Power hidden danger identification method based on image recognition

Abstract

The present application discloses a method for identifying electric hidden dangers based on image recognition, including: obtaining a collection point sequence of a patrol robot in a preset electric scene model, the collection point sequence includes a number of collection points arranged in a preset order, and the collection coordinates include the shooting angle of the patrol robot; controlling the patrol robot to move to each collection point in turn to obtain an actual electric equipment image; based on the actual electric equipment image corresponding to each collection coordinate, determining the target electric equipment image in the electric equipment sample image library according to its shooting angle, inputting the actual electric equipment image and the target electric equipment image into the electric hidden danger identification model, and outputting the hidden danger result; displaying the hidden danger result of each actual electric equipment image at the target position in the electric scene model. In this way, the collection information of different electric equipment in the electric scene can be differentiated, the inspection time cost can be reduced, and the hidden danger identification efficiency and accuracy can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of image recognition, and in particular to a method for identifying hidden dangers of electric power based on image recognition. Background Art

[0002] The power system is one of the most important achievements in the history of human engineering science. It is a power production and consumption system composed of power generation, transmission, transformation, distribution and power consumption. It converts primary energy in nature into electricity through mechanical energy devices, and then supplies electricity to various users through transmission, transformation and distribution. Although the power system provides convenience to users, there are also related power hazards. For example, some power equipment will not directly cause problems when a hidden danger fault occurs, but if the fault persists, it may cause leakage electric shock, equipment damage, etc., causing casualties or financial losses.

[0003] The Chinese invention patent with patent application number 202110623439.X discloses a method and system for identifying power hazards based on image recognition. The method extracts a second device image from a three-dimensional image model according to a preset ratio and a certain shooting angle, and then uses the power equipment image and the second device image to identify the equipment hazards of the power equipment, and uses the power equipment image, the second device image, the shooting angle and the first coordinate position to determine the corresponding second coordinate position and display it in the scene model; the equipment hazards are identified based on the second device image, making the identification of equipment hazards more accurate.

[0004] However, due to the randomness of the shooting angles of power equipment, a single and untargeted shooting angle may not be able to fully monitor the power equipment and obtain the best image reflecting the status of the power equipment, which may lead to missing potential hazards and affect the accuracy of potential hazards identification. In addition, in order to obtain more comprehensive image information of power equipment, the inspection robot performs multi-angle image shooting, which not only increases the monitoring cost, but also reduces the efficiency of the inspection robot in potential hazards detection, and fails to discover and avoid potential hazards in time. Summary of the invention

[0005] This application provides a method for identifying power hazards based on image recognition, which can differentiate the collected information of different power equipment in power scenarios, reduce inspection time costs, and improve the efficiency and accuracy of hazard identification.

[0006] This application provides a method for identifying power hazards based on image recognition, including:

[0007] S101, obtaining a collection point sequence of the inspection robot in a preset power scene model, the collection point sequence including a number of collection points arranged in a preset order, each collection point corresponding to a position coordinate and at least one collection coordinate, each collection coordinate corresponding to a power equipment number, and the collection coordinate including a shooting angle of the inspection robot;

[0008] S102, controlling the inspection robot to move to each collection point in turn, and obtaining an actual power equipment image based on the collection coordinates corresponding to the collection point;

[0009] S103, based on the actual power equipment image corresponding to each acquisition coordinate, determine the target power equipment image in the power equipment sample image library according to its shooting angle, convert the actual power equipment image and the target power equipment image into a pixel matrix and input it into a pre-trained power hidden danger identification model, and output the hidden danger result of the power equipment corresponding to the acquisition coordinate;

[0010] S104, displaying the hidden danger result of each actual power equipment image at a target position in the power scene model, where the target position is determined as: the power equipment corresponding to the acquisition coordinates of the actual power equipment image.

[0011] Preferably, the method for acquiring the collection point sequence includes:

[0012] A1. According to the distribution and inspection requirements of power equipment, a moving path of the inspection robot is pre-set to ensure that the moving path can traverse all power equipment;

[0013] A2. For each power equipment, determine its key area according to its structure and operation characteristics. The key area is set as the area where all parts that can affect the operation of the power equipment are located;

[0014] A3. Based on the direction in which the key area of ​​each power device faces, map it to the ground moving path to obtain a first range, and determine the center point of the moving path defined by the first range as the collection point of the power device;

[0015] A4. According to the moving path of the inspection robot and all the determined collection points, the collection points are arranged in sequence to generate a collection point sequence.

[0016] Preferably, the method for determining the first range in A3 includes:

[0017] Draw a perpendicular line from the center point of the key area to the moving path, use the perpendicular line as the reference line, extract each point on the key area contour, emit a ray parallel to the reference line to the ground, and determine the area enclosed by the intersection of all rays and the ground as the first range.

[0018] Preferably, the power scenario model is provided with a two-dimensional coordinate system for representing the position coordinates of the power equipment and the collection point, the lower left corner of the power scenario model is set as the origin of the two-dimensional coordinate system, the x-axis represents the lateral distance of the power scenario model on the horizontal plane, starting from the origin, the right direction is the positive direction, and the y-axis represents the longitudinal distance of the power scenario model on the horizontal plane, starting from the origin, the upward direction is the positive direction;

[0019] The acquisition method of the acquisition coordinates includes:

[0020] Based on the two-dimensional coordinate system of the power scene model, the collection coordinates of each power equipment are generated, expressed as (a, b, c), where a is the coordinate value of the collection point on the x-axis, b is the coordinate value of the collection point on the y-axis, and c is the preset shooting angle value when the inspection robot is at the collection point. The preset shooting angle value is determined as follows: the inspection robot is simulated and placed at the collection point corresponding to the power equipment, and the camera is facing the center point of the key area of ​​the power equipment.

[0021] Preferably, the hidden danger results include normal equipment and equipment failure, and the pre-trained power hidden danger identification model is specifically constructed as follows:

[0022] A large number of training sets are obtained, wherein the training sets include actual power equipment images for training and sample power equipment images for training as input, and hidden danger result labels as output;

[0023] The training set is used to train and optimize the preset network structure model to obtain a power hazard identification model.

[0024] Preferably, the power equipment image library is composed of three-dimensional image models of various power equipment, and the actual power equipment image corresponding to each acquisition coordinate is determined in the power equipment image library according to its shooting angle, specifically including:

[0025] S201, according to the electric power equipment number corresponding to the collected coordinates, determining the three-dimensional image model corresponding to the electric power equipment number in the electric power equipment image library;

[0026] S202, obtaining two-dimensional images of the three-dimensional image model at various angles, and performing similarity calculations with actual power equipment images respectively;

[0027] S203: Determine the two-dimensional image with the highest similarity to the actual power equipment image as the target power equipment image.

[0028] Preferably, after S103, the method further includes:

[0029] S301, obtaining a key area of ​​each power equipment in advance, and dividing the key area into a plurality of sub-areas according to the component type of the power equipment, each sub-area corresponding to a component type;

[0030] S302, comparing the actual power equipment image with the key area, and calculating the effective contribution value of the acquisition coordinates according to the following formula:

[0031]

[0032] Among them, p is the effective contribution value of the acquisition coordinates, n is the total number of sub-areas in the key area of ​​the corresponding power equipment, is the criticality value of the operation of the power equipment in the i-th sub-area, is the pixel area size of the i-th sub-region, is the pixel area size of the ith sub-region in the actual power equipment image, is the hit rate of the i-th sub-region;

[0033] S303, if the effective contribution value of the acquisition coordinate is less than a preset threshold, using a preset image association guidance algorithm to determine an associated power equipment image of the acquisition coordinate;

[0034] S304: Use the associated power equipment image as the new actual power equipment image of the acquisition coordinates, and execute step S103 again.

[0035] Preferably, the image association guidance algorithm in S303 is specifically:

[0036] S401, determining the associated orientation of the acquisition coordinates according to the hit rate of each sub-region of the key region of the actual power equipment image corresponding to the acquisition coordinates;

[0037] S402: Determine the associated acquisition coordinates according to the associated orientation of the acquisition coordinates, and determine the actual power equipment image corresponding to the associated acquisition coordinates as the associated power equipment image.

[0038] Preferably, the association guidance of the acquisition coordinates is specifically:

[0039] The maximum angle direction formed by a plurality of rays formed by the sub-region with the highest hit rate pointing to each sub-region with a low hit rate is determined as the associated guide;

[0040] The hit rate is determined based on a preset hit rate threshold, which is set to 70%.

[0041] Preferably, in S402, determining the associated acquisition coordinates according to the associated orientation of the acquisition coordinates specifically includes:

[0042] With the current acquisition coordinate as the center, a sector-shaped area is extended toward the associated guide, and the nearest acquisition coordinate is selected in the sector-shaped area as the associated acquisition coordinate of the current acquisition coordinate.

[0043] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0044] By introducing a preset power scene model, the positional relationship of all power equipment in the power scene is displayed, and a two-dimensional coordinate system is set to accurately represent the position of power equipment and collection points, avoiding repeated collection and omissions, and improving the efficiency of robot inspections; by accurately planning collection points and parameters, unnecessary shooting and data processing are reduced, reducing monitoring costs; the hidden danger results of all power equipment are displayed in the power scene model, making the hidden danger distribution and inspection status clear at a glance, facilitating subsequent processing and decision-making;

[0045] The acquisition method of the collection point sequence is more systematic. The moving path is preset according to the distribution of power equipment and inspection requirements. The key areas of each power equipment are mapped to the collection points on the moving path, and the collection point sequence is generated by arranging them in sequence. Each power equipment corresponds to a unique collection coordinate at the collection point, including the position coordinate and the shooting angle value, which ensures the accuracy and pertinence of the image acquisition of each power equipment, differentiates the key characteristics between different power equipment, and improves the accuracy of hidden danger identification.

[0046] For the actual power equipment image collected by the collection coordinates of each power equipment, combined with the pre-set key areas, the importance of different sub-areas in the key areas is differentiated, so as to calculate the effective contribution value of the collection coordinates, which is used to evaluate the information completeness of the collected actual power equipment image;

[0047] When the effective contribution value of the collected coordinates is less than the preset threshold, in order to make up for the loss of key area information caused by the collected coordinates, the image association guidance algorithm is used to determine the associated power equipment image to supplement the missing information. This ensures that even when the information is incomplete at certain shooting angles, the information can be supplemented through associated images, thereby improving the reliability and comprehensiveness of hidden danger identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 The figure is a flow chart of a method for identifying electric power hidden dangers based on image recognition according to an embodiment of the present invention. DETAILED DESCRIPTION

[0049] To facilitate the understanding of the present invention, the present application will be described more comprehensively below with reference to the relevant drawings; the drawings show preferred embodiments of the present invention, but the present invention can be implemented in many different forms and is not limited to the embodiments described herein; on the contrary, the purpose of providing these embodiments is to enable a more thorough and comprehensive understanding of the disclosed content of the present invention.

[0050] It should be noted that the terms “vertical”, “horizontal”, “up”, “down”, “left”, “right” and similar expressions used in this document are only for illustrative purposes and do not represent the only implementation method.

[0051] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by technicians in the technical field to which the present invention belongs; the terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention; the term "and / or" used herein includes any and all combinations of one or more related listed items.

[0052] Embodiment 1: Figure 1 It is a flowchart of a method for identifying electric power hidden dangers based on image recognition according to an embodiment of the present invention.

[0053] like Figure 1 As shown, the method for identifying power hidden dangers based on image recognition includes the following steps:

[0054] S101, obtaining a collection point sequence of the inspection robot in a preset power scene model, the collection point sequence includes a number of collection points arranged in a preset order, each collection point corresponds to a position coordinate and at least one collection coordinate, and each collection coordinate corresponds to a power equipment number.

[0055] Among them, the preset power scene model is used to display the positional relationship of all power equipment in the power scene. The power scene model is established using three-dimensional modeling software. The size of the power scene model is set according to the preset ratio with the actual power scene to ensure that each power equipment and collection point in the power scene model can correspond one-to-one in proportion to the actual power scene. The preset ratio is set according to the actual scene requirements.

[0056] Specifically, the power scenario model is set with a two-dimensional coordinate system to represent the location coordinates of power equipment and collection points. The lower left corner of the power scenario model is set as the origin of the two-dimensional coordinate system. The x-axis represents the lateral distance of the power scenario model on the horizontal plane. Starting from the origin, the right is the positive direction. The y-axis represents the longitudinal distance of the power scenario model on the horizontal plane. Starting from the origin, the upward direction is the positive direction.

[0057] In some embodiments, the method for acquiring the collection point sequence is specifically as follows:

[0058] A1. According to the distribution of power equipment and inspection requirements, a moving path of the inspection robot is pre-set to ensure that the moving path can traverse all power equipment.

[0059] A2. For each power equipment, determine its critical area according to its structure and operation characteristics. The critical area is set as the area where all parts that can affect the operation of the power equipment are located.

[0060] A3. Based on the direction in which the key area of ​​each electric device faces, map it to the ground moving path to obtain a first range, and determine the center point of the moving path defined by the first range as the collection point of the electric device.

[0061] Specifically, step A3 includes: drawing a perpendicular line from the center point of the key area (located at the geometric center of the key area) to the moving path, and using the perpendicular line as a reference line (defining the direction of the projection of the key area onto the ground); emitting a ray parallel to the reference line from each point on the contour of the key area to the ground, and determining the area enclosed by the intersection points of all rays with the ground as the first range.

[0062] Among them, if the overlapping area between the first ranges of two or more power devices is greater than the preset area threshold, the center point of the moving path defined by the corresponding overlapping range is determined as the collection point of the two or more power devices, that is, one collection point corresponds to at least one power device.

[0063] Therefore, the determination of the collection points ensures that the inspection robot covers and pays attention to the key areas of all power equipment to the maximum extent.

[0064] A4. According to the moving path of the inspection robot and all the determined collection points, the collection points are arranged in sequence to generate a collection point sequence.

[0065] Specifically, each collection point corresponds to at least one collection coordinate, each collection coordinate corresponds to an electric power device, and the collection coordinates are obtained in the following manner:

[0066] Based on the two-dimensional coordinate system of the power scene model, the collection coordinates of each power equipment are generated, expressed as (a, b, c), where a is the coordinate value of the collection point on the x-axis, b is the coordinate value of the collection point on the y-axis, and c is the preset shooting angle value when the inspection robot is at the collection point. The preset shooting angle value is determined as follows: the inspection robot is simulated and placed at the collection point corresponding to the power equipment, and the camera is facing the center point of the key area of ​​the power equipment.

[0067] S102, controlling the inspection robot to move to each collection point in turn, and obtaining an actual power equipment image based on the collection coordinates corresponding to the collection point.

[0068] S103, based on the actual power equipment image corresponding to each acquisition coordinate, determine the target power equipment image in the power equipment sample image library according to its shooting angle, convert the actual power equipment image and the target power equipment image into a pixel matrix and input it into a pre-trained power hazard identification model, and output the hazard result of the power equipment corresponding to the acquisition coordinate.

[0069] Among them, hidden danger results include normal equipment and equipment failure.

[0070] Specifically, the pre-trained power hazard recognition model first converts the input image into a pixel matrix to generate a feature vector, and then performs recognition based on the feature vector. The specific construction method of the power hazard recognition model is as follows:

[0071] A large number of training sets are obtained, including actual training power equipment images and training power equipment sample images as input, and hidden danger result labels as output; the preset network structure model is iteratively trained using the training set, the model parameters are updated through the back propagation algorithm, the model parameters are optimized, and the power hidden danger identification model is obtained.

[0072] In some embodiments, the power equipment image library is composed of three-dimensional image models of various power equipment. Based on the actual power equipment image corresponding to each acquisition coordinate, the target power equipment image is determined in the power equipment image library according to its shooting angle, specifically including:

[0073] S201, according to the electric power equipment number corresponding to the collected coordinates, determining the three-dimensional image model corresponding to the electric power equipment number in the electric power equipment image library.

[0074] S202, obtaining two-dimensional images of the three-dimensional image model at various angles, and performing similarity calculations with the actual power equipment images respectively. The similarity calculations may be performed based on contour features and pixel distribution features of the images, which are not elaborated or limited in the present invention.

[0075] S203: Determine the two-dimensional image with the highest similarity to the actual power equipment image as the target power equipment image.

[0076] S104, displaying the hidden danger result of each actual power equipment image at a target position in the power scene model, where the target position is determined as: the power equipment corresponding to the acquisition coordinates of the actual power equipment image.

[0077] Specifically, the power scenario model is used to display the hidden danger results of each power equipment currently monitored by the inspection robot, and synchronize the number of the power equipment whose hidden danger result is equipment failure to the administrator side, so that the administrator side can carry out timely maintenance of the power equipment.

[0078] However, in actual power scenarios, although the actual power equipment images are collected using a single collection coordinate for each power equipment, which can reduce the inspection time cost, it is impossible to completely avoid the possibility that a single shooting angle cannot obtain complete information about the key areas of the power equipment, thereby leading to inaccurate or missed hidden dangers identification.

[0079] In some embodiments, after step S103, the method further includes:

[0080] S301, obtaining a key area of ​​each electric power equipment in advance, dividing the key area into a plurality of sub-areas according to the component type of the electric power equipment, each sub-area corresponding to a component type.

[0081] The key areas of the power equipment are divided in advance according to the actual situation and professional experience, and the key areas and corresponding sub-areas of each power equipment are obtained. The key areas and sub-areas of each power equipment are archived and stored in the power equipment sample image library, which can be retrieved and analyzed according to the power equipment number; for example, the key areas of the power transformer include the oil tank area, the cooling system sub-area, the core sub-area, etc.

[0082] S302, comparing the actual power equipment image with the key area, and calculating the effective contribution value of the acquisition coordinates according to the following formula:

[0083]

[0084] Among them, p is the effective contribution value of the acquisition coordinates, n is the total number of sub-areas in the key area of ​​the corresponding power equipment, is the criticality value of the operation of the power equipment in the i-th sub-area, is the pixel area size of the i-th sub-region, is the pixel area size of the ith sub-region in the actual power equipment image, is the hit rate of the ith sub-region. By calculating the hit rate, we can understand the information completeness of each sub-region in the actual power equipment image.

[0085] S303: If the effective contribution value of the collected coordinates is less than a preset threshold, a preset image association guidance algorithm is used to determine an associated power equipment image of the collected coordinates.

[0086] S304: Use the associated power equipment image as the new actual power equipment image of the acquisition coordinates, and execute step S103 again.

[0087] In some embodiments, the image association guidance algorithm in step S303 is specifically:

[0088] S401, determining the association orientation of the acquisition coordinates according to the hit rate of each sub-area of ​​the key area of ​​the actual power equipment image corresponding to the acquisition coordinates.

[0089] S402: Determine the associated acquisition coordinates according to the associated orientation of the acquisition coordinates, and determine the actual power equipment image corresponding to the associated acquisition coordinates as the associated power equipment image.

[0090] Specifically, the hit rates of the sub-areas are recorded in the order of their locations in the power equipment. The order is usually determined according to the structural characteristics of the power equipment, such as from left to right, from top to bottom, etc. The recorded hit rate data is analyzed to find the direction range of the sub-areas with lower hit rates relative to the sub-areas with higher hit rates, and the direction range (set to the maximum angle of several rays formed by the sub-area with the highest hit rate pointing to each sub-area with low hit rates) is determined as the associated guide; with the current acquisition coordinate as the center, a fan-shaped area is extended to the associated guide (the angle of the fan-shaped area is set to the aforementioned maximum angle), and the closest acquisition coordinate is selected in the fan-shaped area as the associated acquisition coordinate of the current acquisition coordinate. The high and low hit rate is determined according to the preset hit rate threshold. For example, the hit rate threshold is set to 70%, and a hit rate greater than 70% is set as a high hit rate, otherwise, it is set as a low hit rate.

[0091] As an example, assume that the key area of ​​a certain power equipment is divided into five sub-areas: A, B, C, D and E, which are arranged in a roughly left-to-right order. In a certain inspection, in the actual power equipment image corresponding to the acquisition coordinates of the power equipment, the hit rates of sub-areas A and B are higher, while the hit rates of sub-areas C, D and E are lower. The hit rate of each sub-area is calculated and recorded: A (100%), B (90%), C (90%), D (50%), E (20%), and an association guide is formed based on the hit rate: since the hit rates of sub-areas C, D and E are lower, the association guide points to the right; with the current acquisition coordinate as the center, a fan-shaped area is extended to the right (the maximum angle formed by the three rays formed by sub-areas A pointing to sub-areas C, D and E respectively) as the associated acquisition coordinate range, and the nearest one in the fan-shaped area is selected as the associated acquisition coordinate. The actual power equipment image corresponding to the associated acquisition coordinate contains more image information of C, D and E.

[0092] The technical solutions in the above embodiments of the present application have at least the following technical effects or advantages:

[0093] 1. By introducing a preset power scene model, the positional relationship of all power equipment in the power scene is displayed, and a two-dimensional coordinate system is set to accurately represent the position of power equipment and collection points, avoiding repeated collection and omissions, and improving the efficiency of robot inspections; by accurately planning collection points and parameters, unnecessary shooting and data processing are reduced, and monitoring costs are reduced; the hidden danger results of all power equipment are displayed in the power scene model, so that the hidden danger distribution and inspection status are clear at a glance, which is convenient for subsequent processing and decision-making.

[0094] 2. The method of acquiring the collection point sequence is more systematic. The moving path is preset according to the distribution of power equipment and inspection requirements. The key areas of each power equipment are mapped to the collection points on the moving path, and the collection point sequence is arranged in sequence. Each power equipment corresponds to a unique collection coordinate at the collection point, including the location coordinate and the shooting angle value, which ensures the accuracy and pertinence of the image acquisition of each power equipment, differentiates the key characteristics between different power equipment, and improves the accuracy of hidden danger identification.

[0095] 3. For the actual power equipment image collected by the collection coordinates of each power equipment, combined with the pre-set key areas, the importance of different sub-areas in the key areas is differentiated, so as to calculate the effective contribution value of the collection coordinates, which is used to evaluate the information completeness of the actual power equipment image collected.

[0096] 4. When the effective contribution value of the collected coordinates is less than the preset threshold, in order to make up for the loss of key area information caused by the collected coordinates, the image association guidance algorithm is used to determine the associated power equipment image to supplement the missing information. This ensures that even when the information is incomplete at certain shooting angles, the information can be supplemented through associated images, thereby improving the reliability and comprehensiveness of hidden danger identification.

[0097] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for identifying power hazards based on image recognition, characterized in that: include: S101, obtaining a collection point sequence of the inspection robot in a preset power scene model, the collection point sequence including a number of collection points arranged in a preset order, each collection point corresponding to a position coordinate and at least one collection coordinate, each collection coordinate corresponding to a power equipment number, and the collection coordinate including a shooting angle of the inspection robot; The method for acquiring the collection point sequence comprises: A1. According to the distribution and inspection requirements of power equipment, a moving path of the inspection robot is pre-set to ensure that the moving path can traverse all power equipment; A2. For each power equipment, determine its key area according to its structure and operation characteristics. The key area is set as the area where all parts that can affect the operation of the power equipment are located; A3. Based on the direction in which the key area of ​​each power device faces, map it to the ground moving path to obtain a first range, and determine the center point of the moving path defined by the first range as the collection point of the power device; A4. According to the moving path of the inspection robot and all the determined collection points, the collection points are arranged in sequence to generate a collection point sequence; The method for determining the first range in A3 includes: Draw a perpendicular line from the center point of the key area to the moving path, use the perpendicular line as a reference line, extract each point on the key area contour and emit a ray parallel to the reference line to the ground, and determine the area surrounded by the intersection points of all rays and the ground as the first range; S102, controlling the inspection robot to move to each collection point in turn, and obtaining an actual power equipment image based on the collection coordinates corresponding to the collection point; S103, based on the actual power equipment image corresponding to each acquisition coordinate, determine the target power equipment image in the power equipment sample image library according to its shooting angle, convert the actual power equipment image and the target power equipment image into a pixel matrix and input it into a pre-trained power hidden danger identification model, and output the hidden danger result of the power equipment corresponding to the acquisition coordinate; S104, displaying the hidden danger result of each actual power equipment image at a target position in the power scene model, where the target position is determined as: the power equipment corresponding to the acquisition coordinates of the actual power equipment image.

2. The method for identifying power hidden dangers based on image recognition according to claim 1, characterized in that: The electric power scenario model is provided with a two-dimensional coordinate system for representing the position coordinates of the electric power equipment and the collection point. The lower left corner of the electric power scenario model is set as the origin of the two-dimensional coordinate system. The x-axis represents the lateral distance of the electric power scenario model on the horizontal plane. Starting from the origin, the right direction is the positive direction. The y-axis represents the longitudinal distance of the electric power scenario model on the horizontal plane. Starting from the origin, the upward direction is the positive direction. The acquisition method of the acquisition coordinates includes: Based on the two-dimensional coordinate system of the power scene model, the collection coordinates of each power equipment are generated, expressed as (a, b, c), where a is the coordinate value of the collection point on the x-axis, b is the coordinate value of the collection point on the y-axis, and c is the preset shooting angle value when the inspection robot is at the collection point. The preset shooting angle value is determined as follows: the inspection robot is simulated and placed at the collection point corresponding to the power equipment, and the camera is facing the center point of the key area of ​​the power equipment.

3. The method for identifying power hidden dangers based on image recognition according to claim 1, characterized in that: The hidden danger results include normal equipment and equipment failure. The pre-trained power hidden danger identification model is specifically constructed as follows: A large number of training sets are obtained, wherein the training sets include actual power equipment images for training and sample power equipment images for training as input, and hidden danger result labels as output; The training set is used to train and optimize the preset network structure model to obtain a power hazard identification model.

4. The method for identifying power hazards based on image recognition according to claim 2, characterized in that: The power equipment image library is composed of three-dimensional image models of various power equipment. The actual power equipment image corresponding to each acquisition coordinate is determined in the power equipment image library according to its shooting angle, specifically including: S201, according to the electric power equipment number corresponding to the collected coordinates, determining the three-dimensional image model corresponding to the electric power equipment number in the electric power equipment image library; S202, obtaining two-dimensional images of the three-dimensional image model at various angles, and performing similarity calculations with actual power equipment images respectively; S203: Determine the two-dimensional image with the highest similarity to the actual power equipment image as the target power equipment image.

5. The method for identifying power hidden dangers based on image recognition according to claim 1, characterized in that: After S103, the method further includes: S301, obtaining a key area of ​​each power equipment in advance, and dividing the key area into a plurality of sub-areas according to the component type of the power equipment, each sub-area corresponding to a component type; S302, comparing the actual power equipment image with the key area, and calculating the effective contribution value of the acquisition coordinates according to the following formula: Among them, p is the effective contribution value of the acquisition coordinates, n is the total number of sub-areas in the key area of ​​the corresponding power equipment, is the criticality value of the operation of the power equipment in the i-th sub-area, is the pixel area size of the i-th sub-region, is the pixel area size of the ith sub-region in the actual power equipment image, is the hit rate of the i-th sub-region; S303, if the effective contribution value of the acquisition coordinate is less than a preset threshold, using a preset image association guidance algorithm to determine an associated power equipment image of the acquisition coordinate; S304: Use the associated power equipment image as the new actual power equipment image of the acquisition coordinates, and execute step S103 again.

6. The method for identifying power hidden dangers based on image recognition according to claim 5, characterized in that: The image association guidance algorithm in S303 is specifically: S401, determining the associated orientation of the acquisition coordinates according to the hit rate of each sub-region of the key region of the actual power equipment image corresponding to the acquisition coordinates; S402: Determine the associated acquisition coordinates according to the associated orientation of the acquisition coordinates, and determine the actual power equipment image corresponding to the associated acquisition coordinates as the associated power equipment image.

7. The method for identifying power hazards based on image recognition according to claim 6, characterized in that: The association guidance of the acquisition coordinates is specifically: The maximum angle direction formed by a plurality of rays formed by the sub-region with the highest hit rate pointing to each sub-region with a low hit rate is determined as the associated guide; The hit rate is determined based on a preset hit rate threshold, which is set to 70%.

8. The method for identifying power hidden dangers based on image recognition according to claim 7, characterized in that: In S402, determining the associated acquisition coordinates according to the associated orientation of the acquisition coordinates specifically includes: With the current acquisition coordinate as the center, a sector-shaped area is extended toward the associated guide, and the nearest acquisition coordinate is selected in the sector-shaped area as the associated acquisition coordinate of the current acquisition coordinate.

Images