Unmanned aerial vehicle electric power inspection method and system based on multi-source navigation and Yo-v8

By integrating A* algorithm, reinforcement learning and penalty function optimization, combined with Yolo-v8 model and multi-source navigation, the problem of flexibility and insufficient response speed of path planning and obstacle avoidance during power inspection of drones is solved, and the problem of insufficient response speed is achieved. Efficient identification and analysis of power line defects improves patrol efficiency and safety.

CN118938982APending Publication Date: 2024-11-12GUANGZHOU UNIVERSITY
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202410980605.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-22
Publication Date
2024-11-12

AI Technical Summary

Technical Problem

The existing drone power inspection technology has problems of flexibility and insufficient response speed in path planning and obstacle avoidance. It is not feasible to use reinforcement learning or penalty function optimization alone in practical applications.

Method used

Combining the A* algorithm, reinforcement learning and penalty function optimization, combining the Yolo-v8 model and multi-source navigation method, through pre-training defect recognition and reinforcement learning experience pool, multi-source navigation and power line defect recognition of drones are realized, and using GPS global positioning and depth cameras for obstacle detection and path optimization.

Benefits of technology

It improves the flexibility and response speed of drones in power inspection, realizes the rapid identification and analysis of power line defects, reduces the influence of human factors, and ensures the safe operation and stable power supply of power lines.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118938982A_ABST
    Figure CN118938982A_ABST
Patent Text Reader

Abstract

The embodiment of the invention provides an unmanned aerial vehicle electric power inspection method and system based on multi-source navigation and Yo-v8, and the method comprises the steps: training a defect recognition experience pool and a reinforcement learning experience pool in advance, and formulating a cruise target point; the unmanned aerial vehicle uses a GPS global positioning fusion multi-source navigation method to avoid obstacles according to the cruising target point and flies to the target point; and photographing each target point, and carrying out electric power line defect identification according to the obtained pictures. According to the method, the penalty function thought is combined, reinforcement learning is integrated into a local planning penalty function, so that the unmanned aerial vehicle can quickly make flight actions, and the situation that too much time is wasted due to waiting for penalty factor updating is avoided; the method has the capability of autonomously identifying power inspection defects, and can efficiently and autonomously identify the defects of power equipment.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This document relates to the field of power inspection technology, and in particular to a UAV power inspection method and system based on multi-source navigation and Yolo-v8. Background Art

[0002] In recent years, drones have entered the field of power inspection, but there are many challenges in power inspection, such as complex obstacles. The identification of power equipment defects still relies on manual identification after drone aerial photography. In terms of path planning, some scholars have proposed using reinforcement learning alone to train drones, but reinforcement learning is usually optimized for specific environments during training. Once the environment changes, the model may need to be retrained to adapt to the new environment, which poses a challenge to the flexibility and adaptability of drones. Some scholars have proposed using the A* algorithm alone for drone path planning, but this method makes it impossible for drones to avoid sudden obstacles in time. Other scholars have proposed the idea of ​​penalty functions to perform local path planning, but the size of the penalty factor of the penalty function is automatically updated too slowly, which will make the drone respond too slowly.

[0003] Therefore, it is not feasible to use reinforcement learning or penalty function optimization alone in power inspection. Therefore, the present invention integrates the A* algorithm, reinforcement learning and penalty function optimization together, so that the drone has the ability of multi-source navigation and is easier to adapt to the current environment. At the same time, it avoids the problem of slow timeliness caused by the slow automatic update of the penalty factor size, and has certain engineering application significance. Summary of the invention

[0004] One or more embodiments of this specification provide a UAV power inspection method based on multi-source navigation and Yolo-v8, including:

[0005] S1. Pre-train the defect recognition experience pool and reinforcement learning experience pool to set the cruise target point;

[0006] S2. The UAV uses GPS global positioning fusion multi-source navigation method to avoid obstacles and fly to the target point according to the cruise target point;

[0007] S3. Take photos of each target point and identify power line defects based on the photos obtained.

[0008] Furthermore, the specific method for training the defect recognition experience pool is as follows:

[0009] The insulators on the transmission line are collected using a camera head gimbal mounted on a drone, and image data of faulty insulators are selected and expanded.

[0010] Then, the expanded image data is annotated with the insulator defect types, including normal and abnormal defect types. The annotated data set is used as a sample library and divided into a training set and a test set.

[0011] The SE dual-channel attention mechanism is introduced into the Yolo-v8 network model, and the image data and annotation files in the training set are used to iteratively train the Yolo-v8 network model with the SE dual-channel attention mechanism, wherein the Yolo-v8 network model includes a feature extraction layer, a feature fusion layer and an output layer;

[0012] During the training process, the model weights with the highest accuracy are saved to complete the construction of the UAV defect recognition experience pool.

[0013] Furthermore, the specific method for training the reinforcement learning experience pool is:

[0014] Constructing a Markov space for the drone, the Markov space including the current drone state space, the next execution action of the drone, the current reward value of the drone, and the state space transition of the drone;

[0015] Construct a reward function for the drone, which includes two rewards, one for reaching the target point and the other for failure. When the drone reaches the target point margin, the drone receives a positive reward, and when the reading of the drone's distance sensor is below the threshold, it is punished;

[0016] Repeat the above process for training according to the cruise route obtained from the cruise target point to obtain the reinforcement learning experience pool of the drone.

[0017] Furthermore, the UAV uses GPS global positioning fusion multi-source navigation method to avoid obstacles and fly to the target point according to the cruise target point. The specific method is:

[0018] The drone is equipped with a GPS module to obtain the current coordinates of the drone and the coordinates of the target inspection point, and a second-order Bezier curve is used to generate a smooth curve from the current coordinate point to the target coordinate point. The drone moves along the curve;

[0019] During the movement, the drone is equipped with a depth camera to obtain depth images, and uses the depth images for Slam mapping to detect whether there are obstacles in front of the drone;

[0020] If an obstacle is detected, the A* algorithm is used to replan the route for the collision segment on the generated second-order Bezier curve;

[0021] When a collision occurs in an obstacle, the vertically extended ray intersects the previously planned route, and a penalty item constraint is imposed on the new route. The penalty items include collision item penalty and reinforcement learning experience pool penalty, and the optimized best path is obtained.

[0022] Further, the steps of detecting power line defects according to the obtained photos are as follows:

[0023] Get real-time images through the camera on the drone and input the images to the edge device equipped with the Yolo-v8 model;

[0024] Based on the experience gained from the defect identification experience pool, the defective insulators are classified and located through the Yolo-v8 model to achieve defect identification.

[0025] Furthermore, the specific method for marking the insulator defect type on the expanded image data is as follows:

[0026] Create two labels in the annotation interface, corresponding to normal insulators and defective insulators respectively, then use rectangular boxes to mark the insulator targets in each picture, select the label corresponding to the target, and finally generate an xml annotation file with the annotation information;

[0027] Each XML file corresponds to each annotated picture. The annotation information of the file contains 4 position attributes and 1 category attribute. The position attributes are the horizontal coordinate x_min and vertical coordinate y_min of the upper left corner of the rectangular box, and the horizontal coordinate x_max and vertical coordinate y_max of the lower right corner. The category attribute is the label selected when annotating the picture.

[0028] Furthermore, the SE dual-channel attention mechanism is introduced. The specific steps are as follows:

[0029] First, the input feature map is globally pooled to change its size to 1×1×the number of channels;

[0030] Then, the 1×1×channel feature map is adjusted through the fully connected layer and the Sigmoid activation function to generate the weight of each feature map;

[0031] These weights are multiplied element-by-element with the input features. The compression operation uses global average pooling to generate the channel statistic Z_c. The excitation operation uses a fully connected layer and a Sigmoid activation function to generate a weight value for each feature channel. The calculation formula of Z_c is as follows:

[0032]

[0033] One or more embodiments of this specification provide a UAV power inspection system based on multi-source navigation and Yolo-v8, including:

[0034] Target point determination module: used to pre-train the defect recognition experience pool and reinforcement learning experience pool and set the cruise target point;

[0035] Cruise module: used to enable the UAV to avoid obstacles and fly to the target point according to the cruise target point by using GPS global positioning fusion multi-source navigation method;

[0036] Defect recognition module: used to take photos of each target point and identify power line defects based on the photos obtained.

[0037] One or more embodiments of the present specification provide an electronic device, including:

[0038] processor; and,

[0039] A memory arranged to store computer executable instructions, which, when executed, enable the processor to implement the steps of the above-mentioned UAV power inspection method based on multi-source navigation and Yolo-v8.

[0040] One or more embodiments of the present specification provide a storage medium for storing computer-executable instructions, which, when executed, implement the steps of the above-mentioned UAV power inspection method based on multi-source navigation and Yolo-v8.

[0041] By adopting the embodiment of the present invention, the defect recognition experience pool and the reinforcement learning experience pool are pre-trained, so that the drone has a certain flight "experience", which greatly improves the efficiency of the drone's inspection of power lines. Based on this "experience", combined with the penalty function idea, reinforcement learning is integrated into the local planning penalty function, so that the drone can quickly make flight actions and avoid wasting too much time waiting for the penalty factor to update; at the same time, the drone also has the ability to autonomously identify power inspection defects, and can accurately analyze the status of the power line in a short time, identify various defects and problems that may exist, so as to take necessary repair measures in time to ensure the safe operation and stable power supply of the power line. The present invention not only improves the efficiency and accuracy of inspection, but also can greatly reduce the impact of human factors, improve the efficiency and safety of power line operation and maintenance, and has very important practical significance and application prospects for the modern power industry.

[0042] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented according to the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate one or more embodiments of this specification or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0044] Figure 1 A flowchart of a UAV power inspection method based on multi-source navigation and Yolo-v8 provided for one or more embodiments of this specification;

[0045] Figure 2 A specific flow chart of a UAV power inspection method based on multi-source navigation and Yolo-v8 provided for one or more embodiments of this specification;

[0046] Figure 3 A schematic diagram of the composition of a UAV power inspection system based on multi-source navigation and Yolo-v8 provided for one or more embodiments of this specification;

[0047] Figure 4 A schematic diagram of the structure of an electronic device provided for one or more embodiments of this specification. DETAILED DESCRIPTION

[0048] In order to enable those skilled in the art to better understand the technical solutions in one or more embodiments of this specification, the following will be combined with the drawings in one or more embodiments of this specification to clearly and completely describe the technical solutions in one or more embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on one or more embodiments of this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this document.

[0049] Method Embodiment

[0050] According to an embodiment of the present invention, a UAV power inspection method based on multi-source navigation and Yolo-v8 is provided. Figure 1 A flowchart of a UAV power inspection method based on multi-source navigation and Yolo-v8 is provided for one or more embodiments of this specification, such as Figure 1 As shown, the UAV power inspection method based on multi-source navigation and Yolo-v8 according to an embodiment of the present invention specifically includes:

[0051] S1. Pre-train the defect recognition experience pool and reinforcement learning experience pool to set the cruise target point.

[0052] First, the 1,000 photos obtained by taking photos are manually preprocessed, such as dividing the original pictures in equal proportions, mirroring and flipping, etc., to expand the data set, and then manually annotate the data, frame the defective parts and categories: normal and abnormal. In this embodiment, the size of the expanded data set is 8,910 photos, and then the defective pictures of the original data set and the expanded data set are put together as a sample library. The data set is divided into a training set and a test set, and 30% of them are selected as a test set and 70% as a training set.

[0053] Create two labels in the annotation interface, corresponding to normal insulators and defective insulators respectively, then use rectangular boxes to mark the insulator targets in each picture, and select the label corresponding to the target. Finally, generate an xml annotation file with the annotation information.

[0054] Each XML file corresponds to each annotated image. The annotation information of the file contains 4 position attributes and 1 category attribute: the position attributes are the horizontal coordinate x_min and vertical coordinate y_min of the upper left corner of the rectangular box, and the horizontal coordinate x_max and vertical coordinate y_max of the lower right corner; the category attribute is the label selected when annotating the image. The position attribute and category attribute represent the position and category of the insulator area in the image (i.e. normal insulator or defective insulator).

[0055] First, the input feature map is globally pooled to change its size to 1×1×number of channels. Then, this 1×1×number of channels feature map is adjusted through the fully connected layer and the Sigmoid activation function to generate the weight of each feature map. Next, these weights are multiplied element-by-element with the input features. As shown in the figure, the compression operation uses global average pooling to generate the channel statistic Z_c, and the excitation operation uses the fully connected layer and the Sigmoid activation function to generate a weight value for each feature channel. The calculation formula of Z_c is as follows:

[0056]

[0057] Using the image data and annotation files in the training set, the Yolo-v8 network model with SE attention mechanism is iteratively trained. The model includes feature extraction layer, feature fusion layer and output layer. During the training process, the model weights with the highest accuracy are saved.

[0058] After the above processing and training, the drone can acquire a defect recognition experience pool.

[0059] Preferably, the specific steps of strengthening the learning experience pool in S1 are as follows:

[0060] The current state of the drone can be expressed as:

[0061] Where (v x ,v y ,v z ) represents the speed of the drone in three directions, Represents the pitch, roll and yaw angles of the drone.

[0062] Using Markov space it can be expressed as:

[0063] P = {s now ,a,k,r};

[0064] Among them, a represents the next random flight action of the drone, k represents the probability of the drone choosing this action, and r represents the reward function, which represents the reward value of the current drone.

[0065] where r = r + r get , d represents the distance measurement result of the distance measurement sensor, d max Represents the maximum tolerable collision radius. If the ranging result is less than the maximum tolerable radius, it is considered that the drone has collided and a penalty is imposed. Otherwise, a reward is imposed. Select d max =0.5.

[0066] Finally, the above process is repeated so that the drone can obtain a reinforcement learning experience pool.

[0067] S2. The UAV uses GPS global positioning fusion multi-source navigation method to avoid obstacles and fly to the target point according to the cruise target point.

[0068] Figure 2 A specific flow chart of a UAV power inspection method based on multi-source navigation and Yolo-v8 provided in one or more embodiments of this specification, such as Figure 2 As shown in the figure, after the UAV training obtains the reinforcement learning experience pool, it obtains the current positioning and target positioning according to GPS. The UAV is equipped with a GPS module to obtain the current latitude and longitude P0 = (UAV longitude 、UAV latitude ), and at the same time obtain the latitude and longitude of the target inspection point through the satellite map P2 = (D longitude , D latitude ), the midpoint P1. Select P0 = {23.07, 120.03}, P2 = {23.09, 120.08}, P1 = {23.08, 120.055}.

[0069] Then, based on the three points mentioned above, a smooth curve l1 is generated using a second-order Bezier curve (without considering obstacles), where the path of the quadratic Bezier curve is given by the function B(t) of the given points P0, P1 and P2:

[0070] B(t)=(1-t) 3 P0+3t(1-t) 2 P1+3t 2 (1-t)P2+t 3 P3,t∈[0,1];

[0071] The drone is equipped with a depth camera, so it can obtain depth images and use them for Slam mapping to obtain the position of obstacles in front of the drone. The Slam algorithm is the open source algorithm ORB-SLAM2:

[0072] The drone then detects the portion of the l1 curve inside the obstacle and defines this portion as the collision zone O1. The open source A* algorithm is then used to regenerate the guide line segment c1 for the collision zone O1.

[0073] There are n control points in the collision zone {A1,A2,...,A n}, and generate rays {A1I1,A2I2,...,A n I n} intersects the guide line segment c1 at the point {I1,I2,...,I n}, then for the points {I1,I2,...,I n} Optimize the trajectory so that the guide line segment c1 is away from obstacles. For two interval control points, the velocity, acceleration, and jerk can be obtained by differentiating the time respectively:

[0074]

[0075] Consider using the penalty function idea and transform it into a nonlinear unconstrained optimization problem. The penalty function formula is as follows:

[0076] L min =η1L1+L2+η2L3;

[0077] Among them, L1 is the smoothness penalty, L2 is the reinforcement learning penalty, L3 is the collision penalty, η1=η2 is the penalty factor weight. Considering that the size of the penalty factor is difficult to select, in order to make the drone avoid obstacles first based on "experience", a reinforcement learning penalty term is added.

[0078] Selecting η1=η2=0.3 and minimizing high-order derivatives can make the entire trajectory smooth, so the smooth term is composed of the sum of the acceleration and the cube of the jerk.

[0079]

[0080] The drone obtains flight "experience" based on the reinforcement learning experience pool and selects the best flight action. The specific form of the reinforcement learning penalty term is as follows:

[0081] L2=r=r+r get ,

[0082] The collision penalty "pushes" the drone away from obstacles. We introduce a hyperbolic penalty, which takes the following form:

[0083]

[0084] where d i Represents the distance from the control point to the obstacle, and s is the pre-set safety threshold. Take s = 0.5m.

[0085] Then we get the curve ω, which is the optimal path after optimization.

[0086] S3. Take photos of each target point and identify power line defects based on the photos obtained.

[0087] Get real-time images through the camera on the drone and input the images to the edge device equipped with the Yolo-v8 model;

[0088] Based on the experience gained from the defect identification experience pool, the defective insulators are classified and located through the Yolo-v8 model to achieve defect identification.

[0089] The beneficial effects of the present invention are as follows:

[0090] By adopting the embodiment of the present invention, the defect recognition experience pool and the reinforcement learning experience pool are pre-trained, so that the drone has a certain flight "experience", which greatly improves the efficiency of the drone's inspection of power lines. Based on this "experience", combined with the penalty function idea, reinforcement learning is integrated into the local planning penalty function, so that the drone can quickly make flight actions and avoid wasting too much time waiting for the penalty factor to update; at the same time, the drone also has the ability to autonomously identify power inspection defects, and can accurately analyze the status of the power line in a short time, identify various defects and problems that may exist, so as to take necessary repair measures in time to ensure the safe operation and stable power supply of the power line. The present invention not only improves the efficiency and accuracy of inspection, but also can greatly reduce the impact of human factors, improve the efficiency and safety of power line operation and maintenance, and has very important practical significance and application prospects for the modern power industry.

[0091] System Example

[0092] According to an embodiment of the present invention, a UAV power inspection system based on multi-source navigation and Yolo-v8 is provided. Figure 3A schematic diagram of the composition of a UAV power inspection system based on multi-source navigation and Yolo-v8 provided in one or more embodiments of this specification, such as Figure 3 As shown, the UAV power inspection system based on multi-source navigation and Yolo-v8 according to an embodiment of the present invention specifically includes:

[0093] Target point determination module 30: used to pre-train the defect recognition experience pool and the reinforcement learning experience pool and formulate the cruise target point;

[0094] Cruise module 32: used to enable the UAV to avoid obstacles and fly to the target point using GPS global positioning fusion multi-source navigation method according to the cruise target point;

[0095] Defect recognition module 34: used to take photos of each target point and identify power line defects based on the obtained photos.

[0096] The embodiment of the present invention is a system embodiment corresponding to the above-mentioned method embodiment. The specific operations of each module can be understood by referring to the description of the method embodiment, which will not be repeated here.

[0097] Device Example 1

[0098] An embodiment of the present invention provides an electronic device, such as Figure 4 As shown, it includes: a memory 40, a processor 42, and a computer program stored in the memory 40 and executable on the processor 42. When the computer program is executed by the processor 42, the following method steps are implemented:

[0099] S1. Pre-train the defect recognition experience pool and reinforcement learning experience pool to set the cruise target point;

[0100] S2. The UAV uses GPS global positioning fusion multi-source navigation method to avoid obstacles and fly to the target point according to the cruise target point;

[0101] S3. Take photos of each target point and identify power line defects based on the photos obtained.

[0102] Device Example 2

[0103] An embodiment of the present invention provides a computer-readable storage medium, on which a program for implementing information transmission is stored. When the program is executed by the processor 42, the following method steps are implemented:

[0104] S1. Pre-train the defect recognition experience pool and reinforcement learning experience pool to set the cruise target point;

[0105] S2. The UAV uses GPS global positioning fusion multi-source navigation method to avoid obstacles and fly to the target point according to the cruise target point;

[0106] S3. Take photos of each target point and identify power line defects based on the photos obtained.

[0107] The computer-readable storage medium in this embodiment includes, but is not limited to, ROM, RAM, magnetic disk or optical disk, etc.

[0108] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A UAV power inspection method based on multi-source navigation and Yolo-v8, characterized in that: include: S1. Pre-train the defect recognition experience pool and reinforcement learning experience pool to set the cruise target point; S2. The UAV uses GPS global positioning fusion multi-source navigation method to avoid obstacles and fly to the target point according to the cruise target point; S3. Take photos of each target point and identify power line defects based on the photos obtained.

2. The method according to claim 1, characterized in that The specific method for training the defect recognition experience pool is: The insulators on the transmission line are collected using a camera gimbal mounted on a drone, and image data of faulty insulators are selected and expanded. Then, the expanded image data is annotated with the insulator defect types, including normal and abnormal defect types. The annotated data set is used as a sample library and divided into a training set and a test set. The SE dual-channel attention mechanism is introduced into the Yolo-v8 network model, and the image data and annotation files in the training set are used to iteratively train the Yolo-v8 network model with the SE dual-channel attention mechanism, wherein the Yolo-v8 network model includes a feature extraction layer, a feature fusion layer and an output layer; During the training process, the model weights with the highest accuracy are saved to complete the construction of the UAV defect recognition experience pool.

3. The method according to claim 1, characterized in that The specific method for training the reinforcement learning experience pool is: Constructing a Markov space for the drone, the Markov space including the current drone state space, the next execution action of the drone, the current reward value of the drone, and the state space transition of the drone; Construct a reward function for the drone, which includes two rewards, one for reaching the target point and the other for failure. When the drone reaches the target point margin, the drone receives a positive reward, and when the reading of the drone's distance sensor is below the threshold, it is punished; Repeat the above process for training according to the cruise route obtained from the cruise target point to obtain the reinforcement learning experience pool of the drone.

4. The method according to claim 1, characterized in that: The UAV uses GPS global positioning fusion multi-source navigation method to avoid obstacles and fly to the target point according to the cruise target point. The specific method is: The drone is equipped with a GPS module to obtain the current coordinates of the drone and the coordinates of the target inspection point, and a second-order Bezier curve is used to generate a smooth curve from the current coordinate point to the target coordinate point. The drone moves along the curve; During the movement, the drone is equipped with a depth camera to obtain depth images, and uses the depth images for Slam mapping to detect whether there are obstacles in front of the drone; If an obstacle is detected, the A* algorithm is used to replan the route for the collision segment on the generated second-order Bezier curve; When a collision occurs in an obstacle, the vertically extended ray intersects the previously planned route, and a penalty item constraint is imposed on the new route. The penalty items include collision item penalty and reinforcement learning experience pool penalty, and the optimized best path is obtained.

5. The method according to claim 1, characterized in that The specific identification of power line defects based on the obtained photos is as follows: Get real-time images through the camera on the drone and input the images to the edge device equipped with the Yolo-v8 model; Based on the experience gained from the defect identification experience pool, the defective insulators are classified and located through the Yolo-v8 model to achieve defect identification.

6. The method according to claim 2, characterized in that The specific method for marking the insulator defect type on the expanded image data is as follows: Create two labels in the annotation interface, corresponding to normal insulators and defective insulators respectively, then use rectangular boxes to mark the insulator targets in each picture, select the label corresponding to the target, and finally generate an xml annotation file with the annotation information; Each XML file corresponds to each annotated picture. The annotation information of the file contains 4 position attributes and 1 category attribute. The position attributes are the horizontal coordinate x_min and vertical coordinate y_min of the upper left corner of the rectangular box, and the horizontal coordinate x_max and vertical coordinate y_max of the lower right corner. The category attribute is the label selected when annotating the picture.

7. The method according to claim 1, characterized in that The SE dual-channel attention mechanism is introduced. The specific steps are as follows: First, the input feature map is globally pooled to change its size to 1×1×the number of channels; Then, the 1×1×channel feature map is adjusted through the fully connected layer and the Sigmoid activation function to generate the weight of each feature map; These weights are multiplied element-by-element with the input features. The compression operation uses global average pooling to generate the channel statistic Z_c. The excitation operation uses a fully connected layer and a Sigmoid activation function to generate a weight value for each feature channel. The calculation formula of Z_c is as follows:

8. A UAV power inspection system based on multi-source navigation and Yolo-v8, characterized in that: include: Target point determination module: used to pre-train the defect recognition experience pool and reinforcement learning experience pool and set the cruise target point; Cruise module: used to enable the UAV to avoid obstacles and fly to the target point according to the cruise target point by using GPS global positioning fusion multi-source navigation method; Defect recognition module: used to take photos of each target point and identify power line defects based on the photos obtained.

9. An electronic device, characterized in that: include: processor; as well as, A memory arranged to store computer executable instructions, which, when executed, enable the processor to implement the steps of the UAV power inspection method based on multi-source navigation and Yolo-v8 as described in any one of claims 1 to 7.

10. A storage medium, characterized in that: Used to store computer executable instructions, which, when executed, implement the steps of the UAV power inspection method based on multi-source navigation and Yolo-v8 as described in any one of claims 1 to 7.

Citation Information

Cited By

  • Distribution line inspection method and device based on unmanned aerial vehicle and medium

    CN120178940A