Distribution network unmanned aerial vehicle inspection autonomous navigation method based on GPS information and visual information

Abstract

The invention relates to the technical field of distribution network unmanned aerial vehicle autonomous tour-inspection navigation, in particular to a distribution network unmanned aerial vehicle tour-inspection autonomous navigation method based on GPS information and visual information, which comprises the following steps: acquiring a GPS and smoothing projection coordinates, constructing a Thiessen region random index, optimizing a route and acquiring a flight image, evaluating a risk and adjusting exposure parameters, and generating a light environment value. According to the method, an environment model is constructed by fusing GPS information, Gaussian projection is adopted to eliminate earth curvature errors, moving average filtering is adopted to suppress signal noise, Thiessen polygon is utilized to quantify routing inspection point and obstacle distribution, a tornado optimization algorithm is combined to generate an optimal route, the shortest path without repetition and obstacle avoidance are realized, and obstacles are identified based on a YOLOv12 model. A risk avoiding strategy is made according to the risk value, a route is dynamically optimized, a light environment fingerprint database is constructed, illumination change is predicted in combination with LSTM, aperture shutter sensitivity and white balance are adjusted in a self-adaptive mode, image brightness and definition are optimized, and inspection efficiency and data reliability are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of distribution network unmanned aerial vehicle autonomous inspection navigation, and particularly relates to a distribution network unmanned aerial vehicle inspection autonomous navigation method based on GPS information and visual information. BACKGROUND

[0002] Distribution network unmanned aerial vehicle inspection refers to the use of unmanned aerial vehicles carrying visual collection equipment to carry out inspection in the complex environment of distribution network lines and related equipment according to a preset task or autonomously planned route. The core is to replace or assist manual distribution network inspection by the flexible movement and data collection capability of the unmanned aerial vehicle. It is of great practical significance to carry out distribution network unmanned aerial vehicle inspection research. As a key link between the power system and users, distribution network has a wide range of application scenarios and is often in complex scenes such as mountains, forests and urban building groups.

[0003] The traditional distribution network inspection method mainly relies on manual inspection. The staff needs to carry detection tools to the location of the distribution network equipment and record the equipment status by visual observation and instrument measurement. For high towers and cross-mountain lines, climbing equipment or walking through complex terrain is also needed. Not only is the inspection cycle long and the labor intensity great, but also the inspection is significantly affected by weather and terrain, and the completeness and timeliness of data collection cannot be guaranteed. Even if unmanned aerial vehicle inspection is introduced in some scenarios, a simple preset route flight mode is still used, and the route is not optimized in combination with environmental information. The visual parameter adjustment relies on manual preset rather than dynamic self-adaptation. Obstacle recognition mostly uses traditional image recognition algorithms, which have insufficient recognition accuracy and risk avoidance response speed. The overall inspection effect and safety need to be improved.

[0004] However, the current distribution network unmanned aerial vehicle still faces many problems to be solved when performing inspection tasks in real complex environments. In terms of inspection route, the inspection route is not optimized, resulting in many repeated routes, which causes the distribution network unmanned aerial vehicle to fly back and forth in the same area, wasting battery endurance resources and greatly increasing the time cost of the overall inspection task, reducing the inspection efficiency. In terms of obstacle avoidance, the distribution network unmanned aerial vehicle has weak perception, recognition and avoidance response capability for surrounding obstacles, and it is difficult to quickly and accurately determine the position and movement trend of obstacles, which is prone to collision risks. Not only can the equipment be damaged, but also the inspection task can be interrupted, affecting the progress of the work. In terms of visual adaptation, the distribution network unmanned aerial vehicle lacks the ability to dynamically adjust the shooting system parameters according to the environmental brightness. When facing different brightness scenes such as strong light, weak light and backlight, it cannot adaptively match the optimal shooting brightness parameters, which makes the inspection images prone to overexposure and weak light problems, affecting the quality of the inspection image at the task point. SUMMARY

[0005] The application aims to solve the problems in the prior art and provides a network distribution unmanned aerial vehicle inspection autonomous navigation method based on GPS information and visual information.

[0006] To achieve the above-mentioned purpose, the application adopts the following technical scheme: a network distribution unmanned aerial vehicle inspection autonomous navigation method based on GPS information and visual information, comprising the following steps: S1: obtaining the GPS coordinates of the inspection task points and surrounding obstacles, calling the airborne GPS to obtain the take-off point coordinates, smoothing the coordinate sequence by the sliding average filtering algorithm, converting the latitude and longitude into plane rectangular coordinates by Gauss projection, and generating plane coordinate data; S2: constructing a Delaunay triangulation network based on the plane coordinate data to determine the topological relationship of adjacent points, forming a Voronoi polygon to divide the area according to the circumscribed radius, numbering each area and randomly shuffling the index by Fisher-Yates shuffling algorithm to generate an inspection area index sequence; S3: calculating the Euclidean distance sum between the task points as the route length according to the inspection area index sequence, taking the reciprocal as the fitness value to sort and divide the population into tornado layer, thunderstorm layer and storm layer, updating the position of the storm layer individuals by difference vector weighting and applying random disturbance, correcting the thunderstorm layer according to the tornado layer, realizing the conversion of continuous solution to discrete index through range mapping and legality repair, and iteratively optimizing to obtain the optimal route index sequence; S4: controlling the unmanned aerial vehicle to fly in sequence according to the optimal route index sequence, collecting forward images in real time to identify the type, size and distance of obstacles, combining the recognition confidence, pixel density and safety threshold to calculate the environmental risk value through a risk assessment formula, triggering obstacle avoidance when the risk value exceeds the threshold and returning to the route planning step to regenerate the optimal route; S5: based on the unmanned aerial vehicle reaching the task point, collecting images to extract light environment information, constructing a light environment fingerprint library and generating a fingerprint sequence, matching the most similar historical fingerprint from the dynamic fingerprint library by cosine similarity and extracting the corresponding time series data to form a historical light sequence, predicting the future light intensity, direction and spectral changes based on the sequence using LSTM, adjusting the aperture, exposure and white balance according to the prediction results, and generating optimized light environment parameters.

[0007] As a further scheme of the application, the specific calculation formula of the circumscribed radius is: , wherein, is the center, is the radius, , , p i , p j , pk The coordinates.

[0008] As a further aspect of the present invention, the specific calculation formula for smoothing coordinate sequence denoising using the moving average filtering algorithm is as follows: , in, These are the filtered coordinates. To adjust the sliding window size, The original GPS coordinate sequence is taken from the first position. The original coordinates are the endpoints, and the continuous... The original coordinates (i.e., from) arrive The average value of () is used as the filtered coordinates. .

[0009] As a further aspect of the present invention, the specific formula for calculating the route length is as follows: , in, Adjacent Thiessen polygons and The Euclidean distance between them It is the penalty coefficient. For the route and the first The penalty value when colliding with an obstacle. Representing the An obstacle.

[0010] As a further aspect of the present invention, the position of the storm layer individuals is updated using a weighted difference vector method, and the specific position calculation formula is as follows: , in, It is the first After the nth iteration, the th Storm locations, It is the first During the nth iteration, the 1st Storm locations, These are random coefficients. It is an adjustment coefficient used to control the update amplitude. It is the location index of the random storm. , These represent the lower and upper bounds of the search space, respectively, used to limit the range of random perturbations.

[0011] As a further aspect of the present invention, the thunderstorm layer is corrected based on the tornado layer, and the specific location is calculated using the following formula: , in, is the position of the i-th tornado in the j-th iteration, is the position of the i-th thunderstorm in the j-th iteration, is the position of the i-th tornado in the j-th iteration, is the position of the i-th thunderstorm in the j-th iteration, is a random index to ensure search diversity, is a random coefficient to adjust the step size moving towards the tornado, is the position of the i-th tornado in the j-th iteration, is the position of the i-th tornado in the j-th iteration, is the position of the i-th thunderstorm in the j-th iteration.

[0012] As a further scheme of the present application, the environmental risk value comprises an identification result and an algorithmic environmental risk, and the specific calculation formula is: , wherein, is the obstacle identification confidence; is the obstacle identification box size; , , are weight coefficients, respectively; is the obstacle threat level.

[0013] As a further scheme of the present application, the environmental risk value further comprises different risk avoidance steps: (1) when , generally a larger size obstacle or bird is identified, and the distribution network unmanned aerial vehicle needs to be vertically pulled up for risk avoidance, and the control distribution network unmanned aerial vehicle braking response time is controlled to vertically rise at the maximum climbing speed , and the specific movement formula is: , wherein, , is the height of the distribution network unmanned aerial vehicle after risk avoidance and before risk avoidance; is the vertical pulling up speed of the distribution network unmanned aerial vehicle, is the risk avoidance action duration; (2) when , generally a tree or interwoven cable is identified, and the distribution network unmanned aerial vehicle needs to be routed and slowly pulled up for risk avoidance according to the flying radius, and the specific position movement formula is: , wherein, is the distance between the distribution network unmanned aerial vehicle and the obstacle; is the maximum turning radius of the distribution network unmanned aerial vehicle; ​​​​​​Is the radius that the unmanned aerial vehicle needs to detour when avoiding obstacles; (3) When , it is generally to identify small obstacles such as stationary dry branches, and the distribution network unmanned aerial vehicle does not need to take emergency measures.

[0014] As a further scheme of the present application, the light environment fingerprint library includes environmental light intensity, light direction angle, spectral distribution proportion The "dynamic fingerprint" of each light environment scene is defined as a multi-dimensional feature vector, and the specific formula is: , Among them, is the normalized light intensity, is the normalized light direction angle, is the proportion of each spectral field; After the distribution network unmanned aerial vehicle collects a new light environment fingerprint , it needs to be compared with the existing light environment fingerprint in the dynamic light environment library, and the cosine similarity is matched. The specific formula for calculating the cosine similarity is: , If the cosine similarity , it is determined that it is a new light environment fingerprint, and it is added to the light environment fingerprint library; if the cosine similarity , there is a similar fingerprint in the light environment fingerprint library, and the light environment fingerprint is discarded.

[0015] As a further scheme of the present application, the specific steps of adjusting the aperture, exposure and white balance according to the prediction result are: (1) Extract the future light intensity from the prediction fingerprint, calculate the future light intensity gradient change rate, and the specific formula is: , Among them, represents the time interval (also called time step) between the two adjacent time points, that is, the time difference between the two time points, is the time point predicted in the future; The light intensity gradient change rate, the positive gradient indicates that the light intensity rises and is easy to overexpose; the negative gradient indicates that the light intensity decreases and is easy to be weak light; by adjusting the aperture , shutter speed , sensitivity of the distribution network unmanned aerial vehicle, the specific formula is: , Among them, , , , , , respectively are the parameter values of aperture, shutter speed and ISO before and after adjustment; respectively are the parameter values of aperture, shutter speed and ISO before and after adjustment; respectively are the parameter values of aperture, shutter speed and ISO before and after adjustment; respectively are the parameter values of aperture, shutter speed and ISO before and after adjustment; (2) Extract the future light direction angle from the predicted fingerprint , and calculate the direction angle change amount, and the specific calculation formula is: When , is the direction angle threshold value, and it is determined that the light direction changes dramatically, and additional exposure compensation is required, and the specific exposure adjustment method is: wherein, is the adjusted exposure value; is the current exposure value; is the compensation coefficient, and the positive compensation is increased when the direction angle changes positively. (3) Extract the future spectral proportion from the predicted fingerprint , and adjust the white balance parameter according to the spectral component change, and the specific adjustment calculation method is: wherein, is the white balance value before adjustment; is the white balance value after adjustment; is the white balance coefficient corresponding to each spectrum, such as reducing the color temperature when the red light proportion rises, and increasing the color temperature when the blue light proportion rises.

[0016] Compared with the prior art, the advantages and positive effects of the present application are: In the present application, the environment model is constructed by fusing GPS information, the earth curvature error is eliminated by using Gaussian projection, the signal noise is suppressed by using sliding average filtering, the optimal route is generated by using the Voronoi polygon quantization patrol point and obstacle distribution combined with the tornado optimization algorithm, the shortest path without repetition and obstacle avoidance is realized, the obstacle is identified based on the YOLOv12 model, the risk avoidance strategy is formulated according to the risk value and the route is dynamically optimized, the light environment fingerprint library is constructed, the light change is predicted by using LSTM, the aperture, shutter and ISO are adaptively adjusted, the white balance is adjusted, the image brightness and clarity are optimized, and the patrol efficiency and data reliability are improved. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 is the overall process schematic diagram of the present application; Figure 2 is the step process schematic diagram of the present application.​​​​​ DETAILED DESCRIPTION

[0018] In order to make the purpose, technical scheme and advantages of the present application more clear, the present application is further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application and do not limit the present application.

[0019] Please refer to Figure 1 , the autonomous navigation method of the distribution network unmanned aerial vehicle inspection based on GPS information and visual information, comprising the following steps: S1: obtaining the GPS coordinates of the inspection task points and the surrounding obstacles, calling the airborne GPS to obtain the take-off point coordinates, smoothing the coordinate sequence by using the moving average filtering algorithm, converting the latitude and longitude into plane rectangular coordinates by using the Gauss projection, and generating the planar coordinate data; Firstly, the GPS position coordinates of a plurality of inspection task points are obtained from the distribution network unmanned aerial vehicle inspection platform, denoted as: At the same time, the GPS position coordinates of a plurality of static obstacles such as trees and buildings are obtained, denoted as: The GPS coordinates of the take-off point are obtained by the GPS module carried by the distribution network unmanned aerial vehicle, denoted as: The original GPS coordinate sequence is smoothed by using the moving average filtering algorithm to remove the random noise and jump in the GPS signal. The specific formula is: , Among them, is the filtered coordinate, is the size of the sliding window, is the original GPS coordinate sequence, taking the average value of the original coordinates from the th original coordinate as the endpoint, and the continuous original coordinates (i.e. from to ) as the filtered coordinate ; Since the latitude and longitude coordinates provided by the GPS signal belong to the spherical coordinate system, direct use in space calculation will cause large error due to the curvature of the earth. The latitude and longitude coordinates are converted into plane rectangular coordinates by using the Gauss projection to eliminate the influence of the curvature of the earth. The obstacles and the inspection points have a certain obstacle range. However, in the environmental modeling process, the data point information cannot directly reflect its regional range. The Voronoi polygon has a unique characteristic: each polygon contains only one original point in the point set, and the distance from any point on the polygon boundary to the adjacent two terminal points is completely equal. In this way, the plane space is divided into a plurality of polygons that do not overlap and can completely cover the whole, so that the regional range of each data point is accurately quantified, and finally the planar coordinate data is obtained; S2: Construct a Delaunay triangulation based on planar coordinate data to determine the topological relationship between adjacent points. Divide the region into regions by forming Thiessen polygons according to the radius of the circumcircle. Number each region and randomly shuffle the index using the Fisher-Yates shuffle algorithm to generate an inspection region index sequence. Based on planar coordinate data, a Delaunay triangulation is constructed to calculate the topological relationship between task points and obstacles. The Delaunay triangulation satisfies the circumcircle property: the circumcircle of any triangle does not contain other points. A Thiessen polygon is constructed based on the circumcircle radius R, and its circumcircle center is... and radius The specific calculation formula is as follows:

[0020] , , , Let p be any three points on the circumcircle of the Delaunay triangulation. i p j p k The coordinates; The Thiessen polygon is defined as: for any point All points within its Thiessen polygon region to The distances to all points are less than the distances to other points. The boundary of the Thiessen polygon is the perpendicular bisector of the line connecting adjacent points. In this way, the planar space is divided into... Let the three non-overlapping and completely covered polygonal regions be denoted as follows: Each region corresponds to an inspection task point, and each individual in the population is encoded as an index sequence of Thiessen polygon regions. The Fisher-Yates shuffle algorithm is used to randomly shuffle P. For i ranging from n to 2, the following steps are performed: Random selection Index in ,exchange and A randomly arranged index sequence is obtained as the initial inspection order of the power distribution drone. In this embodiment, the population size is initialized to 30 individuals, each individual represents a possible inspection path sequence, and finally the inspection area index sequence is generated. S3: Calculate the total Euclidean distance between task points based on the inspection area index sequence to get the route length. Take its reciprocal as the fitness value to sort the population and divide it into tornado layer, thunderstorm layer and storm layer. Individuals in storm layer update their positions through difference vector weighting and apply random perturbation. The thunderstorm layer is corrected according to the tornado layer. Through range mapping and legality repair, the continuous solution is transformed into a discrete index. Iterative optimization is used to obtain the optimal route index sequence. The total Euclidean distance between task points is calculated based on the inspection area index sequence to determine the route length. The specific calculation formula is as follows: , in, are adjacent Thiessen polygons and The Euclidean distance between them It is the penalty coefficient. For the route and the first The penalty value when colliding with an obstacle. Representing the There are several obstacles; if the flight path passes through the Thiessen polygon region corresponding to the obstacle, it is determined to be an infeasible solution. Based on fitness values, individuals in the population are divided into three strata: tornadoes, thunderstorms, and windstorms. Tornadoes represent the current optimal solution in the population, thunderstorms represent a relatively optimal solution, and windstorms represent potential solutions after removing tornadoes and thunderstorms. Windstorm individuals evolve towards tornado and thunderstorm individuals, with the specific position update formula as follows: , in , , , They are the first Before and after the iteration, the... The and the first Storm location; , It means the first The th iteration in the The first tornado and the first The location of the thunderstorm; It is a random coefficient used to adjust the step size of the movement towards the tornado; It is the location index of the random storm, increasing the diversity of movement directions; This is the storm speed, which simulates the physical effects of the Coriolis force, centrifugal force, etc. The specific calculation formula is as follows: , in, Indicates the first The storm was in the 1st The velocity vector of the next iteration; Indicates the first The current velocity vector of the storm; The convergence factor has a value of 0.7298. It is a fuzzy adaptive kinetic energy parameter, with a value range of [value range missing]. ; These are random coefficients; is the Coriolis parameter, where is calculated, is the earth rotation speed, taking the value of ; , are the radii of curvature of the storm tracks in the northern and southern hemispheres, respectively, calculated by the formula , simulation; , are the Coriolis-related terms in the northern and southern hemispheres, calculated as: , where, is the position difference between the storm and the random tornado, in addition to the storm evolving towards the thunderstorm and the tornado evolving towards the thunderstorm, the specific position calculation formula is: , where, is the position of the th thunderstorm at the th iteration, is the position of the th thunderstorm at the th iteration, and are random indexes to ensure search diversity, is a random coefficient used to adjust the step size of moving towards the tornado, is the position of the th tornado at the th iteration, is the position of the th thunderstorm at the th iteration, and to enhance the exploration ability, the storm can randomly generate a new position, the specific position calculation method is: , where, is the position of the th storm after the th iteration, is the position of the th storm at the th iteration, is a random coefficient, is an adjustment coefficient used to control the update amplitude, is the position index of the random storm, , represent the lower and upper bounds of the search space, respectively, used to limit the range of random disturbance; The population position after evolution presents a continuous decimal form, which cannot be directly corresponded to the discrete index sequence of the Thiessen polygon edge. The conversion from continuous solution to discrete index needs to be realized through two steps of range mapping and legality repair. The continuous position value after evolution is mapped to the effective interval of the Thiessen polygon edge index, and the index interval range is ; for the population individual , linear scaling is mapped to the interval, and the specific calculation formula is: , wherein, is a scaling coefficient, which is fitted by the upper and lower bounds of the decision space and the index interval , and the specific calculation formula is: , For the individual in the population which exceeds the index interval range after mapping, truncation processing is adopted, and the specific processing formula is: , The population individual is mapped in the index interval through the scaling mapping operation, but there may be problems of repetition or non-adjacency. If the converted sequence has repeated indexes, the first appearing index is retained, and the remaining repeated positions are supplemented with the edge indexes not contained in a random order until all the to-be-inspected points are covered, and the optimal flight route index sequence is finally obtained. S4: According to the optimal flight route index sequence, the unmanned aerial vehicle is controlled to fly in sequence, real-time collection of forward images identifies the type, size and distance of obstacles, and combines the recognition confidence, pixel density and safety threshold to calculate the environmental risk value through the risk evaluation formula. When the risk value exceeds the threshold, the obstacle avoidance is triggered and the flight route planning step is returned to regenerate the optimal flight route. According to the optimal flight route index sequence, the unmanned aerial vehicle is controlled to fly in sequence, and the forward images of the on-board camera are continuously collected, 30 frames per second. The target detection algorithm is used to identify the obstacles in the surrounding environment. For the jth detected obstacle, the following information is obtained: Obstacle type: trees, buildings, cables, flying birds, etc. Recognition confidence: detection confidence output by the model. Obstacle size: pixel width and height in the image. Obstacle distance: estimated distance through binocular vision or monocular ranging. According to the obstacle recognition result, the environmental risk value is calculated through the risk evaluation formula.

[0021] , wherein, is the obstacle recognition confidence. is the size of the obstacle recognition box; , , are weight coefficients, respectively; is the obstacle threat level, the tree type obstacle is less threatening to the distribution network unmanned aerial vehicle, and the threat level is 1; the construction machinery has potential threat, and the threat level is 2; the cable is generally located between the power transmission towers, and the distribution range is relatively miscellaneous, and the threat level is 3; the flying speed of the flying bird is fast, and the unmanned aerial vehicle needs to take risk avoidance operation in time, and the threat level is 4, different risk avoidance operations are made according to the environmental risk, when , generally, a larger size obstacle or a flying bird is recognized, the distribution network unmanned aerial vehicle needs to be vertically pulled up to avoid risk, the braking response time of the distribution network unmanned aerial vehicle is controlled, and the distribution network unmanned aerial vehicle is vertically lifted at the maximum climbing speed , the specific movement calculation formula is: , wherein, , is the height of the distribution network unmanned aerial vehicle after risk avoidance and before risk avoidance; is the vertical lifting speed of the distribution network unmanned aerial vehicle, is the duration of the risk avoidance action; when , generally, a tree or an interwoven cable is recognized, the distribution network unmanned aerial vehicle needs to be routed and slowly pulled up to avoid risk according to the flying radius, and the specific position movement calculation formula is: , wherein, is the distance between the distribution network unmanned aerial vehicle and the obstacle; is the maximum turning radius of the distribution network unmanned aerial vehicle; is the radius indicating that the unmanned aerial vehicle needs to be routed when avoiding obstacles; when , generally, a small obstacle such as a static dry branch is recognized, and the distribution network unmanned aerial vehicle does not need to take emergency risk avoidance; In the distribution network unmanned aerial vehicle inspection process, the core goal of local route planning is: when the distribution network unmanned aerial vehicle detects that the surrounding environmental risk value , it is determined that the position and the surrounding area of the current unmanned aerial vehicle are safe, and there is no significant collision threat, at this time, the core algorithm in the "distribution network unmanned aerial vehicle route planning based on tornado optimization algorithm" needs to be called again, the inspection route is optimized and adjusted, and the optimal route is regenerated to ensure the optimality and continuity of the route; S5: After the UAV arrives at the mission point, it collects images to extract light environment information, builds a light environment fingerprint database and generates fingerprint sequences. It matches the most similar historical fingerprints from the dynamic fingerprint database using cosine similarity and extracts the corresponding time series data to form a historical illumination sequence. Based on this sequence, it uses LSTM to predict future changes in light intensity, direction and spectrum. It adjusts the aperture, exposure and white balance according to the prediction results to generate optimized light environment parameters. After the power distribution network drone arrives at the inspection task point, it acquires the lighting environment information from images taken along the flight path, specifically including ambient light intensity, illumination direction angle, and spectral distribution percentage. The "dynamic fingerprint" for each lighting environment scene is defined as a multi-dimensional feature vector, with the specific formula as follows: , in, It is the normalized light intensity. It is the normalized illumination direction angle. It refers to the proportion of each spectral field; power distribution drones are collecting new light environment fingerprints. Then, it needs to be compared with the existing light environment fingerprints in the dynamic light environment database. Cosine similarity matching comparison; the specific formula for calculating cosine similarity is as follows: , If cosine similarity If it is determined to be a new light environment fingerprint, it will be added to the light environment fingerprint database; if the cosine similarity is... The optical environment fingerprint database already contains a similar fingerprint, so this optical environment fingerprint is discarded; the optical environment fingerprint collected by the distribution network drone at time t is... Sampling interval Seconds, historical data window is the previous fingerprint sequence at each moment Simultaneously, cosine similarity matching is used to retrieve fingerprints from the dynamic lighting environment fingerprint database that match the fingerprints obtained from the database. Most similar The complete time sequence corresponding to each historical fingerprint forms a historical illumination sequence. Using historical fingerprint sequences and temporal features of similar fingerprints matched in the fingerprint database as model input, a Long Short-Term Memory (LSTM) network is employed to predict the ambient light fingerprint at the next five time points. Each predicted fingerprint contains a normalized value of future illumination intensity. Normalized value of future illumination direction angle The proportion of each spectral field in the future Adjust the aperture, exposure, and white balance based on the prediction results: (1) Extracting future illumination intensity from predicted fingerprints The formula for calculating the rate of change of future light intensity gradient is as follows: , , is the time interval between the adjacent two time points (also called time step), that is, the time difference between the two adjacent time points, is the time point to be predicted in the future; Light intensity gradient change rate, positive gradient indicates that the light intensity rises, which is easy to overexposure; negative gradient indicates that the light intensity decreases, which is easy to weak light; adjust the aperture, shutter speed and ISO of the aerial unmanned vehicle shooting through the future light intensity gradient change rate , , Sensitivity, the specific calculation formula is: , , , , , , , are the parameter values of the adjusted and unadjusted aperture, shutter speed and ISO respectively; , , , are the aperture adjustment coefficient, shutter speed adjustment coefficient and ISO adjustment coefficient respectively; , (2) Extract the future light direction angle from the predicted fingerprint , calculate the direction angle change, the specific calculation formula is: , When , is the direction angle threshold value, which determines that the light direction changes dramatically and needs additional compensation for exposure, and the specific adjustment of exposure is: , , is the adjusted exposure value; is the current exposure value; is the compensation coefficient, which increases the positive compensation when the direction angle changes positively; (3) Extract the future spectral proportion from the predicted fingerprint , adjust the white balance parameter according to the spectral component change, and the specific adjustment calculation method is: , , is the white balance value before adjustment; is the white balance value after adjustment; is the white balance coefficient corresponding to each spectrum, such as reducing color temperature when red light proportion rises, and increasing color temperature when blue light proportion rises; According to the above multiple parameters, the optimized light environment parameters can be generated to complete image acquisition, the task point image acquisition of the network unmanned aerial vehicle, the next inspection task point is arrived, and all the inspection task points are completed.

[0022] The above is only a preferred embodiment of the present application, and does not limit the present application in other forms. Any person skilled in the art can use the disclosed technical content to make changes or modifications as equivalent embodiments applied to other fields, but any simple modification, equivalent change and modification made to the above embodiments without departing from the technical solution content of the present application, according to the technical essence of the present application, still belongs to the protection scope of the technical solution of the present application.

Claims

1.A method for autonomous navigation of a UAV for distribution network inspection based on GPS information and visual information, characterized in that, The method comprises the following steps: S1: obtaining the GPS coordinates of the inspection task points and surrounding obstacles, calling the onboard GPS to obtain the takeoff point coordinates, smoothing the coordinate sequence with a moving average filtering algorithm, converting the latitude and longitude into plane rectangular coordinates using Gauss projection, and generating planar coordinate data; S2: constructing a Delaunay triangulation based on the planar coordinate data to determine the topological relationship of adjacent points, forming a Voronoi polygon to divide the area according to the circumscribed radius, numbering each region and randomly shuffling the index through the Fisher-Yates shuffle algorithm to generate an inspection region index sequence; S3: calculating the sum of the Euclidean distances between the task points as the route length according to the inspection region index sequence, taking the reciprocal as the fitness value, sorting the population into tornado layer, thunderstorm layer and storm layer, updating the position of the storm layer individuals through weighted difference vectors and applying random disturbance, correcting the thunderstorm layer according to the tornado layer, converting continuous solutions to discrete indexes through range mapping and legality repair, and iteratively optimizing to obtain the optimal route index sequence; S4: controlling the UAV to fly in sequence according to the optimal route index sequence, collecting forward images in real time to identify the type, size and distance of obstacles, combining the recognition confidence, pixel density and safety threshold to calculate the environmental risk value through a risk assessment formula, triggering obstacle avoidance and returning to the route planning step to regenerate the optimal route when the risk value exceeds the threshold; S5: based on the UAV reaching the task point, collecting images to extract light environment information, constructing a light environment fingerprint library and generating a fingerprint sequence, matching the most similar historical fingerprint from the dynamic fingerprint library through cosine similarity and extracting the corresponding time series data to form a historical light sequence, predicting future light intensity, direction and spectral changes based on the sequence using LSTM, adjusting the aperture, exposure and white balance according to the prediction results, and generating optimized light environment parameters. 2.The GPS information and visual information based network distribution unmanned aerial vehicle inspection autonomous navigation method according to claim 1, characterized in that, The specific calculation formula of the circumscribed radius is: ， wherein, is the center of the circle, and r is the radius, , , are the coordinates of any three points p i , p j , p k on the circumcircle of the Delaunay triangulation. 3.The GPS information and visual information based network distribution unmanned aerial vehicle inspection autonomous navigation method according to claim 1, characterized in that, The specific calculation formula of the moving average filtering algorithm for smoothing the coordinate sequence is: ， wherein, is the filtered coordinate, is the size of the sliding window, is the sequence of original GPS coordinates, taken from the th original coordinate as the end point, the average of the consecutive original coordinates, i.e. from to is the filtered coordinate . 4.The GPS information and visual information based network distribution unmanned aerial vehicle inspection autonomous navigation method according to claim 1, characterized in that, The specific calculation formula of the route length is: ， wherein is the Euclidean distance between and , is a penalty coefficient, is the penalty value when the route collides with the th obstacle, represents the th obstacle. 5.The GPS information and visual information based network distribution unmanned aerial vehicle inspection autonomous navigation method according to claim 1, characterized in that, The storm layer individuals update their positions through weighted difference vectors, and the specific position calculation formula is: ， wherein, is the th storm position after the th iteration, is the th storm position at the th iteration, is a random coefficient, is a tuning coefficient for controlling the update magnitude, is the index of the random storm, , respectively represent the lower bound and the upper bound of the search space for limiting the range of random perturbation. 6.The GPS information and visual information based network distribution unmanned aerial vehicle inspection autonomous navigation method according to claim 1, characterized in that, The specific position calculation formula of the thunderstorm layer according to the tornado layer correction is: ， in, It means the first During the nth iteration, the 1st The location of the thunderstorm. It means the first In the nth iteration The location of the thunderstorm. and It uses a random index to ensure search diversity. These are random coefficients used to adjust the step size for moving towards the tornado. It means the first The th iteration The location of the tornado It means the first The th iteration The location of the thunderstorm. 7.The GPS information and visual information based network distribution unmanned aerial vehicle inspection autonomous navigation method according to claim 1, characterized in that, The environmental risk value includes the recognition result and the algorithm environmental risk, and the specific calculation formula is: ， wherein, is an obstacle recognition confidence; is an obstacle recognition box size; , , are weight coefficients, respectively; is an obstacle threat level. 8.The GPS information and visual information based network distribution unmanned aerial vehicle inspection autonomous navigation method according to claim 7, characterized in that, The environmental risk value also includes different risk avoidance steps: (1) When , it is generally identified that the size of the obstacle or bird is large, and the distribution network unmanned aerial vehicle needs to be vertically pulled up to avoid danger, the control distribution network unmanned aerial vehicle brake response time , the maximum climbing speed is vertically lifted, and the specific movement formula is: , wherein, , is the height of the distribution network unmanned vehicle after the risk avoidance and before the risk avoidance, is the vertical pulling speed of the distribution network unmanned vehicle, is the duration of the risk avoidance action; (2) When , it is generally identified that the tree or the interwoven cable, and the distribution network unmanned aerial vehicle needs to fly around and slowly pull up to avoid danger according to the flying radius. The specific position moving formula is: ， wherein, is the distance between the distribution network unmanned aerial vehicle and the obstacle; is the maximum turning radius of the distribution network unmanned aerial vehicle; is the radius representing the need for the unmanned aerial vehicle to detour when avoiding obstacles; (3) When , it is generally to identify small obstacles such as stationary dead branches, and the distribution network unmanned aerial vehicle does not need to take emergency measures. 9.The GPS information and visual information based network distribution unmanned aerial vehicle inspection autonomous navigation method according to claim 1, characterized in that, The light environment fingerprint library includes an ambient light intensity, a light direction angle, and a spectral distribution proportion The "dynamic fingerprint" of each light environment scene is defined as a multidimensional feature vector, and the specific formula is: , in, It is the normalized light intensity. It is the normalized illumination direction angle. It refers to the proportion of each spectral field; power distribution drones are collecting new light environment fingerprints. Then, it needs to be compared with the existing light environment fingerprints in the dynamic light environment database. Cosine similarity matching comparison; the specific formula for calculating cosine similarity is as follows: ， If the cosine similarity , it is determined that the new light environment fingerprint, and added to the light environment fingerprint library; if the cosine similarity , there is a similar fingerprint in the light environment fingerprint library, and the light environment fingerprint is discarded. 10.The GPS information and visual information based network distribution unmanned aerial vehicle inspection autonomous navigation method according to claim 1, characterized in that, The specific steps of adjusting the aperture, exposure and white balance according to the prediction results are: (1) Extracting future light intensity from the predicted fingerprint , calculating the future light intensity gradient change rate, the specific formula is: ， Wherein, Represent the time interval between two adjacent time, also called time step, that is, the time difference between the two time, Is the future of the first time; Light intensity gradient change rate, positive gradient indicates that the light intensity rises, which is easy to overexposure; Negative gradient indicates that the light intensity decreases, which is easy to weak light; Adjust the aperture of the network unmanned aerial vehicle through the future light intensity gradient change rate , shutter speed , Sensitivity, the specific calculation formula is: ， wherein, , , , , , are the parameter values of the adjusted and unadjusted aperture, shutter speed, sensitivity, respectively; , , are the aperture adjustment coefficient, shutter speed adjustment coefficient, sensitivity adjustment coefficient, respectively. (2) Extracting future light direction angle from predicted fingerprint , and the specific calculation formula is: ， When time, is the direction angle threshold value, which determines that the illumination direction changes dramatically and additional compensation for exposure is needed. The specific adjustment of the exposure method is: ， wherein, is the adjusted exposure value; is the current exposure value; is a compensation factor, increasing positive compensation when the orientation angle changes positively; (3) Extracting future spectral proportion from the predicted fingerprint Adjusting the white balance parameter according to the spectral component change, and the specific adjustment calculation manner is: ， wherein, is the white balance value before adjustment; is the white balance value after adjustment; is the white balance coefficient corresponding to each spectrum, such as reducing the color temperature when the proportion of red light rises, and increasing the color temperature when the proportion of blue light rises.

Images