A Grid High-Voltage Harness Inspection Method Based on Information Ablation

Abstract

The present invention relates to a method for inspecting high-voltage wire harnesses of a power grid based on information ablation. During the inspection of high-voltage wire harnesses, the high-voltage wire harnesses are identified by calibration, identification, and path planning, and a suitable sliding selection window is determined to optimize the consistency and recognizability of the cropped graphics. Then, by comparing the differences between each cropped graphic, the cropped graphics can be quickly stored. At the same time, the images are processed through two ablation strategies to retain their simplest information content, so that the entire video stream is segmented into a combination of an image and multiple groups of information. During the whole process, no recognition of situations such as line damage is required, which improves the reliability of the inspection from the data side. After the feedback data, only the abnormal data needs to be restored to quickly analyze the reasons while retaining relatively high image clarity.

Description

Technical Field

[0001] The present invention relates to power grid inspection technology, and more specifically, to a power grid high-voltage wire harness inspection method based on information ablation. Background Art

[0002] Currently, intelligent power grid inspection has become the mainstream inspection method. Since some power grid facilities are located in remote areas with few people, if manual inspection is carried out, the cost and price required are very high. Moreover, generally, power grid lines are erected relatively high, so manual climbing is also required, and aerial work cannot comprehensively and very carefully inspect the lines, especially the middle section of the lines cannot be inspected. Therefore, the current mainstream method is to use drones for auxiliary inspection to complete the identification of line damage. The method is to take images by drones, then transmit the taken images back, and then perform anomaly identification through the background. However, this method has a problem that drones need to store a large amount of image data and have high requirements for storage space. Secondly, if the images taken by drones are incorrect or unclear, re-inspection is required for re-taking, and the cost is relatively high. The background applicable to the original method is that since the battery capacity of drones is also limited, generally, after shooting a small area, they directly return. The influence of battery power on the endurance of drone shooting is reflected prior to the problem of insufficient endurance caused by data problems. Therefore, this problem has not been faced up to and solved. A self-release and recovery system, control method and method of an inspection drone with the publication number CN113156997A and a mobile drone and nest control system for distribution network circuit inspection with the publication number CN115686063A both disclose that the combination of drones and inspection vehicles provides endurance for drones, while data transmission and image analysis require relatively large computing power, which obviously cannot be carried out on inspection vehicles. Therefore, a power grid high-voltage wire harness inspection method based on information ablation that can greatly reduce the inspection data volume of drones is needed. Summary of the Invention

[0003] In view of this, the purpose of the present invention is to provide a power grid high-voltage wire harness inspection method based on information ablation.

[0004] To solve the above technical problems, the technical solution of the present invention is: a power grid high-voltage wire harness inspection method based on information ablation,

[0005] Step A1: Identify the high-voltage wire harness and complete position and focal length calibration at the same time;

[0006] Step A2: Determine the corresponding sliding selection window according to the characteristics of the high-voltage wire harness;

[0007] Step A3: Plan the flight path according to the sliding selection window and control the drone to fly along the flight path;

[0008] Step A4: Crop the segmented cropped graphics from the sliding selection window;

[0009] Step A5: Define the first segmented cropped graphic as the core cropped graphic;

[0010] Step A6: Process each segmented cropped graphic obtained through a preset first ablation strategy to generate first ablation information, where the first ablation information reflects the deviation of the segmented cropped graphic from the core segmented cropped graphic;

[0011] Step A7: Compare all the first ablation information to re - determine the new core cropped graphic, and re - determine the corresponding first ablation information for each through the first ablation strategy;

[0012] Step A8: Compare the first ablation information through a preset second ablation strategy to generate second ablation information and corresponding rule marks, where the rule marks reflect the regular relationship between the first ablation information;

[0013] Step A9: Delete all segmented cropped graphics other than the core segmented cropped graphic from the memory of the unmanned aerial vehicle.

[0014] Further, in step A1, it includes

[0015] Step A1 - 1: Pre - configure a relative deflection angle and a reference visual range, and determine the positions of the tower pole extension line and the high - voltage wire harness extension line in the virtual coordinate system during the ascent of the unmanned aerial vehicle to construct a reference coordinate system;

[0016] Step A1 - 2: Determine the positioning path of the unmanned aerial vehicle according to the position of the relative deflection angle in the reference coordinate system;

[0017] Step A1 - 3: Control the unmanned aerial vehicle to fly along the positioning path, and measure the distance from the initial positioning position of the high - voltage wire harness in real - time through an infrared ranging sensor;

[0018] Step A1 - 4: When the measured distance of the infrared sensor meets the reference visual range, it is regarded that the position calibration is completed;

[0019] Step A1 - 5: Calculate the feature kurtosis value of the image through a preset feature abundance recognition algorithm, and adjust the focal length of the unmanned aerial vehicle so that the feature abundance value of the image is greater than the preset reference abundance value, which is regarded as the completion of the focal length calibration.

[0020] Further, in step A2, it includes

[0021] Step A2 - 1: Pre - store a single - strand feature table, where the single - strand feature table stores several single - strand features, and each single - strand feature corresponds to a framed shape. Compare the closest single - strand feature in the image taken from the high - voltage wire harness to determine the corresponding framed shape;

[0022] Step A2-2: Divide the captured image of the high-voltage wire harness into several single-strand feature map areas according to the determined single-strand features;

[0023] Step A2-3: Calculate the feature abundance value of each single-strand feature map area through a preset feature abundance recognition algorithm, and determine the corresponding box selection length according to the feature abundance value;

[0024] Step A2-4: Generate the sliding selection window according to the box selection length and the box selection shape.

[0025] Further, the step A3 includes

[0026] Step A3-1: Obtain the extension direction of the high-voltage wire harness from the image;

[0027] Step A3-2: Generate a corresponding positioning line instruction according to the extension direction. When the drone receives the positioning line instruction, it emits a positioning ray at the corresponding position of the high-voltage wire harness;

[0028] Step A3-3: Generate the flight path of the drone according to the extension direction, and ensure that the positioning ray is always located at the preset positioning position in the captured image during the flight process.

[0029] Further, the step A4 includes

[0030] Step A4-1: Calculate the difference information between adjacent single-strand feature map areas to generate a corresponding difference abundance value;

[0031] Step A4-2: Determine the intercepted overlap length according to the difference abundance value, and retrieve the corresponding interception strategy according to the intercepted overlap length;

[0032] Step A4-3: Intercept the segmented cropped graphics from the sliding selection box through the interception strategy.

[0033] Further, the interception strategy includes generating an interception line instruction. When the drone receives the interception line instruction, it emits an interception ray at the corresponding position of the high-voltage wire harness; when the interception ray in the captured image of the drone reaches a predetermined trigger position during the flight process, the corresponding segmented cropped graphics are intercepted.

[0034] Further, the positions of adjacent segmented cropped graphics corresponding to the high-voltage wire harness have overlapping map areas, and the shapes and sizes of each overlapping map area are the same.

[0035] Further, the step A5 also includes

[0036] Step A5-1: Calculate the difference information of the overlapping map areas belonging to different segmented cropped graphics;

[0037] Step A5-2: A difference correction index table is pre-constructed. The difference correction index table stores a number of correction parameters. Each correction parameter is indexed by a difference feature. By identifying the difference feature in the difference information, the corresponding correction parameter can be obtained.

[0038] Step A5-3: Correct the corresponding block cropped graph with the correction parameter to reduce the environmental deviation between the block cropped graph and the block cropped graph at the previous moment.

[0039] Further, the first ablation strategy is configured to calculate the difference information of adjacent block cropped graphs, generate absolute difference information with the core block cropped graph as the standard according to the difference information, and compress the absolute difference information through a preset information compression algorithm to generate the first ablation information.

[0040] Further, the second ablation strategy pre-constructs a regular marking table. The regular marking table pre-stores a number of regular compression sub-algorithms. Each regular compression sub-algorithm is indexed by a regular trigger condition. The second ablation strategy identifies the regular features in the first ablation information. When the corresponding regular features meet the regular trigger condition, the corresponding regular compression sub-strategy is obtained, and the corresponding first ablation information is compressed through the regular compression sub-algorithm to generate the second ablation information, and a regular mark of the regular compression sub-strategy is generated.

[0041] The technical effects of the present invention are mainly reflected in the following aspects: By setting like this, during the high-voltage wire harness inspection process, the high-voltage wire harness can be identified by calibration, identification, and path planning, and a suitable sliding selection window can be determined, so that the consistency and recognizability of the cropped graph reach the optimal. Then, by comparing the differences of each cropped graph, the cropped graph can be quickly stored. At the same time, the image is processed through two ablation strategies to retain its simplest information content, so that the entire video stream is cut into a combination of an image and multiple groups of information. And during the whole process, there is no need to identify situations such as line damage. From the data side, the reliability of the inspection is improved. After the feedback data, only the abnormal data needs to be restored, and the reason can be quickly analyzed, while maintaining a high image clarity. Description of the Drawings

[0042] Figure 1 : Schematic flow chart of a method for inspecting high-voltage wire harnesses of a power grid based on information ablation of the present invention. Detailed Embodiments

[0043] The following further details the specific embodiments of the present invention with reference to the drawings, so that the technical solutions of the present invention are easier to understand and master.

[0044] A method for inspecting high-voltage wire harnesses of a power grid based on information ablation

[0045] Step A1: Identify the high-voltage wire harness and complete position and focal length calibration simultaneously; First, the purpose of this step is to ensure that the image corresponding to the identification of the high-voltage wire harness has sufficient characteristic textures. Otherwise, problems such as unidentifiable due to insufficient feature abundance and graphic processing will occur in subsequent comparisons. On the other hand, a better viewing point is required, and the viewing point is unified. In this way, high-voltage wire harnesses of different thicknesses are in

[0046] Step A1 includes

[0047] Step A1-1: A relative deflection angle and a reference viewing distance are pre-configured. During the ascent of the UAV, the positions of the tower extension line and the high-voltage wire harness extension line are determined in the virtual coordinate system to construct a reference coordinate system; First, the UAV constructs a virtual coordinate system with the initial position as the origin. During the flight of the UAV, the flight route is known, so the position of the UAV in the coordinate system is known. Then, when the UAV ascends and approaches the tower, a line perpendicular to the ground can be obtained. Until it flies over the tower, the extension line of the high-voltage wire harness can be obtained. Theoretically, these two lines are perpendicular. However, due to certain bending of the high-voltage wire harness, the tower extension line is translated towards the endpoint of the high-voltage wire harness to form a reference plane. Taking the intersection point of the translated high-voltage wire harness extension line and the tower extension line as the origin, a reference coordinate system can be constructed.

[0048] Step A1-2: Determine the positioning path of the UAV according to the position of the relative deflection angle in the reference coordinate system; The relative deflection angle is artificially configured. First, to ensure that the shadow of the UAV does not block the high-voltage wire harness, and second, to ensure that the UAV can take relatively clear pictures in the direction of solar radiation. Therefore, generally, the relative deflection angle is the mirror image of the solar radiation direction on the reference plane marked on the reference. Then, it is fine-tuned according to the solar radiation intensity and angle to form the corresponding relative deflection angle. That is, starting from the origin, the initial position of the UAV is a ray that forms a relative deflection angle with the reference plane. The pre-configuration of the relative deflection angle is input according to the environmental recognition big data sample. By calculating the relationship between the environmental information and the angle of obtaining the optimal captured image through the deep learning algorithm, the actual environmental information is collected to obtain the optimal relative deflection angle at that time.

[0049] Step A1-3: Control the UAV to fly along the positioning path and measure the distance from the initial positioning position of the high-voltage wire harness in real time through an infrared ranging sensor;

[0050] Step A1-4: When the measured distance of the infrared sensor meets the reference viewing distance, it is regarded as the completion of position calibration; In this way, the shooting distance of the UAV can be controlled to be unified, and the thickness, length and other characteristics of the high-voltage wire harness can be judged from the image, so as to accurately identify foreign objects, damage, icing and other situations.

[0051] Step A1-5: Calculate the characteristic kurtosis value of the image through a preset characteristic abundance recognition algorithm, and adjust the focal length of the drone so that the characteristic abundance value of the image is greater than the preset reference abundance value, which is regarded as the completion of focal length calibration. Currently, the characteristic abundance of an image is calculated from three dimensions: the complexity and distribution of color values, the complexity and distribution of shapes, and the complexity and distribution of textures. Currently, there are already relatively mature various characteristic abundance recognition algorithms, that is, to calculate the richness of an image or a region, so it will not be elaborated here. Since the clarity will change due to the focal length adjustment, the characteristic abundance value will increase. Therefore, by determining the reference abundance value, it can be ensured that when the drone is flying, the focal length will not change and a clear image of the high-voltage wire harness can be captured. The reference abundance value of the characteristic abundance recognition algorithm is obtained by training the image data of the samples. These image data are pre-labeled with the standard of the abundance value level, and then the deviation between the training result and the actual result is used to obtain the reference abundance value of the image.

[0052] Step A2: Determine the corresponding sliding selection window according to the characteristics of the high-voltage wire harness;

[0053] The step A2 includes

[0054] Step A2-1: A single-strand feature table is pre-stored. The single-strand feature table stores several single-strand features, and each single-strand feature corresponds to a box selection shape. Compare the closest single-strand feature in the image of the high-voltage wire harness to determine the corresponding box selection shape; First, determine the position area of the single-strand wire harness from the image taken at the initial position, and then intercept it. According to this area, single-strand features can be obtained, such as the curvature of the texture curve, color distribution, single-strand edge shape, texture quantity, the shape presented by the single-strand in the image, etc. These features are identified by comparing to determine the closest feature. The determination of the closest feature is carried out through a feature recognition neural network, and the recognition result can be stored as historical data. After confirmation in the background, the original features can be supplemented. There is a pre-corresponding box selection shape to make the box selection shape match the features of the wire harness best, that is, the box selection shape can ensure that the complete area of the wire harness is box-selected and can clearly reflect the wire harness texture, or the box selection area can generally clearly reflect changes such as damage and icing, so that the observer can identify.

[0055] Step A2-2: Divide the image of the high-voltage wire harness into several single-strand feature map areas according to the determined single-strand features; After determining the shape, it is necessary to determine the corresponding box selection length. If the feature richness of the single-strand feature is high, the box selection length can be appropriately reduced, and vice versa increased.

[0056] Step A2-3: Calculate the feature abundance value of each single-strand feature map area through a preset feature abundance recognition algorithm, and determine the corresponding box selection length according to the feature abundance value; thus, the box selection length is determined by the feature abundance value. Without changing the shape, the length of the sliding selection window is changed in the extension direction of the high-voltage wire harness, and the relationship between the feature abundance value and the selection box length is preset to ensure that the selection box can just cover an integer number.

[0057] Step A2-4: Generate the sliding selection window according to the box selection length and the box selection shape.

[0058] Step A3: Plan the flight path according to the sliding selection window and control the drone to fly along the flight path; the purpose of this step is to determine the flight path and ensure the consistency of image shooting.

[0059] The step A3 includes

[0060] Step A3-1: Obtain the extension direction of the high-voltage wire harness from the image.

[0061] Step A3-2: Generate a corresponding positioning line instruction according to the extension direction. When the drone receives the positioning line instruction, it emits a positioning ray at the position corresponding to the high-voltage wire harness; the positioning ray is based on a certain determined position to ensure that the ray is always marked at that position. For the drone, it plays an auxiliary positioning effect. Auxiliary flight calibration. There is no need to perform repeated image recognition.

[0062] Step A3-3: Generate the flight path of the drone according to the extension direction, and ensure that the positioning ray is always located at the preset positioning position of the captured image during the flight process.

[0063] Step A4: Intercept a block cropped graphic from the sliding selection window.

[0064] The step A4 includes

[0065] Step A4-1: Calculate the difference information between adjacent single-strand feature map areas to generate a corresponding difference abundance value; first, find the graphics with differences to form a difference graphic as the difference information, and calculate the feature abundance value of the difference graphic as the difference abundance value.

[0066] Step A4-2: Determine the intercepted overlap length according to the difference abundance value, and retrieve the corresponding interception strategy according to the intercepted overlap length; if the difference abundance value is larger, it means that the difference between single strands is larger, then a longer overlap length is required. That is, for example, the position of the first selection box is from 0 to 3 units in the length direction. Then, if the overlap length is large, the second selection box is at 1 to 4 units, and so on. If the overlap length is small, the second selection box is at 2 to 5 units, and so on.

[0067] Step A4-3: Intercept the segmented cropping graphics from the sliding selection box through an interception strategy.

[0068] The interception strategy includes generating an interception line instruction. When the drone receives the interception line instruction, it emits an interception ray at the corresponding position of the high-voltage wire harness. When the interception ray of the image captured by the drone during flight reaches a predetermined trigger position, the corresponding segmented cropping graphics are intercepted. Since the position of the next interception is known, the interception ray is used to assist in marking. In this way, when the drone is flying, the interception efficiency can be improved and positioning can be assisted.

[0069] Adjacent segmented cropping graphics have overlapping regions corresponding to the positions of the high-voltage wire harness, and the shapes and sizes of each overlapping region are the same. The purpose of ensuring that the sizes of the overlapping regions are the same is to calculate environmental variables based on the differences in the overlapping regions. Since the images theoretically should be the same when moving from position A to position B, but for example, there are changes in illumination or fog that cause changes in the captured images, then these can be discovered by identifying the overlapping images. Specifically:

[0070] Step A5: Define the first segmented cropping graphic as the core cropping graphic;

[0071] The step A5 also includes

[0072] Step A5-1: Calculate the difference information of the overlapping regions belonging to different segmented cropping graphics; that is, a difference image can be obtained.

[0073] There is a pre-constructed difference correction index table. The difference correction index table stores several correction parameters, and each correction parameter is indexed by a difference feature. By identifying the difference features in the difference information, the corresponding correction parameter can be obtained. For example, in the difference image, there are different difference situations such as brightness differences and brightness distribution differences. Then the brightness of the subsequent images can be corrected by the brightness correction parameter. The construction of the difference features is carried out through big data training. Taking samples under known different illumination intensities and illumination angles as an example, actual images are captured to form training samples, and the relationships between the training samples are learned through machine learning to obtain the laws of the difference features, thereby generating the corresponding difference features.

[0074] Step A5-3: Correct the corresponding segmented cropping graphics by the correction parameters to reduce the environmental deviation between the segmented cropping graphic and the previous moment's segmented cropping graphic. For example, for the brightness distribution difference, the brightness image is corrected according to this difference law, so that the environmental deviation is smaller, that is, the degree of image consistency is increased, and thus the data volume can be reduced as much as possible when the images are ablated.

[0075] Step A6: Process each piecewise cropped graph obtained by the preset first ablation strategy to generate first ablation information, where the first ablation information reflects the deviation between the piecewise cropped graph and the core piecewise cropped graph;

[0076] Step A7: Compare all the first ablation information to re-determine a new core cropped graph, and re-determine the corresponding first ablation information for each one through the first ablation strategy; the first ablation strategy is configured to calculate the difference information between adjacent piecewise cropped graphs, generate absolute difference information based on the core piecewise cropped graph according to the difference information, and compress the absolute difference information through a preset information compression algorithm to generate the first ablation information. That is, the difference information is saved. For example, originally there were graphs A, B, and C. Now, graph A is retained, but graphs B and C are recorded as the difference information between graph A + graph B and graph A + graph C. The corresponding first ablation information is generated through the difference information.

[0077] Step A8: Compare the first ablation information through a preset second ablation strategy to generate second ablation information and corresponding rule marks, where the rule marks reflect the regular relationship between the first ablation information; the second ablation strategy pre-constructs a rule mark table, and the rule mark table pre-stores several rule compression sub-algorithms. Each rule compression sub-algorithm is indexed by a rule trigger condition. The second ablation strategy identifies the regular features in the first ablation information. When the corresponding regular features meet the rule trigger condition, the corresponding rule compression sub-strategy is obtained, and the corresponding first ablation information is compressed through the rule compression sub-algorithm to generate the second ablation information, and the rule mark of the rule compression sub-strategy is generated. The second ablation information reflects the change rule of the first ablation information. For example, the wire harness gradually becomes brighter. There is no need to record each difference information, only the change rule needs to be recorded. First, identify through the preset regular features to find the corresponding rule mark, and then find the data compression algorithm through the rule mark to compress the data of the first ablation information to obtain the second ablation information.

[0078] Step A9: Delete other piecewise cropped graphs except the core piecewise cropped graph from the memory of the drone. That is, after the drone inspects a complete wire harness, the core piecewise cropped graph, part of the first ablation information, and the second ablation information are retained. That is, the graph can be recorded as graph A, the first ablation information of graph A + graph B, the first ablation information of graph A + graph C, and the change trend of graph ABC.

[0079] Of course, the above are only typical examples of the present invention. In addition, the present invention can also have many other specific implementation manners. Any technical solutions formed by equivalent replacement or equivalent transformation fall within the scope of protection required by the present invention.

Claims

1. A high-voltage wire harness inspection method for power grids based on information ablation, characterized in that: Step A1: Identify the high-voltage wire harness and complete position and focal length calibration simultaneously; Step A2: Determine the corresponding sliding selection window according to the characteristics of the high-voltage wire harness; Step A3: Plan the flight path according to the sliding selection window and control the drone to fly along the flight path; Step A4: Intercept the segmented cropped graphics from the sliding selection window; Step A5: Define the first segmented cropped graphic as the core cropped graphic; Step A6: Process each segmented cropped graphic obtained through a preset first ablation strategy to generate first ablation information, and the first ablation information reflects the deviation between the segmented cropped graphic and the core segmented cropped graphic; Step A7: Compare all the first ablation information to re-determine the new core cropped graphic, and re-determine the corresponding first ablation information for each through the first ablation strategy; Step A8: Compare the first ablation information through a preset second ablation strategy to generate second ablation information and the corresponding rule mark, and the rule mark reflects the rule relationship between the first ablation information; Step A9: Delete other segmented cropped graphics from the memory of the drone except the core segmented cropped graphic; The first ablation strategy is configured to calculate the difference information between adjacent segmented cropped graphics, generate absolute difference information based on the core segmented cropped graphic according to the difference information, and compress the absolute difference information through a preset information compression algorithm to generate the first ablation information; The second ablation strategy pre-constructs a rule mark table, and the rule mark table pre-stores a number of rule compression sub-algorithms. Each rule compression sub-algorithm is indexed by a rule trigger condition. The second ablation strategy identifies the rule features in the first ablation information. When the corresponding rule features meet the rule trigger condition, obtain the corresponding rule compression sub-strategy, compress the corresponding first ablation information through the rule compression sub-algorithm to generate the second ablation information, and generate the rule mark of the rule compression sub-strategy.

2. A high-voltage wire harness inspection method for power grids based on information ablation according to claim 1, characterized in that: In step A1, it includes Step A1-1: Pre-configure a relative deflection angle and a reference viewing distance, and determine the positions of the tower extension line and the high-voltage wire harness extension line in the virtual coordinate system during the ascending process of the drone to construct a reference coordinate system; Step A1-2: Determine the positioning flight path of the drone according to the position of the relative deflection angle in the reference coordinate system; Step A1-3: Control the drone to fly along the positioning flight path, and measure the distance from the initial positioning position of the high-voltage wire harness in real time through an infrared ranging sensor; Step A1-4: When the measured distance of the infrared sensor meets the reference viewing distance, it is regarded as the completion of position calibration; Step A1-5: Calculate the feature kurtosis value of the image through a preset feature abundance recognition algorithm, and adjust the focal length of the drone so that the feature abundance value of the image is greater than the preset reference abundance value, which is regarded as the completion of focal length calibration.

3. A high-voltage wire harness inspection method for power grids based on information ablation according to claim 1, characterized in that: In step A2, it includes Step A2-1: A single-strand feature table is pre-stored. The single-strand feature table stores several single-strand features, and each single-strand feature corresponds to a selected shape. The closest single-strand feature is compared in the image taken of the high-voltage wire harness to determine the corresponding selected shape; Step A2-2: Divide the image taken of the high-voltage wire harness into several single-strand feature map areas according to the determined single-strand features; Step A2-3: Calculate the feature abundance value of each single-strand feature map area through a preset feature abundance recognition algorithm, and determine the corresponding selected length according to the feature abundance value; Step A2-4: Generate the sliding selection window according to the selected length and the selected shape; 4. A method for inspecting high-voltage wire harnesses of a power grid based on information ablation according to claim 1, characterized in that: In the said step A3, it includes Step A3-1: Obtain the extension direction of the high-voltage wire harness from the image; Step A3-2: Generate a corresponding positioning line instruction according to the extension direction. When the unmanned aerial vehicle receives the positioning line instruction, it emits a positioning ray at the corresponding position of the high-voltage wire harness; Step A3-3: Generate the flight path of the unmanned aerial vehicle according to the extension direction, and ensure that the positioning ray is always located at a preset positioning position in the taken image during the flight process.

5. A method for inspecting high-voltage wire harnesses of a power grid based on information ablation according to claim 3, characterized in that: The said step A4 includes Step A4-1: Calculate the difference information between adjacent single-strand feature map areas to generate a corresponding difference abundance value; Step A4-2: Determine the intercepted overlapping length according to the difference abundance value, and retrieve the corresponding interception strategy according to the intercepted overlapping length; Step A4-3: Intercept the block cropping graphic from the sliding selection frame through the interception strategy.

6. A method for inspecting high-voltage wire harnesses of a power grid based on information ablation according to claim 5, characterized in that: The interception strategy includes generating an interception line instruction. When the unmanned aerial vehicle receives the interception line instruction, it emits an interception ray at the corresponding position of the high-voltage wire harness; when the interception ray in the taken image of the unmanned aerial vehicle reaches a predetermined trigger position during the flight process, the corresponding block cropping graphic is intercepted.

7. A method for inspecting high-voltage wire harnesses of a power grid based on information ablation according to claim 6, characterized in that: The adjacent block cropping graphics have an overlapping map area corresponding to the position of the high-voltage wire harness, and the shape and size of each overlapping map area are the same.

8. A method for inspecting high-voltage wire harnesses of a power grid based on information ablation according to claim 1, characterized in that: In the said step A5, it also includes Step A5-1: Calculate the difference information of the overlapping map areas belonging to different block cropping graphics; Step A5-2: A difference correction index table is pre-constructed. The difference correction index table stores several correction parameters, and each correction parameter is indexed by a difference feature. The corresponding correction parameter is obtained by identifying the difference feature in the difference information; Step A5-3: Correct the corresponding block cropping graphic through the correction parameter to reduce the environmental deviation between the block cropping graphic and the block cropping graphic at the previous moment.

Images