Power construction environment sensing method

Abstract

The invention discloses an electric power construction environment sensing system and method, the system comprises an electric power construction environment sensing system and an intelligent monitoring module, the electric power construction environment sensing system comprises a presentation layer, an intelligent monitoring layer, a communication layer and a data layer, ZigBee nodes are arranged at an electric power construction site, electric power construction safety quality information is collected through the ZigBee nodes, and the information is sent to the intelligent monitoring module. Information collected by the nodes is transmitted to the intelligent monitoring layer through a wireless receiving and transmitting module of the communication layer, an intelligent monitoring module of the intelligent monitoring layer processes data through a convolutional neural network, monitors whether a safety quality problem exists in a power construction site or not, transmits a monitoring result to the presentation layer and displays the monitoring result to a user, and a sensing-processing closed loop is completed. The image is enhanced by using a mosaic data enhancement mode, and the attention mechanism module is introduced, so that the accuracy of obtaining the visual image features of the power construction scene is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of electric power construction supervision, and in particular to an electric power construction environment perception and method. Background Art

[0002] With the rapid development of society and the economy, the demand for electricity continues to grow, and the number of power projects is increasing. During the construction of power projects, environmental factors have a significant impact on construction quality and progress. The power construction environment is highly complex, and manual methods are unable to effectively monitor the safety and quality of power construction.

[0003] A Chinese patent discloses a method and system for intelligent perception of the power transmission and transformation project environment (Announcement No. CN118196659A). This patented technology can realize intelligent perception of the power transmission and transformation project environment by dividing the system into a model building module, a parameter acquisition module, a multi-source feature acquisition module, a segmentation result acquisition module and an environment perception module. The modularization concept makes each module independent of each other, which facilitates the unified management of each module. However, the power construction site environment is complex and the efficiency of personnel inspection is low. When monitoring through images, when there is image blur and occlusion, the recognition accuracy is low and the feature accuracy needs to be improved. Therefore, the present invention provides a power construction environment perception and method to solve the problems raised in the above background technology. Summary of the Invention

[0004] The purpose of the present invention is to provide a power construction environment perception and method to solve the problems raised in the above background technology.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A method and system for sensing and understanding an electric power construction environment includes an electric power construction environment sensing system and an intelligent monitoring module. The electric power construction environment sensing system includes a presentation layer, an intelligent monitoring layer, a communication layer, and a data layer. ZigBee nodes are set at the electric power construction site to collect electric power construction safety and quality information through the ZigBee nodes. The information collected by the nodes is transmitted to the intelligent monitoring layer using a wireless transceiver module of the communication layer. The intelligent monitoring module of the intelligent monitoring layer processes data using a convolutional neural network to monitor whether there are safety and quality issues at the electric power construction site, and transmits the monitoring results to the presentation layer for display to the user.

[0007] The intelligent monitoring module combines deep learning technology to abstractly design the monitoring requirements of power construction scenarios, converts the corresponding monitoring targets into model inputs corresponding to deep learning, inputs the trained neural network model, and uses the target detection weights of the YOLOv5 algorithm to locate the target position. After using the convolutional network to obtain the sub-image features of the target area, it can output the violation judgment results related to the power operation scenario in combination with the corresponding judgment logic, and then perform binary classification processing on the images of dangerous behaviors, safe behaviors and facility layouts, and then use the output end to output the visual image detection results of the power construction scene.

[0008] Gradual control measures are taken for the test results. Gradual control means that the specific content of the construction environment is implemented in different professions. Professional experts from various professions formulate targeted on-site control plans or work cards, and divide the risk control levels into five levels: provincial company level, municipal company level, county company department, team and position. They are pushed separately and handled by the main person in charge of the corresponding level unit of the company. After processing, they are distributed to the relevant responsible persons of the rest of the supervision system and guarantee system through the system. It is necessary to implement job responsibilities and control on-site safety. The project responsible unit should appoint a dedicated person to supervise the entire project process, and the corresponding departments and teams will carry out control. After the control is completed, it will be recorded through the system to complete the perception-processing closed loop.

[0009] Preferably, the power construction safety and quality information includes personnel behavior and facility layout information, and data is collected through cameras and sensors.

[0010] Preferably, the system adopts B / S architecture, .NET platform, and is based on MVC layered design concept to avoid high coupling between system layers, making system development simpler and easier to maintain.

[0011] Preferably, when the test result is safe, the monitoring result is intuitively displayed with a green mark. When there are safety and quality issues, the system uses a yellow mark to display it to the monitoring personnel, who can conduct subsequent inspections and processing based on the monitoring results.

[0012] Preferably, the intelligent monitoring module collects power construction safety and quality information from the ZigBee node and inputs it. When the visual image of the power construction scene is blurred and occluded, the image is enhanced using a mosaic data enhancement method. The convolutional neural network obtains the visual image features of the power construction scene through the convolution layer of the YOLOv5 algorithm, extracts the collected power construction site images, extracts power construction safety and quality related features in the image, and sparsely extracts the extracted features to obtain the final power construction safety and quality monitoring results.

[0013] xi and yi are the neuron input and output safety quality signals of the power construction safety quality in the convolutional neural network, respectively. The relationship between the neuron input and output in the convolutional neural network is expressed as follows:

[0014]

[0015] Where ω and f() are the weight matrix and activation function respectively; b is the bias;

[0016] When the convolutional neural network monitors the safety quality of power construction, the loss function K is used to measure the deviation between the convolutional neural network and the actual safety quality monitoring results; the convolutional neural network weight matrix ω is updated using the gradient descent method, and the update expression is as follows:

[0017]

[0018] Where μ is the learning rate. The learning rate of the convolutional neural network is used to measure the learning efficiency and learning speed of the algorithm when monitoring the safety quality of power construction.

[0019] Preferably, since there are many visual image data sets of electric power construction scenes input to the input end of the YOLOv5 algorithm, in order to reduce the amount of computation for extracting image features at the input end, the separable convolution method is used to improve the input end. When the visual image of the electric power construction scene is blurred and occluded, the image is enhanced using a mosaic data enhancement method. The process of mosaic enhancing the image is: randomly selecting 4 images, randomly cropping them, splicing the 4 cropped images in a clockwise direction, and then scaling the spliced ​​images to obtain a clearer visual image of the electric power construction scene.

[0020] The process of using mosaic enhancement to improve the input of the YOLOv5 algorithm is as follows: the separable convolution method is used to convert m visual images of the power construction scene of size Uf*Uf into images of size Uk*Uk. At this time, the output of the YOLOv5 algorithm will output a network channel feature map of size. The feature map expression formula of the output of the YOLOv5 algorithm is as follows:

[0021]

[0022] Where Z represents the output visual image features of the electric power construction scene; k and l both represent the image resolution; m represents the number of channels; δ represents the convolution sum; F represents the input visual image features of the electric power construction scene; i and j represent the pixel coordinates of the feature map.

[0023] Preferably, the convolutional layer of the YOLOv5 algorithm is responsible for acquiring visual image features of power construction scenes. An attention mechanism module is introduced within this layer to improve its accuracy in acquiring visual image features of power construction scenes. Let F = TC * Q * W represent the intermediate feature map of a given visual image of a power construction scene, where Q, W, and C represent the height, width, and channel parameters, respectively. The attention mechanism is expressed as follows:

[0024]

[0025] Where Mc(F) represents the attention mechanism of the intermediate feature map of the given power construction scene visual image; σ represents the sigmoid function; Represent the feature maps in the average pooling and maximum pooling spaces respectively; μ0 and μ1 both represent the shared weights of the intermediate feature maps.

[0026] Introducing the spatial attention module into the above formula, we have:

[0027]

[0028] Among them, Ms(F) represents the first-level weight, Represent the average pooling and maximum pooling features within the channel respectively; f 6*6 Represents a 6×6 convolution operation. After the attention mechanism module is introduced into the convolution layer of the YOLOv5 algorithm, the connection formula of the convolutional neural network is as follows:

[0029] I e =φ e ([I0,I1,....,I e-1 ])

[0030] Among them, I e-1 Represents the residual network of the e-1th convolutional layer of the YOLOv5 algorithm; φ e Denotes the linear transformation operation of the residual network. After linearly transforming and normalizing the convolutional neural network using the above formula, the planning function within the convolutional network can be transferred to subsequent networks through multiple layers, facilitating the extraction of visual image features of power construction scenes at multiple scales.

[0031] Preferably, the system architecture of the power construction environment perception system is as follows:

[0032] (1) Presentation layer. The presentation layer is located at the top of the system and is the outermost layer displayed to the user. The system receives user-submitted data through the presentation layer. The presentation layer is an important operating interface for interaction between the system and the user. The power construction safety and quality monitoring results are intuitively displayed to the user through the presentation layer.

[0033] (2) Communication layer. The communication layer realizes remote control and data transmission of power construction safety and quality monitoring. The wireless transceiver module can realize the retrieval and control of data in the database server at different operation points. The system uses ZigBee communication to transmit monitoring data from different power construction sites to each layer of the system in real time.

[0034] (3) Intelligent Monitoring Layer. The intelligent monitoring layer is the core of the system and is an important part for processing various functional requests submitted by users of power construction safety and quality monitoring. The intelligent monitoring layer plays a crucial role in the system. The intelligent monitoring layer has high scalability and can also call data from the data layer.

[0035] (4) Data layer. The data layer is the data foundation for monitoring the safety and quality of power construction. It receives and processes data from various layers of the system and monitors the safety and quality of power construction in real time based on data analysis results. The system data layer can generate reports in real time, providing the system with the historical data required by users.

[0036] Preferably, the intelligent monitoring layer of the system utilizes the ASP.NET MVC framework as a web page development technology.

[0037] Preferably, the system selects ORACLE database as the system database, and ORACLE database can be well integrated with ASP.NET MVC framework.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] 1. The present invention monitors the power construction environment through the power construction environment perception system, perceives and judges personnel behavior and facility layout information, monitors whether there are safety and quality problems at the power construction site, and reports and handles them in a timely manner.

[0040] 2. When the visual image of the power construction scene is blurred and occluded, the mosaic data enhancement method is used to enhance the image, and an attention mechanism module is introduced to improve the accuracy of obtaining the visual image features of the power construction scene. DETAILED DESCRIPTION

[0041] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0042] In an embodiment of the present invention, a method for sensing an electric power construction environment includes an electric power construction environment sensing system and an intelligent monitoring module. The electric power construction environment sensing system includes a presentation layer, an intelligent monitoring layer, a communication layer, and a data layer. ZigBee nodes are set at the electric power construction site, and the ZigBee nodes are used to collect electric power construction safety and quality information. The information collected by the nodes is transmitted to the intelligent monitoring layer using a wireless transceiver module of the communication layer. The intelligent monitoring module of the intelligent monitoring layer processes data using a convolutional neural network to monitor whether there are safety and quality issues at the electric power construction site, and transmits the monitoring results to the presentation layer for display to users. The electric power construction safety and quality information includes personnel behavior and facility layout information, and the data is collected through cameras and sensors.

[0043] The intelligent monitoring module combines deep learning technology to abstractly design the monitoring requirements of power construction scenes, converts the corresponding monitoring targets into the corresponding model inputs of deep learning, inputs the trained neural network model, and uses the target detection weights of the YOLOv5 algorithm to locate the target position. After using the convolutional network to obtain the sub-image features of the target area, it can output the violation judgment results related to the power operation scene in combination with the corresponding judgment logic, and perform binary classification processing on the images of dangerous behaviors, safe behaviors and facility layouts. Then, the output end outputs the visual image detection results of the power construction scene. When the detection result is safe, the monitoring result is intuitively displayed with a green mark. When there is a safety and quality problem, the system uses a yellow mark to display it to the monitoring personnel. The monitoring personnel can conduct subsequent inspections and processing based on the monitoring results.

[0044] Gradual control measures are taken for the test results. Gradual control means that the specific content of the construction environment is implemented in different professions. Professional experts from various professions formulate targeted on-site control plans or work cards, and divide the risk control levels into five levels: provincial company level, municipal company level, county company department, team and position. They are pushed separately and handled by the main person in charge of the corresponding level unit of the company. After processing, they are distributed to the relevant responsible persons of the rest of the supervision system and guarantee system through the system. It is necessary to implement job responsibilities and control on-site safety. The project responsible unit should appoint a dedicated person to supervise the entire project process, and the corresponding departments and teams will carry out control. After the control is completed, it will be recorded through the system to complete the perception-processing closed loop.

[0045] The intelligent monitoring module inputs the power construction safety and quality information collected by the ZigBee node. When the visual image of the power construction scene is blurred or occluded, the mosaic data enhancement method is used to enhance the image. The convolutional neural network obtains the visual image features of the power construction scene through the convolution layer of the YOLOv5 algorithm, extracts the collected power construction site images, extracts the power construction safety and quality-related features in the images, and sparsifies the extracted features to obtain the final power construction safety and quality monitoring results.

[0046] x i with y i are the neuron input and output safety quality signals of power construction safety quality in the convolutional neural network. The relationship between neuron input and output in the convolutional neural network is as follows:

[0047]

[0048] Where ω and f() are the weight matrix and activation function respectively; b is the bias;

[0049] When the convolutional neural network monitors the safety quality of power construction, the loss function K is used to measure the deviation between the convolutional neural network and the actual safety quality monitoring results; the convolutional neural network weight matrix ω is updated using the gradient descent method, and the update expression is as follows:

[0050]

[0051] Where μ is the learning rate. The learning rate of the convolutional neural network is used to measure the learning efficiency and learning speed of the algorithm when monitoring the safety quality of power construction.

[0052] Input improvements:

[0053] Since the YOLOv5 algorithm inputs a large number of visual image datasets of power construction scenes, in order to reduce the amount of computation required to extract image features at the input end, the separable convolution method is used to improve the input end. When the visual images of the power construction scenes are blurred and occluded, the mosaic data enhancement method is used to enhance the images. The process of mosaic enhancement is as follows: 4 images are randomly selected, randomly cropped, and then the 4 cropped images are spliced ​​in a clockwise direction. The spliced ​​images are then scaled to obtain a clearer visual image of the power construction scene.

[0054] The process of improving the input of the YOLOv5 algorithm using mosaic enhancement is as follows: m pixels of size U are transformed into f *U f The visual image of the power construction scene is converted into a U k *U k The output of the YOLOv5 algorithm will output a network channel feature map of size . The feature map expression formula of the output of the YOLOv5 algorithm is as follows:

[0055]

[0056] Where Z represents the output visual image features of the electric power construction scene; k and l both represent the image resolution; m represents the number of channels; δ represents the convolution sum; F represents the input visual image features of the electric power construction scene; i and j represent the pixel coordinates of the feature map.

[0057] Convolutional layers improve extracted features:

[0058] The convolutional layer of the YOLOv5 algorithm is responsible for acquiring the visual image features of the power construction scene. The attention mechanism module is introduced into this layer to improve the accuracy of acquiring the visual image features of the power construction scene. Let F = T C*Q*W represents the intermediate feature map of a given visual image of a power construction scene, where Q, W, and C represent the height, width, and channel parameters, respectively. The attention mechanism is expressed as follows:

[0059]

[0060] Where Mc(F) represents the attention mechanism of the intermediate feature map of the given power construction scene visual image; σ represents the sigmoid function; They represent the feature maps in the average pooling and maximum pooling spaces respectively; μ0 and μ1 both represent the shared weights of the intermediate feature maps.

[0061] Introducing the spatial attention module into the above formula, we have:

[0062]

[0063] Among them, Ms(F) represents the first-level weight, Represent the average pooling and maximum pooling features within the channel respectively; f 6*6 Represents a 6×6 convolution operation. After the attention mechanism module is introduced into the convolution layer of the YOLOv5 algorithm, the connection formula of the convolutional neural network is as follows:

[0064]

[0065] Among them, I e-1 Represents the residual network of the e-1th convolutional layer of the YOLOv5 algorithm; φ e Denotes the linear transformation operation of the residual network. After linearly transforming and normalizing the convolutional neural network using the above formula, the planning function within the convolutional network can be transferred to subsequent networks through multiple layers, facilitating the extraction of visual image features of power construction scenes at multiple scales.

[0066] The system architecture of the power construction environment perception system is as follows:

[0067] (1) Presentation layer. The presentation layer is located at the top of the system and is the outermost layer displayed to the user. The system receives user-submitted data through the presentation layer. The presentation layer is an important operating interface for interaction between the system and the user. The power construction safety and quality monitoring results are intuitively displayed to the user through the presentation layer.

[0068] (2) Communication layer. The communication layer realizes remote control and data transmission of power construction safety and quality monitoring. The wireless transceiver module can realize the retrieval and control of data in the database server at different operation points. The system uses ZigBee communication to transmit monitoring data from different power construction sites to each layer of the system in real time.

[0069] (3) Intelligent Monitoring Layer. The intelligent monitoring layer is the core of the system and is an important part for processing various functional requests submitted by users of power construction safety and quality monitoring. The intelligent monitoring layer plays a crucial role in the system. The intelligent monitoring layer has high scalability and can call data from the data layer. The intelligent monitoring layer of the system uses the ASP.NET MVC framework as the web page development technology.

[0070] (4) Data layer. The data layer is the data foundation for monitoring the safety and quality of power construction. It receives and processes data from various layers of the system and monitors the safety and quality of power construction in real time based on data analysis results. The system data layer can generate reports in real time and provide the system with the historical data required by users. The system selects the ORACLE database as the system database, which can be well integrated with the ASP.NET MVC framework.

[0071] The system adopts B / S architecture, .NET platform, and is based on the MVC layered design concept to avoid high coupling between system layers, making system development simpler and easier to maintain.

[0072] The behaviors that this system can identify include:

[0073] ① Working at heights at pedestrian crossings or in densely populated areas without barriers (fences), designated personnel, or other safety measures. ② Workers working at heights casually throw tools, equipment, and materials up and down. ③ Working on ladders without a person holding them. ④ Failure to verify equipment names and numbers, the presence of live equipment in cabinets, panels, and boxes before work, and failure to check the foundation, pole base, ladders, and guy wires for proper function during line work. ⑤ Workers crossing or crossing safety fences or exceeding safety lines without authorization. ⑥ Failure to use safety belts as required for working at heights, or loss of fall protection during work. ⑦ Workers or equipment fail to maintain the required safe distances from live parts (including lightning arresters) during on-site work. Safety measures such as power outages, electrical testing, and grounding are not implemented as required. Workers are assigned to work at an inadequate safe distance from live equipment without implementing reliable safety measures. ⑧ Failure to test the grounding wire before installing it, installing the grounding wire at a location inconsistent with the test location, or failing to install or remove the grounding wire in the prescribed sequence. ⑨ Failure to strictly follow the "reverse order and item by item restoration" secondary measures list, resulting in safety measures being omitted or incorrectly restored. ⑩ Identification of illegal behavior such as adjusting or removing tower guy wires while someone is working on the line tower.

[0074] The working principle of the present invention is:

[0075] Real-time perception and early warning of frequent and dangerous violations during video surveillance of power construction are achieved. Safety and quality information of power construction is collected through ZigBee nodes. The information collected by the nodes is transmitted to the intelligent monitoring layer using the wireless transceiver module of the communication layer. The intelligent monitoring module of the intelligent monitoring layer uses convolutional neural networks to process data, monitor whether there are safety and quality issues at the power construction site, and transmit the monitoring results to the presentation layer for display to users.

[0076] The intelligent monitoring module combines deep learning technology to abstractly design the monitoring requirements of power construction scenes, converts the corresponding monitoring targets into the corresponding model inputs of deep learning, inputs the trained neural network model, and uses the target detection weights of the YOLOv5 algorithm to locate the target position. After using the convolutional network to obtain the sub-image features of the target area, it can output the violation judgment results related to the power operation scene in combination with the corresponding judgment logic, and perform binary classification processing on the images of dangerous behaviors, safe behaviors and facility layouts. Then, the output end outputs the visual image detection results of the power construction scene. When the detection result is safe, the monitoring result is intuitively displayed with a green mark. When there is a safety and quality problem, the system uses a yellow mark to display it to the monitoring personnel. The monitoring personnel can conduct subsequent inspections and processing based on the monitoring results.

[0077] Gradual control measures are taken for the test results. Gradual control means that the specific content of the construction environment is implemented in different professions. Professional experts from various professions formulate targeted on-site control plans or work cards, and divide the risk control levels into five levels: provincial company level, municipal company level, county company department, team and position. They are pushed separately and handled by the main person in charge of the corresponding level unit of the company. After processing, they are distributed to the relevant responsible persons of the rest of the supervision system and guarantee system through the system. It is necessary to implement job responsibilities and control on-site safety. The project responsible unit should appoint a dedicated person to supervise the entire project process, and the corresponding departments and teams will carry out control. After the control is completed, it will be recorded through the system to complete the perception-processing closed loop.

[0078] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A power construction environment perception and method, including a power construction environment perception system and an intelligent monitoring module, characterized in that: The power construction environment perception system includes a presentation layer, an intelligent monitoring layer, a communication layer, and a data layer. ZigBee nodes are set up at the power construction site to collect power construction safety and quality information through the ZigBee nodes. The information collected by the nodes is transmitted to the intelligent monitoring layer using the wireless transceiver module of the communication layer. The intelligent monitoring module of the intelligent monitoring layer uses convolutional neural networks to process data, monitor whether there are any safety and quality issues at the power construction site, and transmit the monitoring results to the presentation layer for display to users. The intelligent monitoring module uses deep learning technology to abstractly design monitoring requirements for power construction scenarios, convert the corresponding monitoring targets into model inputs corresponding to deep learning, input the trained neural network model, and use the target detection weights of the YOLOv5 algorithm to locate the target position. After using the convolutional network to obtain the sub-image features of the target area, it combines the corresponding judgment logic to output the violation judgment results related to the power operation scenario. It performs binary classification processing on images of dangerous behaviors, safe behaviors, and facility layouts, and then outputs the visual image detection results of the power construction scene through the output terminal; Gradual control measures are taken for the test results. Gradual control means that the specific content of the construction environment is implemented in different professions. Professional experts from various professions formulate targeted on-site control plans or work cards, and divide the risk control levels into five levels: provincial company level, municipal company level, county company department, team and position. They are pushed separately and handled by the main person in charge of the corresponding level unit of the company. After processing, they are distributed to the relevant responsible persons of the rest of the supervision system and guarantee system through the system. It is necessary to implement job responsibilities and control on-site safety. The project responsible unit should appoint a dedicated person to supervise the entire project process, and the corresponding departments and teams will carry out control. After the control is completed, it will be recorded through the system to complete the perception-processing closed loop.

2. The method for sensing and detecting an electric power construction environment according to claim 1, wherein: Power construction safety and quality information includes personnel behavior and facility layout information, and data is collected through cameras and sensors.

3. The method for sensing and detecting an electric power construction environment according to claim 1, wherein: The system adopts B / S architecture, .NET platform and MVC layered design.

4. The method for sensing and detecting an electric power construction environment according to claim 1, wherein: When the test result is safe, the monitoring result is displayed intuitively with a green mark. When there are safety and quality issues, the system uses a yellow mark to display it to the monitoring personnel. The monitoring personnel can conduct subsequent inspections and processing based on the monitoring results.

5. The method for sensing and detecting an electric power construction environment according to claim 1, wherein: The intelligent monitoring module collects power construction safety and quality information from ZigBee nodes and uses mosaic data enhancement to enhance the image when the visual image of the power construction scene is blurred or obscured. The convolutional neural network uses the convolutional layer of the YOLOv5 algorithm to obtain visual image features of the power construction scene, extract the collected power construction site images, extract power construction safety and quality-related features from the images, and then sparsify the extracted features to obtain the final power construction safety and quality monitoring results. xi and yi are the neuron input and output safety quality signals of the power construction safety quality in the convolutional neural network, respectively. The relationship between the neuron input and output in the convolutional neural network is expressed as follows: Where ω and f() are the weight matrix and activation function respectively; b is the bias; When the convolutional neural network monitors the safety quality of power construction, the loss function K is used to measure the deviation between the convolutional neural network and the actual safety quality monitoring results; the convolutional neural network weight matrix ω is updated using the gradient descent method, and the update expression is as follows: Where μ is the learning rate. The learning rate of the convolutional neural network is used to measure the learning efficiency and learning speed of the algorithm when monitoring the safety quality of power construction.

6. The method for sensing and detecting an electric power construction environment according to claim 5, characterized in that: When the visual image of the power construction scene is blurred and occluded, the mosaic data enhancement method is used to enhance the image. The process of mosaic enhancement is as follows: randomly select 4 images, randomly crop them, splice the 4 cropped images in a clockwise direction, and then scale the spliced ​​images to obtain a clearer visual image of the power construction scene. The process of using mosaic enhancement to improve the input of the YOLOv5 algorithm is as follows: the separable convolution method is used to convert m visual images of the power construction scene of size Uf*Uf into images of size Uk*Uk. At this time, the output of the YOLOv5 algorithm will output a network channel feature map of size. The feature map expression formula of the output of the YOLOv5 algorithm is as follows: Where Z represents the output visual image features of the electric power construction scene; k and l both represent the image resolution; m represents the number of channels; δ represents the convolution sum; F represents the input visual image features of the electric power construction scene; i and j represent the pixel coordinates of the feature map.

7. The method for sensing and detecting an electric power construction environment according to claim 5, characterized in that: The convolutional layer of the YOLOv5 algorithm is responsible for acquiring visual image features of power construction scenes. An attention mechanism module is introduced within this layer to improve its accuracy in acquiring visual image features of power construction scenes. Let F = TC*Q*W represent the intermediate feature map of a given visual image of a power construction scene, where Q, W, and C represent the height, width, and channel parameters, respectively. The attention mechanism is expressed as follows: Where Mc(F) represents the attention mechanism of the intermediate feature map of the given power construction scene visual image; σ represents the sigmoid function; Represent the feature maps in the average pooling and maximum pooling spaces respectively; μ0 and μ1 both represent the shared weights of the intermediate feature maps; Introducing the spatial attention module into the above formula, we have: Among them, Ms(F) represents the first-level weight, Represent the average pooling and maximum pooling features within the channel respectively; f 6 *6 Represents a 6×6 convolution operation. After the attention mechanism module is introduced into the convolution layer of the YOLOv5 algorithm, the connection formula of the convolutional neural network is as follows: AND e =φ e ([I0, I1,...., I e-1 ]) Among them, I e-1 Represents the residual network of the e-1th convolutional layer of the YOLOv5 algorithm; φ e Represents the linear transformation operation of the residual network. After the convolutional neural network is linearly transformed and normalized using the above formula, the planning function in the convolutional network can be transmitted to the subsequent network through the multi-layer network, which is convenient for extracting visual image features of power construction scenes at multiple scales.

8. The method for sensing and detecting an electric power construction environment according to claim 1, characterized in that: The system architecture of the power construction environment perception system is as follows: (1) Presentation layer: The presentation layer is located at the top of the system and is the outermost layer displayed to the user; (2) Communication layer: The communication layer realizes remote control and data transmission of power construction safety and quality monitoring. The wireless transceiver module can realize the retrieval and control of data in the database server at different operation points. The system uses ZigBee communication to transmit monitoring data from different power construction sites to each layer of the system in real time. (3) Intelligent monitoring layer: The intelligent monitoring layer is the core part of the system and is an important part for processing various functional requests submitted by users of power construction safety and quality monitoring; (4) Data layer: The data layer is the data basis for monitoring the safety and quality of power construction. The data layer receives and processes data from various layers of the system, and monitors the safety and quality of power construction in real time based on data analysis results. The system data layer can generate reports in real time to provide the system with the historical data required by users.

9. The method for sensing and detecting an electric power construction environment according to claim 1, characterized in that: The system's intelligent monitoring layer uses the ASP.NET MVC framework as web page development technology.

10. The method for sensing and detecting an electric power construction environment according to claim 1, characterized in that: The system selects ORACLE database as the system database, which can be well integrated with ASP.NET MVC framework.

Images