Power distribution network high-precision semantic segmentation and dynamic hidden danger distance measurement method and system

Abstract

The invention discloses a power distribution network high-precision semantic segmentation and dynamic hidden danger distance measurement method and system, and relates to the technical field of power distribution network hidden danger detection. Time sequence data analysis is carried out, and action time sequence characteristics are captured; establishing a three-dimensional model based on the three-dimensional point cloud and deep learning; and an intelligent early warning and management and control system is established. According to the method, the detection precision of the small target is improved, dynamic behavior time sequence signals are fully utilized, high-precision safe distance measurement based on the three-dimensional point cloud is realized, an efficient early warning and management and control system with expandability and adaptability is constructed, meanwhile, the robustness and generalization ability of the system are enhanced, and the method is suitable for large-scale popularization and application. And the dependence of manual intervention is reduced. In addition, the safety and reliability of the power distribution network can be improved, the maintenance cost is reduced, the working efficiency is improved, scientific and technological innovation is promoted, the emergency response capability is enhanced, sustainable development is supported, resource waste and environmental pollution are reduced, and comprehensive benefits are brought to companies and the society.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power distribution network hidden danger detection, in particular to a power distribution network high-precision semantic segmentation and dynamic hidden danger ranging method and system. BACKGROUND

[0002] Currently, power distribution network hidden danger detection mainly relies on manual inspection and traditional image processing technology. Manual inspection relies on a large number of manpower for on-site inspection, but this method has many limitations, especially in complex environments. Inspectors need to face bad weather, complex terrain, and high-altitude work, etc. factors, which makes the inspection efficiency low and there is a risk of missing hidden dangers. Especially in the power distribution network, there are many types of hidden dangers, and some hidden dangers (such as flying wires, fishing rods, etc. small targets) are difficult to be discovered by manual inspection in time, which brings hidden dangers to the safe operation of the power distribution network.

[0003] In addition, traditional image processing technology is mainly used to identify objects in images and determine their positions. Although this technology can identify hidden dangers in simple environments, its performance is not satisfactory in complex environments (such as strong winds, low light, or other adverse weather conditions). In particular, in the power distribution network area, problems such as occlusion and light changes often occur, which makes the traditional image processing technology less accurate when dealing with small targets (such as thin wires, flying wires, or other small objects).

[0004] In the prior art, hidden danger detection and safety distance measurement usually rely on two-dimensional image geometric calculation methods. This method ignores the complex three-dimensional spatial structure in the power distribution network, and it is difficult to fully consider the actual spatial relationship between the hidden danger target and the facility, resulting in large safety distance measurement errors. For example, when measuring the distance between the target and the power facility based on a two-dimensional image, it is easily affected by factors such as viewing angle and occlusion, and the measurement result is often inaccurate, thereby affecting the efficiency and accuracy of hidden danger identification and processing.

[0005] The existing power distribution network hidden danger detection method mainly relies on manual inspection and traditional image processing technology, although these methods are effective in some cases, but have the following significant shortcomings, which are difficult to meet the efficiency, accuracy and intelligentization requirements of power distribution network hidden danger detection:

[0006] (1) Insufficient detection accuracy: The existing hidden danger detection technology, especially when dealing with small targets (such as flying wires, wires, and small obstacles) in the power distribution network, has low accuracy. Traditional image processing technology performs poorly in complex backgrounds, especially when affected by factors such as environmental occlusion, light changes, and object reflections. For example, small targets are difficult to be identified or detected behind the occlusion, resulting in hidden dangers being missed. Therefore, the detection accuracy of the existing technology in the face of complex environments and multi-target scenarios is obviously insufficient, increasing the risk of hidden dangers not being discovered in time.

[0007] (2) Missing timing information: Dynamic hazards (such as personnel climbing power towers, construction work, etc.) are common in distribution networks, and existing technologies often ignore the timing characteristics of these hazards, making it difficult to effectively identify or predict dynamic behavior. For example, during construction, the movement of equipment and personnel can cause electric shock or damage to power facilities, but existing technologies fail to fully utilize the timing information in video sequences, resulting in an inability to accurately capture changes in dynamic behavior and thus failing to timely detect and prevent dynamic hazards. Existing systems lack deep analysis of time series data, making it difficult to accurately predict and timely handle dynamic hazards.

[0008] (3) Large spatial distance measurement error: Existing technologies for measuring safety distances in distribution networks often rely on two-dimensional images and simple geometric calculation methods, ignoring the complexity of three-dimensional space. Traditional methods cannot fully consider the spatial position relationship between power facilities and hazard targets, especially in complex terrain and obstructions, resulting in large measurement errors. For example, two-dimensional image-based distance measurement methods are limited by changes in viewing angle and interference from obstructions, making it difficult to accurately calculate the true distance between targets and power facilities, which can lead to errors in safety distance judgment and affect the timely handling and safety assessment of hazards.

[0009] (4) Poor robustness and generalization ability: Existing hazard detection models often have a significant decrease in accuracy under different environmental conditions, such as insufficient lighting, severe weather, and low temperatures. Existing technologies often cannot cope with complex real-world environments and are easily affected by factors such as changes in lighting, weather conditions, and equipment failures, resulting in poor robustness and generalization ability of the model. For example, in severe weather conditions such as storms and smog, image processing and detection algorithms have increased errors, making it difficult to detect hazards in a timely manner and affecting the safety and stability of the distribution network. Therefore, existing technologies are not stable enough when dealing with dynamic changes and complex environments.

[0010] (5) Slow processing speed: Traditional methods have relatively slow processing speeds, especially in scenarios that require real-time monitoring and rapid response. Traditional manual inspection and image processing methods are difficult to meet real-time requirements due to the need for extensive human intervention and complex calculation processes, resulting in low efficiency in hazard detection. In the dynamic environment of the distribution network, if hazards cannot be detected in the first instance, it may lead to damage to power facilities or the occurrence of safety accidents.

[0011] (6) It is difficult to realize intelligent management: The existing power distribution network hidden danger detection methods mostly rely on manual inspection or alarm systems based on simple rules, lack of intelligent management function. The hidden danger detection results often need manual analysis and judgment, lack of efficient automatic processing mechanism. The existing system usually cannot automatically generate hidden danger disposal suggestions or track hidden danger processing progress, resulting in more manual intervention in hidden danger treatment process, increasing management cost and error rate. Therefore, the existing technology cannot meet the demand of intelligent and automatic management of power distribution network. SUMMARY

[0012] In view of the above problems, the present application is proposed.

[0013] Therefore, the technical problem solved by the present application is that the existing manual inspection and traditional image processing method has the problems of insufficient detection accuracy, large spatial distance measurement error, poor robustness and generalization ability, slow processing speed, and how to realize intelligent management.

[0014] To solve the above technical problems, the present application provides the following technical scheme: a power distribution network high-precision semantic segmentation and dynamic hidden danger distance measurement method, including collecting data and preprocessing. Time series data analysis is performed to capture action time sequence features. A three-dimensional model is established based on three-dimensional point cloud and deep learning. An intelligent early warning and management system is established.

[0015] As a preferred scheme of the power distribution network high-precision semantic segmentation and dynamic hidden danger distance measurement method according to the present application, wherein: the collecting data and preprocessing includes introducing multi-scale feature extraction and small target detection network, storing the collected multi-modal data in the edge server, and performing real-time backup of the data.

[0016] As a preferred scheme of the power distribution network high-precision semantic segmentation and dynamic hidden danger distance measurement method according to the present application, wherein: the collecting data and preprocessing includes denoising, resolution optimization and multi-modal data alignment processing of the collected data, and combining hollow convolution and context information fusion based on the existing YOLO model.

[0017] As a preferred scheme of the power distribution network high-precision semantic segmentation and dynamic hidden danger distance measurement method according to the present application, wherein: the time series data analysis and action time sequence feature capture includes combining long short-term memory network LSTM and 3D convolution network to accurately capture and analyze personnel behavior trajectory and action time sequence features, and identify dynamic hidden danger behavior.

[0018] As a preferred scheme of the power distribution network high-precision semantic segmentation and dynamic hidden danger ranging method, the three-dimensional model is established based on three-dimensional point cloud and deep learning, which includes three-dimensional modeling of point cloud data and deep learning technology, construction of high-precision ranging model, and optimization combining K-D tree algorithm and deep learning model.

[0019] As a preferred scheme of the power distribution network high-precision semantic segmentation and dynamic hidden danger ranging method, the intelligent early warning and management system is established, which includes an intelligent early warning and management module, real-time generation of early warning information according to the type, urgency and specific location of hidden dangers, automatic allocation of processing tasks according to the level of hidden dangers, and tracking of the processing progress of hidden dangers by introducing a hierarchical early warning mechanism.

[0020] As a preferred scheme of the power distribution network high-precision semantic segmentation and dynamic hidden danger ranging method, the intelligent early warning and management system is established, which includes a closed-loop management system, data feedback according to the progress of hidden danger treatment, and continuous optimization of hidden danger detection and prediction model by machine learning method, automatic learning and improvement of prediction ability for future potential hidden dangers.

[0021] Another object of the present application is to provide a power distribution network high-precision semantic segmentation and dynamic hidden danger ranging system which can increase automatic identification, classification, ranging and early warning functions through one of the schemes, and solve the problems of low hidden danger detection efficiency and easy human error in current manual inspection and manual intervention.

[0022] As a preferred scheme of the power distribution network high-precision semantic segmentation and dynamic hidden danger ranging system, it includes a data acquisition module, a preprocessing module, an intelligent recognition module, a three-dimensional ranging module, and an early warning and management module; the data acquisition module is used to acquire multi-modal data around the power distribution network, including video, depth information and three-dimensional point cloud; the preprocessing module is used to denoise, optimize resolution and align multi-modal data of the collected data. The intelligent recognition module is used to enhance the recognition ability of small hidden danger targets, extract the continuous features of personnel actions, apply attention mechanism to focus on key action period, and filter irrelevant background information. The three-dimensional ranging module is used to construct a high-precision ranging model to realize accurate distance calculation of hidden danger position and power facilities. The early warning and management module is used to publish early warning information and push disposal suggestions in real time according to the type and urgency of hidden dangers, set a hierarchical early warning mechanism, trigger alarm according to the risk level of hidden dangers, integrate a task scheduling system, automatically allocate disposal tasks to relevant persons in charge and track the execution.

[0023] A computer device comprises a memory and a processor, the memory stores a computer program, and the processor executes the computer program to implement the steps of the power distribution network high-precision semantic segmentation and dynamic hidden danger ranging method.

[0024] A computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the steps of the power distribution network high-precision semantic segmentation and dynamic hidden danger ranging method.

[0025] The power distribution network high-precision semantic segmentation and dynamic hidden danger ranging method provided by the present application introduces advanced deep learning technology, time series data analysis, three-dimensional modeling and intelligent early warning system to solve the key defects in the prior art. Specifically, the present application aims to solve the following main technical problems:

[0026] (1) Improve the detection accuracy of small targets and dynamic hidden dangers: The prior art has insufficient accuracy in detecting small targets and dynamic hidden dangers (such as flying wires and fallen poles) under complex environmental conditions, severe occlusion and light changes, and has a high rate of missed detection. The present application introduces multi-scale feature extraction and small target detection network, and combines dilated convolution and context information fusion based on the existing YOLO model, which effectively enhances the detection accuracy of small targets in complex environments. By optimizing the detection algorithm, the present application can accurately identify and process small targets under different conditions, greatly reducing the rate of missed detection and improving the reliability and accuracy of hidden danger detection.

[0027] (2) Realize accurate identification and processing of dynamic hidden dangers: Traditional hidden danger detection methods ignore the time sequence characteristics of dynamic hidden dangers in power distribution networks, resulting in the inability to identify dynamic behaviors such as personnel climbing and construction operations under high real-time requirements. Existing technologies often fail to effectively analyze dynamic hidden dangers using time sequence information, often missing the opportunity. To solve this problem, the present application combines long short-term memory network (LSTM) and 3D convolution network to accurately capture and analyze personnel behavior trajectory and action time sequence characteristics, which can accurately identify dynamic hidden danger behaviors such as unsafe climbing and construction operations, ensuring that hidden dangers can be identified and disposed of in a timely manner, reducing the risk of accidents.

[0028] (3) Achieving high-precision three-dimensional safety distance measurement: Existing technologies mainly rely on two-dimensional images and simple geometric calculations when measuring the safety distance of the power distribution network, which leads to low measurement accuracy and is easily affected by factors such as viewing angle and obstruction. Especially in complex terrain and obstructed environments, traditional methods cannot fully consider the spatial relationships in three-dimensional space, affecting the judgment of safety distance. The invention uses three-dimensional modeling of point cloud data and deep learning technology to build a high-precision distance measurement model, and combines K-D tree algorithm and deep learning model for optimization, successfully solving the distance measurement error caused by complex terrain and viewing angle changes, ensuring the accurate calculation of the safety distance between hidden danger targets and power facilities, and automatically judging whether it meets national safety standards.

[0029] (4) Improving the robustness and generalization ability of the system: Existing technologies have significant performance degradation in complex environments, especially in low light and harsh weather conditions, and cannot maintain stable detection accuracy. The invention introduces multi-modal data fusion technology and reinforcement learning mechanism, significantly improving the adaptability and robustness of the system, enabling it to maintain high detection accuracy in different environmental conditions, especially in dynamic changes and harsh weather conditions. By optimizing the generalization ability of the model, the invention can adapt to the hidden danger detection needs of different regions and environments of the power distribution network, ensuring the stable operation of the system.

[0030] (5) Realizing intelligent early warning and efficient management and control: Existing technologies rely on manual intervention and simple rules for hidden danger alarm and disposal, lacking intelligent and automated processing mechanisms, resulting in hidden dangers not being effectively disposed of in a timely manner. The invention uses an intelligent early warning and control module to generate real-time warning information based on the type, urgency, and specific location of hidden dangers, and automatically assigns disposal tasks according to the level of hidden dangers. By introducing a hierarchical early warning mechanism, the system can quickly respond and track the progress of hidden danger disposal, ensuring that hidden dangers are timely and effectively addressed, reducing the need for manual intervention, and improving the efficiency and accuracy of hidden danger disposal.

[0031] (6) Realizing closed-loop management and data tracking: Traditional hidden danger management lacks effective tracking and closed-loop management mechanisms, making it difficult to monitor the progress and effectiveness of hidden danger disposal, and making it difficult to optimize intelligently after data accumulation. The invention uses a closed-loop management system to ensure that the entire process from discovery to disposal of each hidden danger is recorded and tracked. The system provides data feedback based on the progress of hidden danger disposal, and continuously optimizes hidden danger detection and prediction models through machine learning methods, enabling the system to automatically learn and improve its prediction ability for future potential hidden dangers when processing historical hidden danger data. This mechanism not only ensures that hidden dangers are addressed in a timely manner, but also effectively improves the ability to prevent future hidden dangers.

[0032] (7) Reduce the dependence on manual and improve the level of automation: the existing technology relies too much on manual inspection and manual intervention, resulting in low efficiency of hidden danger detection and easy human error. By introducing automatic recognition, classification, ranging and early warning functions, the present application greatly reduces the need for manual intervention, making the hidden danger detection and treatment process more intelligent and efficient. The system can automatically identify and evaluate the severity of the hidden danger, and push it to the relevant personnel for processing according to the type of hidden danger and the processing priority, reducing the possibility of human negligence and improving the overall processing efficiency and accuracy of hidden danger treatment. BRIEF DESCRIPTION OF DRAWINGS

[0033] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0034] Fig. 1 The overall flow chart of a power distribution network high-precision semantic segmentation and dynamic hidden danger ranging method provided for the first embodiment of the present application.

[0035] Fig. 2 The LSTM network structure diagram of a power distribution network high-precision semantic segmentation and dynamic hidden danger ranging method provided for the first embodiment of the present application.

[0036] Fig. 3 The feature pyramid module of a power distribution network high-precision semantic segmentation and dynamic hidden danger ranging method provided for the first embodiment of the present application. DETAILED DESCRIPTION

[0037] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should be within the scope of protection of the present application.

[0038] Embodiment 1, refer to Figs. 1-3 For an embodiment of the present application, a power distribution network high-precision semantic segmentation and dynamic hidden danger ranging method is provided, which comprises:

[0039] S1: Collect data and pre-process.

[0040] Further, collecting and preprocessing data includes storing the collected multi-modal data in the edge server and backing up the data in real time by introducing multi-scale feature extraction and small target detection network. Then, through denoising, resolution optimization and multi-modal data alignment processing of the collected data, and combining the existing YOLO model with the hollow convolution and context information fusion.

[0041] It should be noted that the specific operation is:

[0042] Step 1: Collect data

[0043] (1) Deploy fixed high-resolution cameras and laser radars in the distribution network area to monitor the all-weather environment in real time;

[0044] (2) Use unmanned aerial vehicles to regularly conduct inspections and collect data from corner areas that are difficult to cover through fixed equipment, including video, pictures and three-dimensional point cloud.

[0045] (3) Store the collected multi-modal data in the edge server and backup the data in real time.

[0046] (4) Cooperate with weather sensors to collect environmental parameters (such as light intensity, humidity, wind speed, etc.) to provide auxiliary basis for subsequent data processing.

[0047] Step 2: Preprocess data

[0048] (1) Frame the video and image data, remove redundant frames, and only keep key frames to improve computing efficiency.

[0049] (2) Apply bilateral filtering algorithm to denoise the collected image data, and use histogram equalization algorithm to optimize image contrast.

[0050] (3) Align the point cloud data, use ICP algorithm (Iterative Closest Point) to align the point cloud collected at different times, and generate a unified three-dimensional space model.

[0051] (4) Use optical flow method to preliminarily estimate the trajectory of dynamic targets in video data, and provide time sequence basis for subsequent analysis.

[0052] S2: Perform time series data analysis to capture action time sequence features.

[0053] Further, by combining long short-term memory network LSTM and 3D convolution network, the personnel behavior trajectory and action time sequence features can be accurately captured and analyzed, and dynamic hidden danger behaviors such as unsafe climbing and construction operation can be accurately identified, ensuring that hidden dangers can be identified and disposed in time, reducing the risk of accidents.

[0054] It should be noted that dynamic hazard identification utilizes a 3D convolutional neural network to extract the temporal features of consecutive frames in the video, identifying dynamic behaviors such as personnel climbing, construction operations, etc. The temporal features are input into a long short-term memory network (LSTM) as shown in Fig. 2 The action continuity and feature changes are further analyzed. The spatial position information and behavior trajectory of the hidden danger target are fused to construct a complete dynamic hidden danger analysis model. Finally, the hidden danger behavior category and its occurrence probability are output, providing detailed basis for subsequent disposal.

[0055] S3: Establish a three-dimensional model based on three-dimensional point cloud and deep learning.

[0056] Further, establishing a three-dimensional model based on three-dimensional point cloud and deep learning includes constructing a high-precision ranging model through three-dimensional modeling of point cloud data and deep learning technology, and combining K-D tree algorithm and deep learning model for optimization, successfully solving the ranging error problem caused by complex terrain and changes in viewing angle, ensuring that the safe distance between the hidden danger target and the power facility is accurately calculated, and automatically determining whether it meets the national safety standards.

[0057] It should be noted that the three-dimensional ranging and safety analysis steps are as follows:

[0058] (1) Based on point cloud data, generate a three-dimensional model of the hidden danger area, and realize fast positioning of hidden objects through K-D tree algorithm.

[0059] (2) Apply the deep learning ranging model to control the shortest path calculation error between the hidden danger target and the distribution network facility to the millimeter level.

[0060] (3) Combine relevant national standards to perform threshold judgment on the ranging results, and mark hidden danger areas that do not meet the safe distance.

[0061] (4) Output detailed hidden danger distance report and automatically generate processing suggestions.

[0062] S4: Establish an intelligent early warning and control system.

[0063] Further, establishing an intelligent early warning and control system includes generating real-time warning information through an intelligent early warning and control module, combining the type, urgency and specific location of the hidden danger, automatically assigning processing tasks according to the level of the hidden danger, and introducing a hierarchical early warning mechanism to enable the system to quickly respond and track the processing progress of the hidden danger, ensuring that the hidden danger is timely and effectively managed, reducing the need for manual intervention, and improving the efficiency and accuracy of hidden danger processing;

[0064] Also included is a closed-loop management system that ensures the entire process of each hidden danger from discovery to governance can be recorded and tracked. The system will provide data feedback based on the progress of hidden danger governance, and continuously optimize the hidden danger detection and prediction model through machine learning methods, so that the system can automatically learn and improve the prediction ability of potential hidden dangers in the future when processing historical hidden danger data. This mechanism not only ensures that hidden dangers are timely treated, but also effectively improves the subsequent hidden danger prevention ability.

[0065] It should be noted that, in order to establish intelligent early warning and closed-loop management and control:

[0066] (1) According to the severity of the hidden danger, the system triggers real-time warning, and is divided into emergency, urgent and general three disposal levels.

[0067] (2) Push the disposal notice to the relevant person in charge, and clearly indicate the hidden danger category and processing time limit.

[0068] (3) The system automatically generates hidden danger disposal suggestions (such as increasing the height of the tower, adding an insulating layer, and migrating the line) and records the disposal progress.

[0069] (4) After the disposal is completed, data collection and analysis are performed again to verify the governance effect.

[0070] (5) After the hidden danger governance is completed, the system uploads the governance information to the cloud account and updates the hidden danger management record.

[0071] (6) Regularly inspect the completed hidden danger governance points to ensure that the hidden danger does not recur.

[0072] (7) Based on long-term data accumulation, predict future potential hidden dangers through machine learning and provide early warning.

[0073] Embodiment 2, which is different from the first two embodiments, is a second embodiment of the present application.

[0074] If the function is realized in the form of a software function unit and sold or used as an independent product, it can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application essentially or say the part that contributes to the prior art or part of the technical solutions can be embodied in the form of a software product, which is stored in a storage medium and includes a number of instructions to make a computer device (which can be a personal computer, server, or network device, etc.) execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disk, and various media that can store program codes.

[0075] The logic and / or steps represented in flow diagrams or otherwise described herein, for example, can be considered as a sequence of executable instructions for implementing logic functions, and can be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, processor-containing system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions, or a combination of them. The "computer-readable medium" can be any media that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device. The computer-readable medium can be, for example but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, device, or propagation medium. More specific examples (a non-exhaustive list) of the computer-readable medium include the following: an electrical connection having one or more wires (electronic), a portable computer diskette (magnetic), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber (optical), and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium can even be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, for example via the optical scanner of a device, and then compiled, interpreted, or otherwise processed in a suitable manner, if necessary, and stored in a computer memory.

[0076] More specific examples (a non-exhaustive list) of the computer-readable medium include the following: an electrical connection having one or more wires (electronic), a portable computer diskette (magnetic), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber (optical), and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium can even be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, for example via the optical scanner of a device, and then compiled, interpreted, or otherwise processed in a suitable manner, if necessary, and stored in a computer memory.

[0077] It should be understood that portions of the present application can be implemented in hardware, software, firmware, or combinations thereof. In the above embodiments, a number of steps or methods can be implemented in software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented in hardware, and as in another embodiment, it can be implemented using any or a combination of the following technologies, which are well known in the art: a discrete logic circuit having logic gates for implementing logic functions on data signals, an application specific integrated circuit having appropriate combinational logic gates, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0078] Embodiment 3, which is an embodiment of the present application, provides a power distribution network high-precision semantic segmentation and dynamic hidden danger ranging system, comprising a data acquisition module, a preprocessing module, an intelligent recognition module, a three-dimensional ranging module, a pre-warning and control module;

[0079] The data acquisition module is configured to acquire multi-modal data around the power distribution network, including video, depth information, and three-dimensional point cloud.

[0080] The preprocessing module is used for denoising, resolution optimization and multi-modal data alignment of the collected data. The intelligent recognition module is used for enhancing the recognition ability of small hidden danger targets, and extracting the continuous features of personnel actions, applying the attention mechanism to focus on the key action period, and filtering irrelevant background information. The three-dimensional ranging module is used for constructing a high-precision ranging model to realize accurate distance calculation of hidden danger position and power facilities. The early warning and control module is used for real-time release of early warning information and push of disposal suggestions according to the hidden danger type and emergency degree, setting of a hierarchical early warning mechanism, hierarchical triggering of alarm according to the hidden danger risk level, integration of a task scheduling system, automatic allocation of disposal tasks to relevant persons in charge and tracking of the execution situation.

[0081] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application rather than limit the present application. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced equivalently without departing from the spirit and scope of the technical solutions of the present application, which should be covered in the scope of the claims of the present application.

Claims

1. A method for high-precision semantic segmentation and dynamic hazard ranging in power distribution networks, characterized in that, include: Collect data and preprocess it; Perform time-series data analysis to capture action time-series characteristics; 3D models are built based on 3D point clouds and deep learning; Establish an intelligent early warning and control system.

2. The high-precision semantic segmentation and dynamic hazard ranging method for power distribution networks as described in claim 1, characterized in that: The data collection and preprocessing process includes storing the collected multimodal data in an edge server by introducing a multi-scale feature extraction and small target detection network, and performing real-time data backup.

3. The high-precision semantic segmentation and dynamic hazard ranging method for power distribution networks as described in claim 2, characterized in that: The data collection and preprocessing includes denoising, resolution optimization, and multimodal data alignment of the collected data, and combining dilated convolution and contextual information fusion based on the existing YOLO model.

4. The high-precision semantic segmentation and dynamic hazard ranging method for power distribution networks as described in claim 3, characterized in that: The aforementioned time-series data analysis, which captures action time-series features, includes combining a Long Short-Term Memory (LSTM) network and a 3D convolutional network to accurately capture and analyze personnel behavior trajectories and action time-series features, thereby identifying dynamic potential hazards.

5. The high-precision semantic segmentation and dynamic hazard ranging method for power distribution networks as described in claim 4, characterized in that: The establishment of a 3D model based on 3D point cloud and deep learning includes constructing a high-precision ranging model by using 3D modeling of point cloud data and deep learning technology, and optimizing it by combining KD tree algorithm and deep learning model.

6. The method for high-precision semantic segmentation and dynamic hazard ranging in power distribution networks as described in claim 5, characterized in that: The establishment of the intelligent early warning and control system includes generating early warning information in real time through the intelligent early warning and control module, combining the type, urgency and specific location of the hidden danger, and automatically assigning handling tasks according to the level of the hidden danger. At the same time, the system tracks the progress of the handling of hidden dangers by introducing a hierarchical early warning mechanism.

7. The method for high-precision semantic segmentation and dynamic hazard ranging in power distribution networks as described in claim 6, characterized in that: The establishment of the intelligent early warning and control system also includes using a closed-loop management system to provide data feedback based on the progress of hazard management, and continuously optimizing the hazard detection and prediction model through machine learning methods to automatically learn and improve the ability to predict potential future hazards.

8. A system employing the high-precision semantic segmentation and dynamic hazard ranging method for power distribution networks as described in any one of claims 1 to 7, characterized in that: It includes a data acquisition module, a preprocessing module, an intelligent identification module, a three-dimensional ranging module, and an early warning and control module; The data acquisition module is used to collect multimodal data from the surrounding power distribution network, including video, depth information, and 3D point cloud. The preprocessing module is used to perform noise reduction, resolution optimization, and multimodal data alignment on the collected data; The intelligent identification module is used to enhance the ability to identify small potential hazards, extract continuous features of personnel movements, apply an attention mechanism to focus on key action periods, and filter out irrelevant background information. The three-dimensional ranging module is used to construct a high-precision ranging model to realize accurate distance calculation between the location of potential hazards and power facilities; The early warning and control module is used to issue early warning information and push disposal suggestions in real time according to the type and urgency of the hidden danger, and to set up a graded early warning mechanism to trigger alarms according to the risk level of the hidden danger. It also integrates a task scheduling system to automatically allocate disposal tasks to relevant responsible persons and track the execution status.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the high-precision semantic segmentation and dynamic hazard ranging method for power distribution networks as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the high-precision semantic segmentation and dynamic hazard ranging method for power distribution networks as described in any one of claims 1 to 7.

Images