitoring methods is implemented.

[0020] In summary, the power construction monitoring method provided by the present invention can effectively suppress sudden changes in illumination, equipment occlusion and noise interference in complex construction environments through spatiotemporal calibration and anti-interference processing of multi-source sensor data, significantly improving the fusion quality and stability of multimodal data; the multi-level feature extraction mechanism based on the neural network model can perform high-granularity modeling of the construction personnel's motion trajectory, and at the same time, combined with the dynamic analysis of equipment operating parameters, it can accurately capture high-risk behaviors such as illegal climbing postures and non-compliant use of tools; further, through the collaborative analysis of temperature gradients and electrical characteristics, it can break through the limitations of traditional threshold detection. The system can effectively identify the early overheating hazards of equipment and realize the advance warning of abnormal events; relying on the assessment model of dynamic coupling of spatiotemporal correlation matrix and risk weight, it can transform isolated risk events into probabilistic risk levels in three-dimensional space, thereby constructing a multi-dimensional and continuous dynamic risk map; finally, through the threshold trigger mechanism of graded response instructions, it can adaptively match differentiated emergency strategies such as on-site personnel warnings, equipment emergency power off and evacuation route planning according to the risk evolution trend, forming a full-process closed-loop management and control system from risk perception, assessment to disposal, and significantly improving the real-time performance and decision-making accuracy of power construction safety supervision.

[0021] For better understanding and implementation, the present invention is described in detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 A flowchart of a method for monitoring power construction provided in an embodiment of the present application;

[0023] Figure 2 A schematic diagram of a process for generating a feature graph set including a behavior feature graph and a device state feature graph provided in an embodiment of the present application;

[0024] Figure 3A schematic structural diagram of a power construction monitoring system provided in another embodiment of the present application. DETAILED DESCRIPTION

[0025] To facilitate understanding of the present invention, the present invention will be described more fully below with reference to the accompanying drawings. The drawings illustrate preferred embodiments of the present invention. However, the present invention may be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a more thorough and comprehensive understanding of the disclosure.

[0026] Unless otherwise defined, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which this invention pertains. The terms used in this specification are for the purpose of describing specific embodiments only and are not intended to limit the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0027] In one embodiment, Figure 1 As shown, a method for monitoring power construction is provided. This embodiment uses the method applied to a terminal as an example for illustration. It is understandable that the method can also be applied to a server, or to a system including a terminal and a server, and implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0028] S1: Perform time-space position calibration and anti-interference processing on the construction image data, temperature data, and dust concentration data collected by multi-source sensors at the construction site to eliminate sensor acquisition delays, compensate for spatial perspective differences, and suppress noise, thereby generating a preprocessed multi-source fusion dataset.

[0029] Specifically, the construction site of this embodiment integrates various types of sensors, including high-definition cameras, infrared thermal imagers, dust concentration detectors, temperature sensors, and vibration sensors. High-definition cameras are used to capture the movements of construction workers and the overall situation of the construction site in real time. Their high resolution and wide dynamic range enable clear recording of construction details. Infrared thermal imagers focus on monitoring the temperature distribution of electrical equipment, promptly identifying abnormally hot areas and preventing potential thermal failures. Dust concentration detectors monitor dust levels at the construction site in real time, ensuring the health and safety of construction workers while also preventing damage to electrical equipment caused by excessive dust concentration. Temperature sensors are widely distributed around key equipment, providing accurate environmental and equipment temperature data. Vibration sensors monitor the mechanical stability of electrical equipment and capture abnormal vibrations to prevent equipment failure. These sensors are evenly distributed and rationally configured based on the construction site layout and monitoring requirements to ensure comprehensive and accurate data collection. Through a high-speed communication network, each sensor transmits the collected data to the monitoring system center in real time, enabling rapid data aggregation and preliminary processing, providing a rich data source for subsequent comprehensive analysis and risk assessment.

[0030] Specifically, due to differences in the operating principles, sampling frequencies, and installation locations of each sensor, the collected data is not completely aligned in time and space. To address this, the system can use a timestamp-based synchronization algorithm to time-calibrate the data from different sensors and estimate the time difference between different sensors through interpolation, thereby eliminating acquisition delays. To address differences in spatial perspective, the system can utilize the three-dimensional modeling data of the construction site, apply perspective transformation and interpolation algorithms to geometrically correct the image data, and map images from different perspectives into a unified spatial coordinate system. To address noise interference in the data, the system can use adaptive filtering technology to dynamically adjust the filtering parameters based on the statistical characteristics of the data, thereby achieving effective noise reduction of the sensor signals. After the above steps, the system generates a preprocessed multi-source fusion dataset that integrates information such as construction images, temperature, and dust concentration. This dataset provides accurate and reliable basic data support for subsequent feature extraction and risk identification, ensuring that the system can comprehensively and accurately perceive the dynamic changes of the construction site.

[0031] S2: Based on the neural network model, multi-level feature extraction is performed on the multi-source fusion data set to identify the movement trajectory of construction workers and the operating parameters of equipment, and generate a feature map set including behavioral feature maps and equipment status feature maps.

[0032] Specifically, the neural network model is pre-trained on a large amount of annotated data, resulting in powerful feature extraction capabilities. For construction image data, the system processes the images layer by layer using a convolutional neural network (CNN). It first extracts low-level features, such as edges, textures, and other basic visual elements. As the network depth increases, it gradually abstracts higher-level semantic features, such as the construction worker's posture and the shape and position of tools. During this processing, the system uses sliding window technology to analyze consecutive frames of imagery, capturing the changing trends of the construction worker's movements and thereby constructing a characteristic trajectory of their motion.

[0033] For sensor data such as temperature and dust concentration, the system can use long short-term memory (LSTM) networks to model time series data and extract temporal features from the data, including key information such as rate of change and amplitude of fluctuation, to reflect the dynamic changes in equipment operating status and environmental conditions. These features are ultimately integrated into a collection of feature graphs. The behavioral feature graph details the construction workers' movement patterns, tool usage, and location information, while the equipment status feature graph covers key indicators such as the equipment's temperature distribution, vibration frequency, and electrical operating parameters. This provides a comprehensive and accurate feature description for subsequent risk identification, enabling the system to keenly capture potential risk factors during the construction process.

[0034] S3: Identify high-risk actions and equipment anomalies in the feature graph collection, analyze the movement patterns of human joints and the compliance of tool use, and detect abnormal changes in equipment temperature to generate a construction risk collection containing dangerous behavior events and equipment failure events.

[0035] Specifically, for the construction workers' behavioral characteristic maps, the system can use human posture estimation algorithms to accurately extract human joint coordinate information, and combine dynamic models to analyze the joint's motion trajectory, angle changes, and speed characteristics. By performing pattern matching with preset safety action templates, the system can identify high-risk actions such as climbing high places without wearing a safety belt and illegally operating large machinery. At the same time, the system analyzes the use of tools and uses shape matching and semantic understanding technology to determine whether construction workers use tools in accordance with standard procedures. For example, there are violations such as touching live equipment without wearing insulating gloves or using damaged tools. For equipment status characteristic maps, the system can use a multi-threshold judgment strategy combined with time-frequency analysis methods to detect abnormal temperature changes. By monitoring the thermal imaging data of the surface temperature distribution of the equipment, it can identify local overheating areas and combine historical data to determine whether the temperature change trend exceeds the normal range.

[0036] The system integrates the identified high-risk actions and equipment anomalies into a construction risk set, which records in detail key information such as the type, location, time and severity of each risk event, providing a comprehensive and accurate basis for subsequent risk assessment and early warning processing, ensuring that various risks in the construction process can be identified and responded to in a timely and effective manner.

[0037] S4: Perform spatiotemporal correlation analysis on the construction risk set. By calculating the event distance in three-dimensional space and combining it with continuous monitoring in the time dimension, a risk level assessment result is generated. The risk level assessment result is used to quantify the comprehensive accident probability level within the construction area.

[0038] Specifically, in the three-dimensional spatial dimension, the system can calculate the Euclidean distance and spatial topological relationship between different risk events based on the geographic information system (GIS) data of the construction site, and analyze the distribution pattern and mutual influence of risk events in the construction site. For example, if there is illegal climbing behavior on a high-altitude work platform, and there is an abnormal temperature in the electrical equipment nearby, the system will comprehensively consider the spatial distance and relative position of the two to assess the risk of chain safety accidents that may be caused. In the time dimension, the system continuously monitors risk events, records their occurrence time, duration and change trend. By establishing a time series model, the system analyzes the evolution of risk events over time and determines whether they show a trend of intensification or diffusion. For example, for abnormal equipment temperature events, the system monitors its temperature change curve. If the temperature continues to rise and exceeds the preset danger threshold, the system will promptly increase its risk level.

[0039] Combining spatial distance and temporal trend analysis, the system uses a fuzzy comprehensive evaluation method to generate risk level assessment results. These results present the comprehensive accident probability level within the construction area in the form of quantitative indicators, comprehensively considering the spatiotemporal correlations and coupling effects of multiple risk factors. The system categorizes risk levels into yellow and red warnings, and sets corresponding quantitative standards and assessment rules for each level. This provides a scientific and accurate basis for subsequent graded warning processing, ensuring that construction units can take targeted emergency measures based on risk levels, effectively reducing the probability of construction accidents and the extent of losses.

[0040] S5: Perform graded warning processing based on the risk level assessment results, trigger different levels of emergency response instructions through preset risk thresholds, generate and transmit comprehensive warning notifications to the construction unit's management terminal, and the emergency response instructions include one of the following: on-site personnel warning information push signal, equipment power cut-off instruction and evacuation route guidance signal.

[0041] Specifically, based on the generated risk level assessment results, the system activates a graded early warning mechanism to ensure construction safety. The system pre-sets yellow and red alerts. When the risk level assessment results reach the corresponding threshold, the corresponding emergency response instructions are automatically triggered. In the case of a yellow alert, the system sends a prompt message to the construction unit's management terminal, alerting on-site personnel to the potential risk and recommending preventive measures.

[0042] When a red alert occurs, in addition to sending a warning message, the system also activates the on-site audio and visual alarm devices, alerting construction workers to the danger through flashing lights and sirens. Based on pre-set safety policies, it generates equipment power-off instructions, shutting off power to equipment that could cause an accident to prevent the situation from deteriorating further. A detailed event report is also sent to the management terminal, including the location and type of risk and recommended treatment measures. Signals are sent to relevant equipment via an integrated control interface. Furthermore, the system integrates the construction site's geographic information and the location of safe exits to generate evacuation route guidance signals. These signals are then sent to construction workers via on-site electronic display screens or mobile terminal applications, guiding them to evacuate the danger zone quickly and orderly. These emergency response instructions are transmitted to the construction unit's management terminal via secure and reliable communication protocols, ensuring the timeliness and accuracy of information transmission.

[0043] After the management terminal receives the early warning notification, the construction unit can quickly organize personnel to carry out emergency response based on the detailed information provided by the system, and take measures including evacuation of on-site personnel, emergency repair of equipment, and safe isolation to minimize casualties and property losses caused by construction accidents and ensure the safety and stability of the power construction process.

[0044] In summary, the power construction monitoring method provided by the present invention can effectively suppress sudden changes in illumination, equipment occlusion and noise interference in complex construction environments through spatiotemporal calibration and anti-interference processing of multi-source sensor data, significantly improving the fusion quality and stability of multimodal data; the multi-level feature extraction mechanism based on the neural network model can perform high-granularity modeling of the construction personnel's motion trajectory, and at the same time, combined with the dynamic analysis of equipment operating parameters, it can accurately capture high-risk behaviors such as illegal climbing postures and non-compliant use of tools; further, through the collaborative analysis of temperature gradients and electrical characteristics, it can break through the limitations of traditional threshold detection. The system can effectively identify the early overheating hazards of equipment and realize the advance warning of abnormal events; relying on the assessment model of dynamic coupling of spatiotemporal correlation matrix and risk weight, it can transform isolated risk events into probabilistic risk levels in three-dimensional space, thereby constructing a multi-dimensional and continuous dynamic risk map; finally, through the threshold trigger mechanism of graded response instructions, it can adaptively match differentiated emergency strategies such as on-site personnel warnings, equipment emergency power off and evacuation route planning according to the risk evolution trend, forming a full-process closed-loop management and control system from risk perception, assessment to disposal, and significantly improving the real-time performance and decision-making accuracy of power construction safety supervision.

[0045] In one embodiment, S1 of a power construction monitoring method provided by the present invention specifically includes the following steps:

[0046] S11: Based on clock synchronization technology, the construction image data collected by the multispectral camera is timestamp aligned to eliminate the acquisition delay between sensors and generate a time-synchronized video data stream. The construction image data includes visible light video data and infrared video data.

[0047] Specifically, in the power construction monitoring scenario, different cameras may have slight differences in acquisition delays, which will cause the video data to be out of sync in the time dimension, affecting the subsequent comprehensive analysis. To solve this problem, this embodiment introduces clock synchronization technology, which ensures that all cameras have a unified time base when collecting data by calibrating the built-in clock of each camera. The system first obtains the original construction image data collected by the visible light camera and the infrared thermal imager. These data carry their respective timestamp information. By analyzing the timestamps of each camera, the system calculates the delay differences between different cameras and adjusts the video data based on these delay differences.

[0048] Specifically, the system uses clock synchronization algorithms such as the Persistent Time Protocol (PTP) or Network Time Protocol (NTP), using a high-precision clock server as a reference source, to calibrate the camera's acquisition clock in real time. After timestamp alignment, the system eliminates acquisition delays between sensors and generates a time-synchronized video data stream containing both visible light and infrared video data. This data maintains high temporal consistency, providing a synchronization foundation for subsequent spatiotemporal fusion processing.

[0049] S12: Perform spatial registration processing on the video data stream based on the weighted Kalman filter algorithm to compensate for the camera perspective differences and generate spatiotemporally aligned multispectral video data.

[0050] Specifically, due to the complexity of construction sites and the diverse nature of monitoring requirements, multispectral cameras typically capture the construction area from varying angles and positions, resulting in differences in the spatial perspective of the captured video data. This perspective difference can lead to inconsistent positions and shapes of the same object across different video streams, complicating subsequent feature extraction and analysis. To address this issue, the system employs a weighted Kalman filter algorithm for spatial registration. This algorithm accurately estimates internal parameters such as the camera's focal length and principal point coordinates, as well as external parameters such as the rotation matrix and translation vector, to establish the geometric transformation relationship between the cameras.

[0051] During the processing process, the system first constructs an initial spatial transformation model based on the camera's calibration parameters. It then dynamically optimizes the model parameters using a Kalman filter algorithm. Through two steps, prediction and update, the Kalman filter continuously adjusts the model parameters, enabling the spatial transformation model to better adapt to actual perspective differences. Furthermore, the system incorporates a weighting mechanism that weights the contributions of different cameras based on their data quality and reliability, thereby improving the accuracy of spatial registration. After spatial registration, the system generates spatiotemporally aligned multispectral video data, in which the video data collected by different cameras is precisely aligned in space, providing a reliable spatial foundation for subsequent multi-source data fusion.

[0052] S13: The multispectral video data, the temperature data collected by the environmental sensor, and the dust concentration data of the construction site are fused and processed, and a construction site dataset with environmental labels is generated through three-dimensional coordinate system mapping.

[0053] Specifically, in power construction monitoring, a single type of data often fails to fully reflect the complex conditions at the construction site. Therefore, it is necessary to fuse data from different sensor types to achieve more comprehensive and accurate monitoring. The system first acquires spatiotemporally aligned multispectral video data, temperature data collected by environmental sensors, and dust concentration data. These data respectively carry visual, thermal, and environmental information about the construction site. To effectively fuse these heterogeneous data, the system establishes a unified three-dimensional coordinate system, centered at a fixed point on the construction site and covering the entire construction area. Using geometric modeling and spatial mapping algorithms, the system maps pixels in the multispectral video data to spatial points in the three-dimensional coordinate system and maps temperature and dust concentration data to corresponding spatial locations. During the fusion process, the system employs data interpolation and spatial interpolation techniques to uniformly process data of varying resolutions and sampling frequencies, ensuring that all data have the same sampling density and spatial resolution in the three-dimensional coordinate system. Furthermore, the system assigns environmental labels to each spatial point, including temperature, dust concentration, and other relevant environmental parameters, thereby generating a labeled construction site dataset. This dataset not only contains rich visual information, but also incorporates environmental perception information, providing multi-dimensional data support for subsequent risk identification and assessment.

[0054] S14: Perform illumination compensation processing on the visible light video frames of the construction site dataset based on the dynamic histogram equalization algorithm to balance the differences between bright and dark areas and generate video frames with uniform brightness.

[0055] Specifically, in power construction monitoring, lighting conditions at construction sites are often complex and variable, potentially resulting in areas that are too bright or too dark, affecting the quality of video data and subsequent feature extraction. To address this issue, the system uses a dynamic histogram equalization algorithm to perform illumination compensation on visible light video frames. This algorithm statistically analyzes the pixel brightness values ​​within a video frame and dynamically adjusts the histogram distribution to achieve a more uniform distribution of brightness values. During the processing, the system first calculates a histogram of the pixel brightness values ​​within the video frame and analyzes the distribution of brightness values. Then, based on the histogram distribution characteristics, the system determines the cumulative distribution function of the brightness values ​​and normalizes the cumulative distribution function to generate a new brightness value mapping. Unlike traditional histogram equalization, the dynamic histogram equalization algorithm adaptively adjusts the mapping based on the local characteristics of the video frame and illumination variations, avoiding over-enhancement or under-enhancement. The system also considers the temporal continuity of the video frames, introducing temporal smoothing to ensure smooth and natural brightness transitions between adjacent frames after illumination compensation. After illumination compensation processing, the system generates video frames with uniform brightness, in which the differences between light and dark areas are effectively balanced, and the contrast and clarity of the image are significantly improved, providing higher-quality visual data for subsequent feature extraction and analysis.

[0056] S15: Perform occlusion repair processing on the uniform brightness video frame based on the edge repair algorithm to fill the missing pixels in the occluded area and generate a complete scene video frame.

[0057] Specifically, in construction site monitoring, factors such as construction equipment, material stacking, and human activity may cause occlusion in the video frame, resulting in the loss of pixel information in some areas, affecting the complete understanding and analysis of the construction scene. To address this problem, the system uses an edge repair algorithm to repair occluded areas in the video frame. This algorithm analyzes the edge information in the video frame and fills the occluded area with pixel information from the surrounding unoccluded areas. During the processing, the system can use the Canny edge detection algorithm to identify edge pixels in the video frame and construct an edge map. The system then analyzes the edge structure and texture features around the occluded area in the edge map to determine the boundary and direction of the occluded area.

[0058] Based on this information, the system can employ methods based on curve fitting or texture synthesis to generate pixel information that matches the surrounding area and fill in the missing portions of the occluded area. The system also considers the temporal continuity of video frames, analyzing the changes in occluded areas across adjacent video frames to optimize occlusion repair in the temporal domain, ensuring that the repaired video frames maintain consistency over the time series. After occlusion repair processing, the system generates complete scene video frames, effectively filling in the missing pixels in the occluded area and restoring the integrity of the construction scene, providing more comprehensive visual information for subsequent construction behavior analysis and risk identification.

[0059] S16: Perform noise suppression processing on the complete scene video frames, remove high-frequency noise interference through non-local mean filtering, and generate a multi-source fusion dataset containing an optimized video frame sequence.

[0060] Specifically, in power construction monitoring, despite the previous processing steps, video data may still contain some noise. This noise may originate from electronic noise in the sensor itself, environmental electromagnetic interference, or errors during data transmission. This high-frequency noise can degrade the quality of the video data and affect subsequent feature extraction and analysis. To address this issue, the system employs a non-local mean filtering algorithm to suppress noise in complete scene video frames. This algorithm leverages the principle of non-local self-similarity in images and achieves noise suppression by weighted averaging similar pixel blocks within the video frame. During processing, the system first divides the video frame into multiple small blocks and then searches for blocks similar to the current block. Based on a similarity metric between the blocks, the system calculates a weight for each similar block and performs a weighted average of these blocks to generate the denoised pixel values. The non-local mean filtering algorithm effectively preserves image edge and texture details while removing noise, avoiding the blurring effect that can occur with traditional filtering algorithms. The system also considers the temporal continuity of video frames, further improving the denoising effect by jointly processing similar blocks in adjacent frames in the temporal domain. After noise suppression processing, the system generates a multi-source fusion dataset containing an optimized video frame sequence. This dataset integrates multispectral video data, temperature data, and dust concentration data. After multiple steps of processing such as time-space calibration, illumination compensation, occlusion repair, and noise suppression, it has high temporal synchronization, spatial consistency, illumination uniformity, and data integrity, providing a high-quality data foundation for subsequent construction risk identification, behavior analysis, and equipment status monitoring.

[0061] In one embodiment, Figure 2 As shown, S2 of a power construction monitoring method provided by the present invention specifically includes the following steps:

[0062] S21: Replace the 1x1 convolutional downsampling layer of the original ResNet-50 network with a 3x3 convolutional layer, embed a channel attention mechanism in the residual module, enhance the response strength of the safety equipment area by recalibrating the feature map channel weights, and construct an improved ResNet-50 network.

[0063] Specifically, the 1x1 convolution downsampling layer of the traditional ResNet-50 network may lose some spatial information during the dimensionality transformation of the feature map. This embodiment provides a method of replacing the 1x1 convolution downsampling layer in the ResNet-50 network with a 3x3 convolution layer. This can increase the receptive field while adjusting the number of channels, capture more spatial context information, enhance the ability to express local features in the feature map, and be more conducive to identifying complex and diverse object contours and details in construction scenes, such as the edges of safety equipment and the texture of equipment.

[0064] At the same time, based on the traditional ResNet-50 network, this embodiment embeds a channel attention mechanism in the residual module. By modeling the information in the channel dimension of the feature map, it learns the importance weights of different channel features and realizes the recalibration of the feature map channel weights. In the power construction monitoring scenario, safety equipment areas such as hard hats, insulating gloves, insulating shoes, and key component areas of electrical equipment are the focus of risk identification. The channel attention mechanism can enhance the response strength of these areas in the feature map, allowing the network to focus more on these key areas, thereby improving the ability to identify whether construction workers are wearing safety equipment correctly and whether the equipment is in normal condition.

[0065] S22: Based on the improved ResNet-50 network, spatial feature extraction is performed on the multi-source fusion data set to extract spatial features at different levels and generate multi-scale spatial feature maps.

[0066] Specifically, the improved ResNet-50 network, through its multi-layer convolutional operations and residual connection structure, is able to automatically learn spatial features at different levels in the data. These features range from shallow edge and texture features to deeper semantic features, such as the posture of construction workers and the appearance and shape of equipment. When processing a multi-source fusion dataset, the network comprehensively analyzes the rich visual details in visible light video frames and temperature distribution differences in infrared video frames. This multi-source fusion dataset contains high-quality, pre-processed video frame sequences that are temporally synchronized, spatially aligned, and exhibit uniform illumination and data integrity, providing a solid foundation for the network's feature extraction. After a series of convolution, pooling, and activation operations, the system ultimately generates multi-scale spatial feature maps. These feature maps depict visual information of the construction site at different scales, including large-scale scene layout features such as the overall structure of the substation and the orientation of transmission lines, as well as small-scale local features such as the shapes of construction tools and equipment markings. These provide a multi-dimensional spatial feature representation for more detailed feature analysis and risk identification.

[0067] S23: Perform attention weighting processing on the multi-scale spatial feature map, enhance the weight coefficients of the helmet and insulating tool areas through the channel attention mechanism, and generate a behavioral feature map.

[0068] Specifically, in power construction monitoring, the behavioral norms of construction workers are crucial for safety assurance. Correctly wearing safety helmets and using insulating tools in a standardized manner are important bases for risk assessment. The system uses a channel attention mechanism to weight channels in multi-scale spatial feature maps. The channel attention mechanism calculates the importance weight of each channel by capturing the global information of the feature map. Channels related to key areas such as safety helmets and insulating tools are given higher weight coefficients, thereby enhancing the significance of these areas in the feature map. The calculation of the weight coefficient satisfies the following formula:

[0069] w c =σ(W2*δ(W1*GAP(F c )))

[0070] Among them, w c is the weight coefficient of the cth channel, F c is the cth channel of the multi-scale spatial feature map, GAP represents the global average pooling operation, W1 and W2 are the fully connected layer parameters, δ is the ReLU activation function, and σ is the Sigmoid function. This weighted processing allows the behavioral feature map to focus more on key behavioral characteristics such as the construction workers' wearing of safety equipment and the use of insulating tools. This improves the accuracy of identifying the normative behavior of construction workers, helps to promptly detect violations, and reduces construction risks.

[0071] S24: Based on the three-dimensional wavelet transform algorithm, the infrared video frames in the multi-source fusion data set are subjected to temperature feature extraction processing, the temperature gradient distribution of the equipment area is analyzed, and the equipment status feature map is generated.

[0072] Specifically, infrared video frames can reflect the thermal state of electrical equipment and serve as an important data source for detecting abnormal heating. A three-dimensional wavelet transform algorithm decomposes infrared video frames in both time and space, obtaining temperature signature information at different scales. In the spatial dimension, the algorithm can capture detailed localized temperature variations on the equipment surface, such as sudden temperature changes in a component. In the temporal dimension, it can analyze temperature trends over time, such as gradual increases or sudden decreases in temperature during equipment operation. By extracting temperature signatures from infrared video frames, the system can analyze the temperature gradient distribution across the equipment. This temperature gradient reflects temperature differences between different parts of the equipment. Normal equipment operation typically maintains a relatively stable temperature gradient, while conditions such as equipment failure or overload can result in excessive or abnormally fluctuating local temperature gradients. Based on these temperature signatures, the system generates a device status feature map, which visually displays the thermal status of the equipment and provides a key basis for detecting equipment anomalies.

[0073] S25: Fusing the behavior feature graph and the device state feature graph to generate a feature graph set including the behavior feature graph and the device state feature graph.

[0074] Specifically, the behavioral feature map mainly describes the behavioral characteristics of construction workers, such as the wearing of safety equipment, movement posture, etc.; the equipment status feature map reflects the thermal state and operation status of electrical equipment. In order to achieve the effective fusion of these two feature maps, the system uses methods such as feature splicing or feature fusion network to combine the behavioral feature map and the equipment status feature map in the channel dimension or spatial dimension. In this way, the system constructs a comprehensive feature map set that includes both the behavioral characteristics of construction workers and the status characteristics of equipment, which can more comprehensively reflect the risk factors of the construction site. This fusion processing enables the subsequent risk assessment model to consider both personnel behavior and equipment status factors at the same time, improve the accuracy and comprehensiveness of risk assessment, and provide more reliable decision support for power construction monitoring.

[0075] The above-mentioned power construction monitoring method replaces the 1x1 downsampling layer of the original ResNet-50 network with a 3x3 convolution layer, which can slow down the attenuation rate of the spatial resolution of the feature map. Combined with the channel attention mechanism embedded in the residual module, the channel weights of key areas such as safety helmets and insulating tools can be dynamically calibrated to achieve robust extraction of construction personnel behavior characteristics under complex lighting and occlusion scenes; based on the multi-scale spatial feature extraction of multi-source data by the improved network, it can fuse local details and global semantic information under different receptive fields, and enhance the detection of climbing and illegal operations. The fine-grained characterization capability of high-risk actions such as maneuvering and performing other tasks; further, the time-frequency domain joint analysis of the temperature gradient distribution of infrared video frames through three-dimensional wavelet transform can break through the limitations of traditional temperature threshold detection, capture the transient temperature fluctuation characteristics of early overheating hazards of equipment, and improve the recognition sensitivity of abnormal heating events; finally, through the cross-modal fusion of behavioral feature graphs and equipment status feature graphs, it is possible to construct a feature expression system for the multi-dimensional correlation between man, machine and environment, providing high-confidence data support for the subsequent spatiotemporal correlation analysis of risk events, thereby solving the problem of misjudgment and missed judgment caused by feature isolation in existing methods.

[0076] In one embodiment, S3 of a power construction monitoring method provided by the present invention specifically includes the following steps:

[0077] S31: Perform motion trajectory analysis on the human skeleton key point sequence in the feature map set, calculate the abnormal index of the climbing action through the joint angle change rate model, and generate a climbing risk event.

[0078] Specifically, the system analyzes and processes the motion trajectory of the human skeleton keypoint sequence in the feature map collection. Using a deep learning algorithm, the system pre-extracts the construction worker's skeletal keypoints to form a keypoint sequence. This sequence includes key joints such as the wrist, elbow, shoulder, and hip, accurately reflecting changes in posture and movement. The system uses a kinematic model to analyze the skeletal keypoint sequence, calculating the positional change of each joint across consecutive frames to derive the motion trajectory. Based on this, the system introduces a joint angle change rate model, which calculates the angle change rate between adjacent joints based on the kinematic principles of human joints. For climbing movements, the system's model focuses on analyzing the angle change rate of the worker's limb joints, such as the periodic changes in the shoulder and hip joints during climbing. If the joint angle change rate exceeds a preset threshold for normal climbing movement, the system determines that the climbing movement is abnormal. Based on the degree of abnormality, it calculates an anomaly index, which then generates a climbing risk event.

[0079] S32: Perform tool contact detection on the feature map of the construction worker's hand area in the feature map set, identify the behavior of touching a live object without using an insulating tool through a contact relationship classification model, and generate an illegal operation event.

[0080] Specifically, the system utilizes target detection algorithms from deep learning to accurately locate the construction worker's hand area and generate corresponding feature maps. Within these feature maps, the contact characteristics of the hand with tools and live objects are the focus of analysis. The system constructs a contact relationship classification model, trained with a large amount of sample data, capable of identifying the characteristic patterns of hand contact with different objects. During the detection process, the system extracts multi-dimensional features such as texture, shape, and color from the hand area feature map and inputs them into the classification model. Based on these features, the model determines whether the construction worker used insulated tools to touch live objects. The system can quickly identify and record violations where insulated tools were not used. Once such a violation is detected, the system generates a violation event, detailing the specific circumstances of the incident, such as the time, the identity of the construction worker, and the location of the live object involved, thereby effectively monitoring the construction worker's operational compliance.

[0081] S33: Perform high-temperature region segmentation processing on the device status feature graph of the feature graph set, locate abnormal heating areas of the electrical equipment through the U-Net++ network, and generate a high-temperature region marking map.

[0082] Specifically, the device status feature map contains rich temperature information, acquired through infrared imaging technology, and can reflect the temperature distribution on the surface of electrical equipment. The system uses the advanced U-Net++ network architecture to segment high-temperature areas. The U-Net++ network, with its powerful feature extraction and segmentation capabilities, excels in processing this type of image data. Through multi-path convolution operations and feature fusion mechanisms, the network accurately identifies and segments high-temperature areas on the device surface. During processing, the system inputs the device status feature map into the U-Net++ network, which, through a series of encoding and decoding operations, ultimately outputs a labeled map of the high-temperature areas.

[0083] S34: Perform temperature change rate analysis on the high temperature area marker image, determine transient temperature rise abnormal events through wavelet coefficient thresholds, and generate overheating fault events.

[0084] Specifically, based on the high-temperature area marker map, the system collects temperature change data for the area over a continuous period of time and constructs a temperature time series. In order to accurately determine transient temperature rise anomalies, the system uses a wavelet transform algorithm to analyze the temperature time series. Wavelet transform can decompose signals into different scales, which helps to capture the sudden change characteristics in temperature changes. The system determines whether there is a transient temperature rise anomaly by calculating the wavelet coefficient and comparing it with the preset threshold. When the wavelet coefficient exceeds the set threshold, it indicates that the temperature has risen abnormally rapidly on the corresponding time scale, and the system generates an overheating fault event accordingly. This event records key parameters such as the specific time when the equipment overheated, the location of the high-temperature area, and the temperature change rate, providing timely and accurate information for equipment maintenance and fault handling.

[0085] S35: Aggregate climbing risk events, illegal operation events, and overheating failure events to generate a construction risk set containing dangerous behavior events and equipment failure events. Dangerous behavior events include climbing risk events and illegal operation events, and equipment failure events include overheating failure events.

[0086] Specifically, the system builds a comprehensive risk event database, storing various risk events in a unified data structure and format. For each risk event, the system records its type, time of occurrence, location information, construction personnel or equipment involved, severity of the event and other details. During the aggregation process, the system uses event correlation analysis to determine whether there is a correlation between different events, such as whether a climbing risk event may cause equipment overheating failure. In this way, the system generates a comprehensive construction risk set that integrates dangerous behavior events and equipment failure events. This comprehensive construction risk set provides comprehensive and accurate data support for subsequent risk assessments and the formulation of emergency response strategies, helping construction units to take effective measures in a timely manner to reduce construction risks and ensure construction safety.

[0087] In summary, the present invention uses a joint angle change rate model to perform kinematic modeling on the sequence of key points of the human skeleton, which can realize dynamic monitoring of abnormal angles of the knee and elbow joints during climbing movements, breaking through the limitation of insufficient analysis of posture features by traditional trajectory tracking; based on the contact relationship classification model, tool compliance detection is performed on the hand area feature map, which can accurately identify high-risk behaviors of touching live objects without using insulating tools, so as to solve the problem of rough judgment of operation compliance in existing technologies; high-temperature area segmentation is performed on the equipment status feature map through the U-Net++ network, and the transient characteristic analysis of the temperature change rate by the wavelet coefficient threshold is combined to realize high-sensitivity detection of early overheating hazards of electrical equipment, overcoming the hysteresis defect of conventional temperature threshold detection; finally, through the spatiotemporal aggregation processing of multiple types of risk events, a multi-dimensional risk set covering personnel behavior and equipment status can be constructed, providing a fine-grained data basis for dynamic risk assessment, thereby forming a closed-loop control chain from risk identification, quantification to response, and significantly improving the timeliness and systematicness of safety warnings in complex construction scenarios.

[0088] In one embodiment, S4 of a power construction monitoring method provided by the present invention specifically includes the following steps:

[0089] S41: Perform spatial correlation processing on the construction risk set, determine the proximity relationship between events through three-dimensional Euclidean distance calculation, and generate a spatial correlation matrix.

[0090] Specifically, the system divides the construction site into multiple three-dimensional spatial units, each unit corresponding to a specific construction area. For each risk event in the construction risk set, the system obtains its three-dimensional coordinate information, including the x, y, and z coordinates of the event in the construction site. The system uses the three-dimensional Euclidean distance formula to calculate the spatial distance between any two risk events and records it in the spatial association matrix. The rows and columns of the matrix represent different risk events, and the values ​​of the matrix elements represent the distance between the corresponding events. By setting a distance threshold, the system can determine which events have a spatial proximity relationship, thereby identifying groups of risk events that may affect each other. This spatial association analysis helps to reveal the distribution pattern of risk events in the construction site and the potential chain reaction risks, providing a spatial dimension basis for subsequent comprehensive risk assessment.

[0091] S42: Perform time window analysis on the spatial correlation matrix, count the cumulative frequency of events in the same area through a sliding time window, and generate spatiotemporal coupling risk events.

[0092] Specifically, the system defines a time window, such as 10 minutes, as the time unit for analysis, and slides the window on the time axis of construction monitoring. Within each time window, the system counts the number of risk events in each construction area for the adjacent event groups identified in the spatial association matrix. Through cumulative frequency statistics, the system can identify the concentrated outbreak of risk events in the same area within a specific time period, thereby generating spatiotemporal coupled risk events. These spatiotemporal coupled risk events not only include the spatial location and type of the event, but also record the frequency and duration of the event within the time window. This time window analysis method can effectively capture the dynamic changes of risk events in the spatiotemporal dimensions, help the system to promptly discover concentrated risks that may occur during the construction process, provide dynamic data support for the subsequent risk level quantification, and ensure that the system can timely adjust the warning level and emergency response measures according to the spatiotemporal changes of the risk.

[0093] S43: Quantify the levels of spatiotemporal coupling risk events, calculate the comprehensive risk score through the risk weight accumulation model, and generate the risk level assessment results.

[0094] Specifically, the system assigns a preset weight to each type of risk event based on the event's potential harm and probability of occurrence. For example, different weights are assigned to climbing risk events, illegal operation events, and overheating failure events. The system multiplies each event type in the spatiotemporal coupled risk events by its corresponding weight, then sums the weighted scores for all events within the same area to generate a comprehensive risk score for that area. This comprehensive risk score reflects the overall risk level within the construction area, taking into account the combined effects of multiple risk factors. Based on pre-defined risk classification criteria, the system maps the comprehensive risk score to different risk levels, such as low, medium, high, and extremely high. The risk level assessment results are presented to monitoring personnel in an intuitive manner, helping them quickly understand the overall risk status of the construction site and providing a quantitative basis for subsequent graded warning processing. This ensures that construction units can take appropriate emergency measures based on the risk level, effectively improving the safety and controllability of power construction.

[0095] In one embodiment, S5 of a power construction monitoring method provided by the present invention specifically includes the following steps:

[0096] S51: Perform yellow warning judgment processing on the comprehensive risk score of the risk level assessment result. If the comprehensive risk score is greater than a preset first risk threshold, generate and send an on-site personnel warning information push signal to the mobile terminal of the on-site personnel.

[0097] Specifically, the system pre-sets a first risk threshold, which is determined based on historical construction data, safety regulations, and expert experience. It aims to identify risk levels that require the attention of on-site personnel but have not yet reached an extremely dangerous level. During real-time monitoring, the system continuously obtains the comprehensive risk score output by the risk level assessment module and compares it with the first risk threshold. When the comprehensive risk score exceeds the first risk threshold, the system triggers a yellow warning mechanism. The system generates a warning information push signal for on-site personnel, which contains key information such as the type of risk event, location, severity, and corresponding precautions. The system establishes a communication link with the mobile terminal of the on-site personnel and uses wireless communication technologies such as 4G / 5G networks to send warning information push signals to the smartphones, smart helmets, or other portable terminal devices of the on-site construction personnel, ensuring that the construction personnel can receive warning information in a timely manner, be vigilant, and take preventive measures.

[0098] S52: Perform red warning judgment processing on the comprehensive risk score of the risk level assessment result. If the comprehensive risk score is greater than the preset second risk threshold, generate and send a device power cut-off instruction to the device end, and at the same time generate and send an evacuation route guidance signal to the mobile terminal of the on-site personnel.

[0099] Specifically, the second risk threshold is higher than the first risk threshold, indicating that an extremely serious risk situation has occurred at the construction site, which may cause serious damage to the safety of personnel and equipment. When the comprehensive risk score exceeds the second risk threshold, the system immediately initiates the red warning process. On the one hand, the system generates an equipment power cut-off instruction, which is quickly sent to the control end of the relevant equipment through wired or wireless communication to achieve emergency power cut-off of the equipment in the dangerous area to prevent the expansion of the fault or the occurrence of an accident. On the other hand, the system generates an evacuation route guidance signal, which comprehensively considers factors such as the location of the risk event, the site layout, and the location of the safe exit, and plans the optimal evacuation route for on-site personnel. The system also sends the evacuation route guidance signal to the construction personnel through the mobile terminal. The guidance signal is presented in the form of a map, arrow instructions or voice navigation to guide personnel to evacuate the dangerous area quickly and orderly.

[0100] S53: Log the push signals for warning information to on-site personnel, the power cut-off instructions for equipment, and the evacuation route guidance signals, generate an unalterable comprehensive early warning notice through the blockchain hash chain storage, and send it to the management terminal of the construction unit.

[0101] Specifically, the system uses blockchain hash chain technology to encrypt these signals, along with key information such as transmission time and recipients, to generate an immutable, comprehensive record of early warning notifications. The characteristics of blockchain hash chains ensure the integrity and security of these records. Any modifications to the records are immediately detected, ensuring transparency and traceability of the early warning process. These records are stored in the system's database and transmitted to the construction unit's management terminal via a secure communication protocol. This terminal can be the construction unit's monitoring center server or a manager's mobile device, allowing managers to view detailed logs of early warning processing in real time and conduct post-analysis and review of risk events during the construction process, providing data support for optimizing safety strategies and improving construction processes. Furthermore, these immutable records facilitate accountability and accident investigation in the event of an incident, ensuring the seriousness and effectiveness of construction safety management.

[0102] Preferably, if Figure 3 As shown, the present invention provides a power construction monitoring system 600, which is configured with the following modules:

[0103] The data acquisition preprocessing module 610 is used to perform time-space position calibration and anti-interference processing on the construction image data, temperature data, and dust concentration data collected by multi-source sensors at the construction site, eliminate sensor acquisition delays, compensate for spatial perspective differences, and suppress noise, thereby generating a preprocessed multi-source fusion data set;

[0104] Multi-source feature extraction module 620, used to perform multi-level feature extraction processing on the multi-source fusion data set based on the neural network model, identify the movement trajectory of the construction workers and the operating parameters of the equipment, and generate a feature map set including behavior feature maps and equipment status feature maps;

[0105] Abnormal behavior identification module 630 is used to identify high-risk actions and equipment anomalies in the feature graph set, analyze human joint movement patterns and tool usage compliance, and generate a construction risk set containing dangerous behavior events and equipment failure events by detecting abnormal equipment temperature changes and electrical spark characteristics;

[0106] Comprehensive risk assessment module 640 is used to perform spatiotemporal correlation analysis on the construction risk set. By calculating the event distance in three-dimensional space and combining it with continuous monitoring in the time dimension, it generates a risk level assessment result. The risk level assessment result is used to quantify the comprehensive accident probability level within the construction area.

[0107] The emergency response and warning module 650 is used to perform graded warning processing based on the risk level assessment results, trigger different levels of emergency response instructions through preset risk thresholds, generate and transmit comprehensive warning notifications to the construction unit's management terminal, and the emergency response instructions include one of the following: on-site personnel warning information push signals, equipment power cut-off instructions, and evacuation route guidance signals.

[0108] In summary, the power construction monitoring system provided by the present invention can effectively suppress sudden changes in illumination, equipment occlusion and noise interference in complex construction environments through spatiotemporal calibration and anti-interference processing of multi-source sensor data, significantly improving the fusion quality and stability of multimodal data; the multi-level feature extraction mechanism based on the neural network model can perform high-granularity modeling of the construction personnel's motion trajectory, and at the same time, combined with the dynamic analysis of equipment operating parameters, it can accurately capture high-risk behaviors such as illegal climbing postures and non-compliant use of tools; further, through the collaborative analysis of temperature gradients and electrical characteristics, it can break through the limitations of traditional threshold detection. The system can effectively identify the early overheating hazards of equipment and realize the advance warning of abnormal events; relying on the assessment model of dynamic coupling of spatiotemporal correlation matrix and risk weight, it can transform isolated risk events into probabilistic risk levels in three-dimensional space, thereby constructing a multi-dimensional and continuous dynamic risk map; finally, through the threshold trigger mechanism of graded response instructions, it can adaptively match differentiated emergency strategies such as on-site personnel warnings, equipment emergency power off and evacuation route planning according to the risk evolution trend, forming a full-process closed-loop management and control system from risk perception, assessment to disposal, and significantly improving the real-time performance and decision-making accuracy of power construction safety supervision.

[0109] Preferably, the data acquisition and preprocessing module 610 provided in this embodiment is configured with the following units:

[0110] A time synchronization generation unit is used to perform time stamp alignment processing on the construction image data collected by the multispectral camera based on clock synchronization technology, eliminate the acquisition delay between sensors, and generate a time-synchronized video data stream. The construction image data includes visible light video data and infrared video data;

[0111] The video spatial registration unit is used to perform spatial registration processing on the video data stream based on the weighted Kalman filter algorithm, compensate for the difference in camera viewing angles, and generate multispectral video data that is aligned in time and space;

[0112] The video-environment fusion unit is used to fuse multispectral video data with temperature data collected by environmental sensors and dust concentration data at the construction site, and generate a construction site dataset with environmental labels through three-dimensional coordinate system mapping;

[0113] The video frame illumination compensation unit is used to perform illumination compensation processing on the visible light video frames of the construction site dataset based on the dynamic histogram equalization algorithm, balance the differences between bright and dark areas, and generate video frames with uniform brightness;

[0114] The video frame edge repair unit is used to perform occlusion repair processing on the uniform brightness video frame based on the edge repair algorithm, fill the missing pixels in the occluded area, and generate a complete scene video frame;

[0115] The multi-source fusion denoising unit is used to perform noise suppression on the complete scene video frames, remove high-frequency noise interference through non-local mean filtering, and generate a multi-source fusion dataset containing an optimized video frame sequence.

[0116] Preferably, the multi-source feature extraction module 620 provided in this embodiment is configured with the following units:

[0117] Improved ResNet-50 network building unit: This replaces the 1x1 convolutional downsampling layer of the original ResNet-50 network with a 3x3 convolutional layer. A channel attention mechanism is embedded in the residual module. This improves the response strength of the safety equipment area by recalibrating the feature map channel weights, thus building an improved ResNet-50 network.

[0118] The spatial feature extraction unit is used to perform spatial feature extraction processing on the multi-source fusion data set based on the improved ResNet-50 network, extract spatial features at different levels, and generate multi-scale spatial feature maps;

[0119] The behavioral feature map generation unit is used to perform attention-weighted processing on the multi-scale spatial feature map, enhance the weight coefficients of the helmet and insulating tool areas through the channel attention mechanism, and generate a behavioral feature map;

[0120] The device status feature map generation unit is used to extract temperature features from infrared video frames in the multi-source fusion data set based on a three-dimensional wavelet transform algorithm, analyze the temperature gradient distribution in the device area, and generate a device status feature map;

[0121] The feature graph set generating unit is used to fuse the behavior feature graph and the device state feature graph to generate a feature graph set including the behavior feature graph and the device state feature graph.

[0122] Preferably, the abnormal behavior identification module 630 provided in this embodiment is configured with the following units:

[0123] A climbing risk generation unit is used to analyze and process the motion trajectory of the human skeleton key point sequence in the feature map set, calculate the abnormal index of the climbing action through the joint angle change rate model, and generate a climbing risk event;

[0124] The illegal operation generation unit is used to perform tool contact detection on the feature map of the construction worker's hand area in the feature map set, identify the behavior of using an uninsulated tool to touch a live object through a contact relationship classification model, and generate an illegal operation event;

[0125] The high-temperature region marking unit is used to segment the device status feature map in the feature map set into high-temperature regions, locate abnormal heating areas of electrical equipment through the U-Net++ network, and generate a high-temperature region marking map;

[0126] Overheat fault generation unit, used to analyze and process the temperature change rate of the high-temperature area marker map, determine the transient temperature rise abnormal event through the wavelet coefficient threshold, and generate an overheat fault event;

[0127] The construction risk set generation unit is used to aggregate climbing risk events, illegal operation events and overheating failure events to generate a construction risk set containing dangerous behavior events and equipment failure events. Dangerous behavior events include climbing risk events and illegal operation events, and equipment failure events include overheating failure events.

[0128] Preferably, the comprehensive risk assessment module 640 provided in this embodiment is configured with the following units:

[0129] The risk spatial association unit is used to perform spatial association processing on the construction risk set, determine the proximity relationship between events through three-dimensional Euclidean distance calculation, and generate a spatial association matrix;

[0130] The risk event spatiotemporal coupling unit is used to perform time window analysis on the spatial correlation matrix, calculate the cumulative frequency of events in the same area through a sliding time window, and generate spatiotemporal coupling risk events;

[0131] The risk level assessment unit is used to quantify the level of spatiotemporal coupling risk events, calculate the comprehensive risk score through the risk weight accumulation model, and generate the risk level assessment results.

[0132] Preferably, the emergency response and warning module 650 provided in this embodiment is configured with the following units:

[0133] A yellow warning processing unit is used to perform yellow warning determination processing on the comprehensive risk score of the risk level assessment result. If the comprehensive risk score is greater than a preset first risk threshold, a field personnel warning information push signal is generated and sent to the mobile terminal of the field personnel;

[0134] A red warning processing unit is used to perform a red warning determination process on the comprehensive risk score of the risk level assessment result. If the comprehensive risk score is greater than a preset second risk threshold, a device power cut-off instruction is generated and sent to the device end, and an evacuation route guidance signal is generated and sent to the mobile terminal of the on-site personnel;

[0135] The comprehensive early warning notification unit is used to log and process warning information push signals for on-site personnel, equipment power cut-off instructions, and evacuation route guidance signals. It generates tamper-proof comprehensive early warning notifications through blockchain hash chain storage and sends them to the construction unit's management terminal.

[0136] In one embodiment, the present application further provides a computer device including a memory and a processor, wherein the memory stores a computer program, and the processor implements the above-mentioned power construction monitoring method when executing the computer program.

[0137] In one embodiment, the present application further provides a computer-readable storage medium having a computer program stored thereon, which implements the above-mentioned power construction monitoring method when executed by a processor.

[0138] In the description of this specification, the reference terms "one embodiment," "some embodiments," "example," "specific example," or "some examples" mean that the specific features, structures, materials, or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. Moreover, the specific features, structures, materials, or characteristics described may be combined in any appropriate manner in any one or more embodiments or examples. In addition, those skilled in the art may combine and integrate different embodiments or examples described in this specification, as well as features of different embodiments or examples, unless they are mutually inconsistent.

[0139] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to the partial description of the method embodiments. The device embodiments described above are merely illustrative, wherein the components described as separate parts may or may not be physically separated, and the parts displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the disclosed solution. A person of ordinary skill in the art can understand and implement it without expending creative work.

[0140] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any person skilled in the art can easily conceive of various modifications or substitutions within the technical scope disclosed in this application, and such modifications or substitutions should be included within the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A method for monitoring electric power construction, characterized in that: The following steps are involved: S1: Perform time-space position calibration and anti-interference processing on the construction image data, temperature data, and dust concentration data collected by multi-source sensors at the construction site to eliminate sensor acquisition delays, compensate for spatial perspective differences, and suppress noise, generating a pre-processed multi-source fusion dataset. S2: Perform multi-level feature extraction processing on the multi-source fusion data set based on a neural network model to identify the movement trajectory of construction workers and the operating parameters of equipment, and generate a feature graph set including a behavior feature graph and an equipment status feature graph; S3: Identify high-risk actions and equipment anomalies based on the feature graph set, analyze human joint movement patterns and tool usage compliance, and detect abnormal equipment temperature changes to generate a construction risk set containing dangerous behavior events and equipment failure events; S4: performing spatiotemporal correlation analysis on the construction risk set, generating a risk level assessment result by calculating event distances in three-dimensional space and combining continuous monitoring in the time dimension. The risk level assessment result is used to quantify the comprehensive accident probability level within the construction area; S5: Perform graded warning processing based on the risk level assessment results, trigger different levels of emergency response instructions through preset risk thresholds, generate and transmit comprehensive warning notifications to the construction unit's management terminal, and the emergency response instructions include one of the following: on-site personnel warning information push signal, equipment power cut-off instruction and evacuation route guidance signal.

2. The method according to claim 1, characterized in that Said S1 comprises: S11: performing time stamp alignment processing on the construction image data collected by the multispectral camera based on clock synchronization technology to eliminate acquisition delays between sensors and generate a time-synchronized video data stream, wherein the construction image data includes visible light video data and infrared video data; S12: performing spatial registration processing on the video data stream based on a weighted Kalman filter algorithm to compensate for differences in camera viewing angles and generate spatiotemporally aligned multispectral video data; S13: fusing the multispectral video data, the temperature data collected by the environmental sensor, and the dust concentration data of the construction site, and generating a construction site dataset with environmental labels through three-dimensional coordinate system mapping; S14: performing illumination compensation processing on the visible light video frames of the construction site dataset based on a dynamic histogram equalization algorithm to balance the differences between bright and dark areas and generate video frames with uniform brightness; S15: performing occlusion repair processing on the uniform brightness video frame based on an edge repair algorithm to fill in missing pixels in the occluded area and generate a complete scene video frame; S16: Performing noise suppression processing on the complete scene video frame, removing high-frequency noise interference through non-local mean filtering, and generating a multi-source fusion data set containing an optimized video frame sequence.

3. The method according to claim 1, characterized in that The S2 includes: S21: Replace the 1x1 convolutional downsampling layer of the original ResNet-50 network with a 3x3 convolutional layer, embed a channel attention mechanism in the residual module, and enhance the response strength of the safety equipment area by recalibrating the feature map channel weights to build an improved ResNet-50 network. S22: performing spatial feature extraction processing on the multi-source fusion dataset based on an improved ResNet-50 network, extracting spatial features at different levels, and generating a multi-scale spatial feature map; S23: Performing attention weighted processing on the multi-scale spatial feature map, enhancing the weight coefficients of the helmet and insulating tool areas through a channel attention mechanism, and generating a behavior feature map; S24: performing temperature feature extraction processing on the infrared video frames in the multi-source fusion data set based on a three-dimensional wavelet transform algorithm, analyzing the temperature gradient distribution of the device area, and generating a device status feature map; S25: Fusing the behavior feature graph and the device state feature graph to generate a feature graph set including the behavior feature graph and the device state feature graph.

4. The method according to claim 3, characterized in that In the aforementioned method of enhancing the weight coefficients of the helmet and insulating tool areas through the channel attention mechanism, the calculation of the weight coefficients satisfies the following formula: w c =σ(W2*δ(W1*GAP(F c ))) Among them, w c is the weight coefficient of the cth channel, F c is the cth channel of the multi-scale spatial feature map, GAP represents the global average pooling operation, W1 and W2 are the parameters of the fully connected layer, δ is the ReLU activation function, and σ is the Sigmoid function.

5. The method according to claim 1, wherein The S3 includes: S31: performing motion trajectory analysis on the human skeleton key point sequence in the feature graph set, calculating the abnormality index of the climbing action through a joint angle change rate model, and generating a climbing risk event; S32: performing tool contact detection processing on the feature map of the construction worker's hand area in the feature map set, identifying the behavior of touching a live object with an uninsulated tool through a contact relationship classification model, and generating an illegal operation event; S33: performing high-temperature region segmentation processing on the device state feature graph of the feature graph set, locating abnormal heating areas of the electrical equipment through a U-Net++ network, and generating a high-temperature region marking map; S34: performing temperature change rate analysis on the high temperature area marker image, determining a transient temperature rise abnormality event through a wavelet coefficient threshold, and generating an overheating fault event; S35: Aggregate the climbing risk events, the illegal operation events, and the overheating failure events to generate a construction risk set containing dangerous behavior events and equipment failure events. The dangerous behavior events include climbing risk events and illegal operation events, and the equipment failure events include overheating failure events.

6. The method according to claim 1, wherein The S4 includes: S41: performing spatial correlation processing on the construction risk set, determining the proximity relationship between events through three-dimensional Euclidean distance calculation, and generating a spatial correlation matrix; S42: performing time window analysis on the spatial correlation matrix, counting the cumulative frequency of events occurring in the same area through a sliding time window, and generating spatiotemporal coupling risk events; S43: Quantify the level of the spatiotemporal coupling risk event, calculate the comprehensive risk score through the risk weight accumulation model, and generate a risk level assessment result.

7. The method according to any one of claims 1 to 6, characterized in that The S5 includes: S51: Performing a yellow warning determination process on the comprehensive risk score of the risk level assessment result. If the comprehensive risk score is greater than a preset first risk threshold, generating and sending an on-site personnel warning information push signal to a mobile terminal of the on-site personnel; S52: Performing a red warning determination process on the comprehensive risk score of the risk level assessment result. If the comprehensive risk score is greater than a preset second risk threshold, generating and sending a device power cut-off instruction to the device terminal, and simultaneously generating and sending an evacuation route guidance signal to the mobile terminal of the on-site personnel; S53: Log the on-site personnel warning information push signal, the equipment power cut-off instruction, and the evacuation route guidance signal, generate an unalterable comprehensive early warning notice through blockchain hash chain storage, and send it to the construction unit's management terminal.

8. An electric power construction monitoring system, characterized in that: The system comprises: The data acquisition preprocessing module is used to perform time-space position calibration and anti-interference processing on the construction image data, temperature data, and dust concentration data collected by multi-source sensors at the construction site, eliminate sensor acquisition delays, compensate for spatial perspective differences, and suppress noise, thereby generating a preprocessed multi-source fusion data set; A multi-source feature extraction module is used to perform multi-level feature extraction processing on the multi-source fusion data set based on a neural network model, identify the movement trajectory of construction workers and the operating parameters of equipment, and generate a feature graph set including a behavior feature graph and an equipment status feature graph; An abnormal behavior recognition module is used to identify high-risk actions and equipment anomalies based on the feature graph set, analyze the movement patterns of human joints and the compliance of tool use, and generate a construction risk set containing dangerous behavior events and equipment failure events by detecting abnormal temperature changes and electrical spark characteristics of equipment; A comprehensive risk assessment module is used to perform spatiotemporal correlation analysis on the construction risk set, calculate event distances in three-dimensional space, and combine continuous monitoring in the time dimension to generate risk level assessment results. The risk level assessment results are used to quantify the comprehensive accident probability level within the construction area; The emergency response and warning module is used to perform graded warning processing based on the risk level assessment results, trigger different levels of emergency response instructions through preset risk thresholds, generate and transmit comprehensive warning notifications to the construction unit's management terminal, and the emergency response instructions include one of the following: on-site personnel warning information push signals, equipment power cut-off instructions, and evacuation route guidance signals.

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

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

Images