An unmanned aerial vehicle based power line inspection method

Abstract

The application discloses a kind of power line inspection methods based on unmanned aerial vehicle, it is related to circuit inspection technical field, the method comprises: obtaining time series remote sensing image, based on the time series remote sensing image, construct reference growth model;Obtain three-dimensional space model;Through optical load collection current cycle inspection image, and obtain the current form parameter of current independent vegetation unit;Current form parameter and the history growth sequence of corresponding independent vegetation unit in time series remote sensing image are time series alignment and consistency analysis, obtain real-time growth model;Obtain predicted effective accumulated temperature data, and input into each real-time growth model, potential growth state profile of each vegetation unit at the end of early warning period is predicted;If minimum predicted spatial distance is less than the preset safety distance threshold, generate risk early warning information of encroachment limit.Solve the technical problem that the prior art lacks depth analysis to vegetation growth dynamic process, leading to encroachment limit risk early warning lag.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of circuit inspection, in particular to a power line inspection method based on a UAV. BACKGROUND

[0002] With the continuous expansion of the power system scale, the geographical environment traversed by the power line corridor is increasingly complex, and the threat of vegetation growth in the corridor to the safe operation of the line is increasingly prominent. However, the traditional manual inspection method is limited by the terrain, has low inspection efficiency, and is difficult to accurately predict the growth trend of vegetation in the long term.

[0003] The existing inspection technology based on a UAV mainly focuses on the static detection of the current vegetation state, lacks in-depth analysis of the dynamic process of vegetation growth, and thus leads to a lag in the early warning of potential intrusion risks, which cannot meet the forward-looking needs of the safe operation and maintenance of the power line. SUMMARY

[0004] The embodiments of the present application provide a power line inspection method based on a UAV, which solves the technical problem of a lag in the early warning of intrusion risks caused by the lack of in-depth analysis of the dynamic process of vegetation growth in the prior art.

[0005] The technical solution of the present application to solve the above technical problem is as follows:

[0006] The present application provides a power line inspection method based on a UAV, which comprises:

[0007] Obtaining time-series remote sensing images of a target power line corridor in multiple historical growth periods, and constructing a reference growth model of each independent vegetation unit in the corridor based on the time-series remote sensing images;

[0008] Obtaining a three-dimensional space model of the target power line corridor, wherein the three-dimensional space model at least contains spatial position information of conductors and towers;

[0009] Collecting current period inspection images of the target power line corridor by an optical load carried by a UAV, and identifying and segmenting each current independent vegetation unit from the current period inspection images to obtain current morphological parameters of each current independent vegetation unit;

[0010] Performing time-series alignment and consistency analysis on the current morphological parameters and the historical growth sequence of the corresponding independent vegetation unit in the time-series remote sensing images, dynamically adjusting the reference growth model based on the consistency analysis result, and obtaining a real-time growth model adapted to the current growth stage;

[0011] Obtaining predicted effective accumulated temperature data in a future preset warning period, and inputting the predicted effective accumulated temperature data into each real-time growth model to predict the potential growth state profile of each vegetation unit at the end of the warning period;

[0012] Based on the three-dimensional space model and the potential growth state profile, the minimum predicted spatial distance of each vegetation unit and the nearby conductor is calculated, and if the minimum predicted spatial distance is less than a preset safety distance threshold, an encroachment risk warning information for the corresponding vegetation unit is generated.

[0013] The present application provides one or more technical solutions, at least with the following technical effects or advantages:

[0014] The embodiment of the present application provides a power line inspection method based on a UAV. First, time-series remote sensing images of a power line corridor in the past multiple growth periods are acquired, historical accumulated temperature data is combined, a baseline growth model reflecting the variation law of the morphological parameters of each independent vegetation unit with effective accumulated temperature is constructed for each independent vegetation unit, and historical trend modeling of vegetation growth characteristics is realized. Second, current period inspection images are collected by using a UAV optical load, morphological parameters of each current independent vegetation unit are extracted through preprocessing, feature recognition, instance segmentation and other steps, and data support is provided for growth state analysis. Third, the current morphological parameters and the historical growth sequence are time-aligned, consistency analysis is completed through relative deviation calculation and goodness-of-fit test, the baseline growth model is dynamically adjusted to adapt to the current growth stage, and the timeliness and accuracy of model prediction are ensured. On this basis, predicted effective accumulated temperature in a future warning period is acquired, a vegetation potential growth state profile is obtained by inputting the real-time growth model, and the minimum predicted spatial distance of the vegetation and the conductor is calculated through discretization sampling combined with a power line three-dimensional space model. After comparison with a safety threshold, an encroachment risk warning is generated, and prospective assessment and warning of vegetation growth risk of the power line corridor are realized.

[0015] Through the above technical solutions, the dynamic prediction model of the present application fuses historical growth rules and real-time monitoring data, combines power line spatial layout information, provides a basis for the power line operation and maintenance department to develop differentiated obstacle removal plans, effectively reduces the line failure probability caused by vegetation growth, and improves the safety and economy of power grid operation. BRIEF DESCRIPTION OF DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0017] Figure 1 is a flowchart of a power line inspection method based on a UAV provided by the embodiment of the present application. DETAILED DESCRIPTION

[0018] The embodiment of the present application provides a power line inspection method based on a UAV, and is used for solving the technical problem that the prior art lacks deep analysis on a vegetation growth dynamic process, thereby causing lag of a limit intrusion risk early warning.

[0019] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person skilled in the art without creative work fall within the protection scope of the present application.

[0020] In the description of the present application, the terms "first", "second" are only used for description purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first", "second" can explicitly or implicitly include one or more of the features. In the description of the present application, the meaning of "multiple" is two or more, unless otherwise explicitly and specifically limited.

[0021] In the description of the present application, the term "for example" is used to indicate "as an example, illustration or explanation". Any embodiment described as "for example" in the present application is not necessarily interpreted as more preferred or more advantageous than other embodiments. In order to enable any person skilled in the art to implement and use the present application, the following description is given. In the following description, details are listed for the purpose of explanation. It should be understood that a person skilled in the art can realize the present application without using these specific details. In other examples, well-known structures and processes will not be described in detail to avoid unnecessary details making the description of the present application obscure. Therefore, the present application is not intended to be limited to the shown embodiments, but is consistent with the broadest scope of principles and features disclosed in the present application.

[0022] Embodiments, such as Figure 1 The embodiment of the present application provides a power line inspection method based on a UAV, including:

[0023] S10: acquiring time sequence remote sensing images of a target power line corridor in multiple historical growth periods, and constructing a reference growth model of each independent vegetation unit in the corridor based on the time sequence remote sensing images;

[0024] In the embodiment of the present application, the time sequence remote sensing images of the target power line corridor in the multiple historical growth periods are acquired, and the reference growth model of each independent vegetation unit in the corridor is constructed.

[0025] Specifically, the time-series remote sensing images of the target power line corridor in the past at least 3 complete growth cycles are acquired, and the image data need to include high-resolution optical images of key phenological nodes such as spring budding period, summer vigorous growth period, and autumn leaf falling period, and the spatial resolution is not less than 0.5 meters.

[0026] Firstly, the acquired time-series remote sensing images are preprocessed, including radiation calibration, atmospheric correction, geometric precision correction and image registration, to ensure that the spatial position accuracy deviation of images at different periods is less than 1 pixel. An object-oriented image segmentation method is used to identify and extract each independent vegetation unit from the preprocessed images, wherein the division of the independent vegetation unit is based on plant species, spatial distribution continuity and growth boundary. For arbor vegetation, a single independent vegetation unit is a single tree; for shrub or herbaceous vegetation, it is divided into continuous growth areas according to the community boundary.

[0027] Further, for each independent vegetation unit, its morphological parameters in each historical period are extracted from the time-series remote sensing images, for example, the height is obtained by subtracting the DSM data from the DTM digital terrain model, the crown width is calculated by the minimum circumscribed rectangle method, and the crown volume is estimated based on the height and crown width parameters to construct an ellipsoid or cylindrical model. At the same time, the daily weather data of the target power line corridor region in the corresponding historical growth cycle are collected, including daily mean temperature, maximum temperature, minimum temperature and precipitation, and based on the preset plant growth base point temperature, such as 5℃ for arbor and 10℃ for shrub, the effective accumulated temperature data sequence of each historical period is calculated. The effective accumulated temperature is the cumulative value of daily temperature values higher than the growth base point temperature.

[0028] The historical morphological parameter sequence of the independent vegetation unit is associated with the corresponding historical effective accumulated temperature data sequence to construct a growth model data set with effective accumulated temperature as the independent variable and morphological parameter as the dependent variable. A nonlinear curve fitting method is used to construct a reference growth model.

[0029] The reference growth model is constructed by the following steps:

[0030] extracting the historical morphological parameter sequence of each independent vegetation unit in the time-series remote sensing images in multiple historical growth cycles;

[0031] acquiring the historical effective accumulated temperature data sequence corresponding to the multiple historical growth cycles;

[0032] for each independent vegetation unit, based on the historical morphological parameter sequence of the independent vegetation unit and the corresponding historical effective accumulated temperature data sequence, a growth curve model representing the change of the morphological parameter of the independent vegetation unit with the effective accumulated temperature is fitted to obtain a reference growth model of the independent vegetation unit.

[0033] In the embodiments of the present application, first, the historical morphological parameter sequence of each independent vegetation unit in multiple historical growth cycles is extracted from the time sequence remote sensing image, including at least one of vegetation height, crown diameter, projected area, and target tracking is performed on images of different periods.

[0034] Specifically, after the independent vegetation unit is obtained by image segmentation for the first time, a unique identifier is assigned to each unit, and the same vegetation unit in images of different periods is associated by a feature matching algorithm, such as SIFT feature point matching combined with RANSAC algorithm to remove false matching points, to ensure that the extracted morphological parameter sequence belongs to the same entity. For the extraction of morphological parameters, height data is obtained by the difference between the same period DSM and DTM, and the DSM data is filtered to remove abnormal values caused by terrain undulations or noise; the crown width is calculated by the minimum circumscribed rectangle method, that is, the contour boundary of the vegetation unit is extracted by the Canny edge detection algorithm, and then the rotating puzzle algorithm is used to calculate the minimum rectangle that can completely enclose the contour, and the length of the long axis and the short axis of the rectangle is the size of the two directions of the crown width, and the average value is taken as the final crown width value; the crown volume can be estimated according to the vegetation type by selecting a suitable geometric model, such as the ellipsoid model for trees, and the volume calculation formula is 4 / 3 x pi x (crown width / 2) 2 x (height / 2), and the cylindrical model for shrubs, and the volume is pi x (crown width / 2) 2 x height.

[0035] Secondly, the historical effective accumulated temperature data sequence corresponding to multiple historical growth cycles is obtained, and the start and end time of each historical growth cycle is determined. The historical effective accumulated temperature data sequence needs to be counted according to the vegetation type, and each vegetation type corresponds to an independent growth base point temperature, which is determined based on the experimental data of the physiological characteristics of the vegetation of this type. For deciduous vegetation, the growth cycle usually starts from the date when the spring daily mean temperature stabilizes through the growth base point temperature, and ends on the date when the autumn daily mean temperature is continuously lower than the growth base point temperature; evergreen vegetation takes a year as a cycle. Collect daily weather data from national weather stations or regional automatic stations in the target area, and for areas not covered by weather stations, spatial interpolation methods can be used to obtain data from surrounding weather stations, such as Kriging interpolation method. The corresponding growth base point temperature is determined according to the plant species of the vegetation unit, for example, the growth base point temperature of Pinaceae trees is 5°C, and the growth base point temperature of Rosaceae shrubs is 10°C, and then the effective temperature value is calculated daily. When the daily mean temperature is higher than the growth base point temperature, the effective temperature value is the daily mean temperature minus the growth base point temperature; when the daily mean temperature is lower than or equal to the growth base point temperature, the effective temperature value is 0, and the effective accumulated temperature data sequence of each historical period is accumulated to obtain the effective accumulated temperature data sequence of each historical period.

[0036] Further, for each independent vegetation unit, a baseline growth model is fitted based on its historical morphological parameter sequence and corresponding historical effective accumulated temperature data sequence. Firstly, the data is preprocessed. The outliers in the historical morphological parameter sequence due to poor image quality or extraction algorithm error are removed. The 3σ criterion can be used for judgment, that is, if the deviation of a data point from the average value of the sequence exceeds 3 times the standard deviation, it is considered as an outlier and is removed. For missing morphological parameter data, linear interpolation method or sliding average method based on adjacent period data is used for filling.

[0037] Then, the relationship type between the morphological parameter and the effective accumulated temperature is analyzed. Common relationship types include: Logistic curve suitable for S-shaped growth under limited environmental resources, Gompertz curve with slow initial growth, then acceleration, and finally tend to be flat, and polynomial curve suitable for complex growth mode. By calculating the determination coefficient R², root mean square error RMSE and Akaike information criterion AIC of different models, the model with the best fitting effect is selected as the baseline growth model of the independent vegetation unit.

[0038] Exemplarily, when the R² value of the Logistic model is 0.92, the RMSE is 0.35 m, and the AIC is -12.5, while the R² value of the Gompertz model is 0.89, the RMSE is 0.42 m, and the AIC is -10.2, then the Logistic model is preferentially selected as the baseline growth model. The mathematical expression of the model is: y=K / (1+exp(a-bx)), where y is the morphological parameter such as height or crown width, x is the effective accumulated temperature, K is the saturation value of the morphological parameter, a and b are model fitting parameters, which are obtained by fitting the historical data by Levenberg-Marquardt nonlinear least squares algorithm.

[0039] S20: Obtain a three-dimensional space model of the target power line corridor, wherein the three-dimensional space model at least contains spatial position information of conductors and towers;

[0040] In the embodiment of the application, the three-dimensional space model of the target power line corridor is obtained. Firstly, the unmanned aerial vehicle laser radar load is used to collect point cloud data of the target power line corridor. Parallel flight lines are arranged along the direction of the line. The flight line spacing is set according to the conductor height and the laser point cloud density requirement, so as to ensure that the point cloud coverage rate of the conductors and towers is not less than 95% and the point cloud density is not less than 20 points per square meter.

[0041] Secondly, the collected original point cloud data is preprocessed, including denoising, point cloud registration, ground point separation and power facility classification, that is, the noise points generated by flying birds, clouds and the like are removed first, the point clouds of multiple flight lines are spliced to a unified coordinate system, and then the gradual TIN encryption algorithm is used to extract the ground points, and the lead point set and the tower point set are separated from the point cloud by using the recognition method based on geometric features. For the tower, the three-dimensional coordinates of the center point, the tower height, the cross arm length and the insulator string suspension point position are extracted; for the conductor, the least square method is used to fit the conductor point set to obtain the catenary equation of the conductor in each span, and the equation parameters include span, sag, suspension point height, etc., and the complete three-dimensional model of the conductor is constructed in combination with the tower position information.

[0042] The finally generated three-dimensional space model is stored in an absolute geodetic coordinate system, such as the WGS84 coordinate system. The spatial position accuracy of the conductor model needs to meet: the plan position error is not greater than 0.1 meters, and the height error is not greater than 0.05 meters, so as to ensure the accuracy of subsequent distance calculation.

[0043] S30: acquiring current morphological parameters of each current independent vegetation unit by identifying and segmenting each current independent vegetation unit from the current periodical inspection image of the target power line corridor through the optical load carried by the unmanned aerial vehicle;

[0044] In the embodiment of the application, the current periodical inspection image of the target power line corridor is collected by the optical load carried by the unmanned aerial vehicle, wherein the optical load uses an RGB camera with more than 20 million pixels, and is matched with a mechanical shutter and a global exposure sensor to ensure that the image has no motion blur. When flying, a flight path planning with a 50% heading overlap and an 80% lateral overlap is used to obtain high-resolution images covering the entire corridor area, and the image ground resolution is not less than 0.1 meters.

[0045] The current periodical inspection image is preprocessed, including the same radiation calibration, atmospheric correction and geometric fine correction as the historical time sequence image. The corrected image and the historical image used for constructing the reference growth model use the same coordinate system, and the spatial position deviation is controlled within 0.3 pixels.

[0046] The same object-oriented image segmentation parameters as in S10 are used to identify and segment the current period inspection image into independent vegetation units. The current segmentation result is associated with the historical independent vegetation units through a feature matching algorithm to ensure the consistency of the identifiers of the same vegetation unit in different periods. For new vegetation units, such as naturally growing seedlings or newly planted vegetation, a new unique identifier is assigned and included in the current independent vegetation unit set. The current morphological parameters of each current independent vegetation unit are extracted from the current period inspection image, including the current height calculated by subtracting the historical DTM model from the DSM data generated by the stereo matching technology, the current crown width calculated by the same minimum circumscribed rectangle method as the historical morphological parameter extraction, and the current crown volume estimated based on the current height and crown width, to ensure that the current morphological parameters are completely consistent with the parameter types and calculation methods in the historical baseline growth model, thereby realizing the compatibility of subsequent model dynamic adjustment.

[0047] Specifically, step S30 in the method comprises:

[0048] The current period inspection image is preprocessed to enhance the contrast features of the vegetation region and the background.

[0049] Based on the preset vegetation spectral feature library and spatial texture feature library, potential vegetation regions are identified and preliminarily screened from the preprocessed image.

[0050] The potential vegetation regions are instance segmented to separate and identify each current independent vegetation unit according to the independence and spatial continuity of geometric contours.

[0051] For each current independent vegetation unit obtained by segmentation, a corresponding circumscribed polygon is calculated, and current morphological parameters representing the size and contour of the current independent vegetation unit are extracted based on the circumscribed polygon, wherein the current morphological parameters include at least one of vegetation height, crown diameter, and projected area.

[0052] In the embodiments of the present application, first, the current period inspection image is preprocessed. In addition to the conventional radiation calibration, atmospheric correction, and geometric rectification, a multi-scale Retinex-based image enhancement algorithm is used to highlight the texture details and color contrast of the vegetation region, while suppressing background noise, making the boundaries between vegetation and non-vegetation regions clearer. The preprocessed image needs to be verified by a quality evaluation index to ensure that the average gradient of the vegetation region is improved by no less than 15% and the signal-to-noise ratio is no less than 30 dB.

[0053] Secondly, based on a pre-defined vegetation spectral feature library and spatial texture feature library, potential vegetation areas are identified and initially screened from the pre-processed images. The vegetation spectral feature library is a publicly available remote sensing spectral database; the spatial texture feature library is constructed by analyzing vegetation sample areas in historical images of the target power line corridor and statistically analyzing their image texture features.

[0054] The vegetation spectral feature library contains the reflectance threshold ranges of different vegetation types in the red, green, blue, and near-infrared bands. For example, the near-infrared reflectance of healthy vegetation is usually higher than 0.4, and the red reflectance is lower than 0.2. The spatial texture feature library covers statistical features such as texture entropy, contrast, and energy of the vegetation canopy. The texture feature values ​​of local areas of the image are calculated by sliding window and compared with the standard values ​​in the library to achieve preliminary screening of potential vegetation areas. The screening results must meet the requirements that the vegetation area identification accuracy is not less than 90% and the false detection rate is not higher than 5%.

[0055] Subsequently, instance segmentation was performed on potential vegetation areas using the Mask R-CNN deep learning model. Annotated vegetation sample data was used as the training set, and the model parameters were optimized through multiple rounds of iterative training. This enabled the model to accurately separate and identify each independent vegetation unit based on the independence and spatial continuity of the vegetation's geometric contours. During model training, the Intersection over Union (IoU) ratio was used as the evaluation metric, and training was stopped when the average IoU on the validation set reached 0.85 or higher.

[0056] Furthermore, the Mask R-CNN deep learning model is based on an instance segmentation architecture of a region proposal network and a feature pyramid network. Its core consists of a backbone feature extraction network, a region proposal network (RPN), a RoIAlign feature alignment layer, and parallel branches for classification, bounding box regression, and mask generation.

[0057] Finally, for each current independent vegetation unit obtained by segmentation, an inscribed polygon corresponding to the current independent vegetation unit is calculated. The polygon adopts a Douglas-Peucker algorithm to simplify vegetation contour boundary points, retain key turning points, and reduce redundant data. Based on the inscribed polygon, current morphological parameters representing the size and contour of the current independent vegetation unit are extracted. For vegetation height, a difference between DSM data generated by image stereo matching and a historical DTM model is obtained. When calculating, the DSM data needs to be subjected to Gaussian filtering processing. The window size is dynamically adjusted according to the vegetation type. For example, a 5*5 window is used for arbor type, and a 3*3 window is used for shrub type. The crown diameter is determined by the average value of the long axis and the short axis of the minimum circumscribed rectangle of the inscribed polygon. The projection area is the area of the plane region surrounded by the inscribed polygon, which is calculated by a shoelace formula, and the calculation precision is retained to two decimal places. The shoelace formula is an algorithm for calculating the area of a polygon. The principle is to arrange the coordinates of the polygon vertices in order, calculate the area of the trapezoid formed by adjacent vertices in turn, and sum them up. Finally, the absolute value is taken as one-half of the polygon area.

[0058] Specifically, the Douglas-Peucker algorithm is used to simplify the curve or polygon contour. By recursively retaining the points on the curve that are farthest from a preset straight line, until the distance between all points and the simplified curve is less than a set threshold, the data amount is reduced while the basic form of the contour is ensured. In the vegetation contour boundary simplification, first, the contour boundary of the independent vegetation unit is represented as an ordered pixel point sequence. The first and last points of the sequence are used as the initial straight line segment. The perpendicular distance of all intermediate points to the straight line is calculated. The point with the maximum distance is found out and it is judged whether it exceeds the threshold. If it exceeds, the point is retained and used as a new division point. The two curves composed of the first and last points and the division point are repeatedly processed, until the maximum distance of all subsegments is less than the threshold. Finally, the simplified contour polygon composed of key turning points is obtained.

[0059] The potential vegetation region is subjected to instance segmentation. According to the independence and spatial continuity of the geometric contour, each current independent vegetation unit is separated and identified, including:

[0060] The potential vegetation region is subjected to binaryzation processing to obtain an initial segmentation graph.

[0061] The connected domains in the initial segmentation graph are subjected to boundary tracking to obtain the geometric contours of each connected domain.

[0062] Based on a preset contour integrity threshold, the geometric contours with concave points or breakpoints are repaired.

[0063] The areas, perimeters and minimum circumscribed rectangles of the repaired geometric contours are calculated. Based on the difference in the area-perimeter ratio and the non-overlapping nature of the minimum circumscribed rectangle, the independence of the geometric contours is determined.

[0064] Based on the spatial Euclidean distance between the geometric contours, adjacent contours with a distance less than a preset merging threshold are merged, and each of the current independent vegetation units is finally separated and identified.

[0065] In the embodiments of the present application, first, the potential vegetation area is binarized, the Otsu adaptive threshold algorithm is used to determine the optimal segmentation threshold, and the preprocessed image is converted into an initial segmentation image in which the vegetation area is the foreground and the non-vegetation area is the background.

[0066] Secondly, the boundary tracking is performed on the connected domain in the initial segmentation image, the 8-neighborhood seed filling method is used to traverse all the foreground connected domains, the edge contour points of the connected domain are extracted by the Sobel operator, the contour point coordinate sequence is recorded in the clockwise direction, and a closed geometric contour is formed. For the concave points or breakpoints existing in the contour, based on a preset contour integrity threshold, the contour is repaired, for example, the contour integrity threshold is set to 0.8, that is, the coincidence degree of the repaired contour and the complete geometric contour is greater than or equal to 80%. The concave point repair adopts the morphological dilation operation, a 3*3 rectangular structural element is used to dilate the contour for 1-2 times, and the concave area is filled; the breakpoint repair connects the adjacent contour segments by the B-spline curve interpolation method, so that the contour curve is smooth and continuous.

[0067] Subsequently, the area, perimeter and minimum circumscribed rectangle of each repaired geometric contour are calculated, the shape factor is used to distinguish the independent vegetation unit and the adhesion area, that is, the shape factor of the single vegetation unit is usually close to the circular shape, and the ratio is about 0.785, while the adhesion area is significantly reduced due to the irregular contour, and the ratio is usually less than 0.5. At the same time, based on the non-overlapping judgment of the minimum circumscribed rectangle, if the overlapping area of the circumscribed rectangles of two contours is more than 30% of the smaller rectangle area, it is considered as a potential adhesion area.

[0068] Finally, the spatial Euclidean distance between each geometric contour is calculated based on the contour centroid as the reference, for the adjacent contours with a distance less than a preset merging threshold, for example, the preset merging threshold is 2 meters, which is determined based on the statistical distance between the adjacent vegetation. The merging is combined with the spectral feature similarity, for example, when the cosine similarity is greater than 0.85, the merging is performed, so as to avoid missegmenting the branches of the same vegetation unit into multiple independent units; otherwise, for the contours with a distance greater than the threshold and a shape factor meeting the independent unit characteristics, an independent identifier is assigned, and the separation and identification of all current independent vegetation units are finally completed.

[0069] S40: Time sequence alignment and consistency analysis are performed on the current morphological parameter and a historical growth sequence of the corresponding independent vegetation unit in the time sequence remote sensing image, a real-time growth model adapted to the current growth stage is obtained by dynamically adjusting the reference growth model based on the consistency analysis result;

[0070] In the embodiments of the present application, the current morphological parameters are time-series aligned with the historical growth sequence. First, based on the unique identifier of each independent vegetation unit, the morphological parameters and effective accumulated temperature data of the corresponding unit in previous years are retrieved from the historical database to construct a time series matrix.

[0071] The timestamp is accurate to the month, ensuring that the current period matches the historical data at the same period in terms of phenology, and eliminating the interference of seasonal growth differences on model adjustment. The consistency analysis adopts a three-level verification mechanism: the first-level verification calculates the absolute error between the current morphological parameters and the historical prediction values at the same period, and the error rate is |current value-predicted value| / predicted value. If the error rate is less than 10%, it is determined that the model prediction is reliable and no adjustment is needed. The second-level verification is used for the case where the error rate is between 10% and 20%. A sliding window method is used to analyze the error trend, and if the error of two consecutive periods is monotonically increasing or decreasing, the model parameter fine-tuning is triggered. The third-level verification is used for the case where the error rate exceeds 20% or the error is not monotonous. It is determined that the model structure may have a deviation, and the applicability of the baseline growth model needs to be re-evaluated.

[0072] Specifically, in the dynamic adjustment process, the parameters a and b of the Logistic model are updated incrementally. When the error rate is between 10% and 15%, the parameter b is fine-tuned based on the current morphological parameters and the corresponding effective accumulated temperature by using the gradient descent method, and the learning rate is set to 0.01. When the error rate is between 15% and 20%, the parameters a and b are updated simultaneously, and an elastic network regularization term is introduced to suppress overfitting. If the third-level verification is triggered, the model selection process in S10 is re-executed, the R², RMSE and AIC indicators of the Logistic model and the Gompertz model on the new historical data are compared, and the optimal baseline model is re-determined. The adjusted real-time growth model needs to pass the residual test to ensure that the residual sequence meets the normal distribution, i.e., the Shapiro-Wilk test p>0.05 and the autocorrelation is significantly reduced, to ensure the reliability of the prediction of the subsequent growth trend.

[0073] The current morphological parameters are time-series aligned with the historical growth sequence of the corresponding independent vegetation unit in the time-series remote sensing image, and the consistency analysis includes:

[0074] For each current independent vegetation unit, the historical morphological parameter sequence corresponding to the independent vegetation unit is obtained.

[0075] The current morphological parameters are time-series aligned with the multiple historical morphological parameters at the same period in the same growth season in the historical morphological parameter sequence;

[0076] The relative deviation between the current morphological parameters and the average of the multiple historical morphological parameters at the same period is calculated, and it is determined whether the relative deviation exceeds the preset allowed deviation range.

[0077] If the relative deviation exceeds the allowed deviation range, the current morphological parameter is subjected to overall goodness-of-fit test with the historical morphological parameter sequence, and the test result is taken as the consistency analysis result.

[0078] In the embodiments of the present application, first, for each current independent vegetation unit, the corresponding historical morphological parameter sequence is retrieved from the database through its unique identifier, which contains the height, crown width and volume data collected every quarter in the past 3-5 growth cycles, and is accompanied by the phenological period labels corresponding to each collection date, such as the germination period, growth period, mature period, etc.

[0079] Secondly, the current morphological parameter is time-series aligned with multiple historical contemporaneous data in the same phenological period in the historical morphological parameter sequence, for example, if the current collection date is the summer growth period, all morphological parameters of the summer growth period in the historical sequence are selected as comparison samples to reduce the influence of abnormal values in a single year on the analysis result.

[0080] Thirdly, the relative deviation of the current morphological parameter from the average value of the multiple historical contemporaneous morphological parameters is calculated, the formula is (current value-historical average value) / historical average value x 100%, and the preset allowed deviation range is dynamically set according to the vegetation type, ±12% for arbor, ±15% for shrub, and ±20% for herb.

[0081] Further, if the relative deviation does not exceed the allowed range, it is determined that the current growth state conforms to the historical law, and the consistency analysis passes; if it exceeds the range, the overall goodness-of-fit test of the current morphological parameter and the historical morphological parameter sequence is performed using the determination coefficient R² and the root mean square error RMSE, wherein R² needs to be greater than 0.8 and RMSE needs to be less than 1 / 3 of the standard deviation of the historical parameters, and if the test passes, it is determined as normal fluctuation, and if it does not pass, it is marked as consistency anomaly, triggering the dynamic adjustment process of the reference growth model.

[0082] Further, if the relative deviation exceeds the allowed deviation range, the current morphological parameter is subjected to overall goodness-of-fit test with the historical morphological parameter sequence, and the test result is taken as the consistency analysis result, including:

[0083] The current morphological parameter is subjected to synchronicity fitting based on the reference growth model with the historical morphological parameter sequence, to obtain the fitting residual error corresponding to the current growth stage;

[0084] The fitting residual error is compared with the fitting residual error distribution of the historical morphological parameter sequence in the corresponding growth stage, and the standard deviation multiple of the fitting residual error deviating from the mean value of the residual error distribution is calculated;

[0085] If the standard deviation multiple is less than or equal to a preset stability threshold, it is determined that the growth trend is consistent, and the consistency analysis result is that the model can be continued.

[0086] If the standard deviation multiple is greater than the stability threshold, it is determined that the growth trend is inconsistent, and the consistency analysis result is that the model needs to be rebuilt, and the adjustment direction and amplitude of the model parameters are determined based on the fitting residual.

[0087] In the embodiments of the present application, first, the current morphological parameter and the historical morphological parameter sequence are input into the reference growth model for synchronization fitting, taking time as the independent variable and morphological parameter as the dependent variable. The theoretical prediction value of the model in the current growth stage is solved by nonlinear least squares method, and the difference between the current actual value and the prediction value is calculated as the fitting residual.

[0088] Subsequently, the fitting residual data of the historical morphological parameter sequence at the same growth stage is called, the mean μ and the standard deviation σ are calculated, and the normal residual distribution model N(μ,σ²) is constructed. The deviation value Δ = |current residual-μ| of the current fitting residual from μ is calculated, and is converted into a standard deviation multiple k = Δ / σ.

[0089] Exemplarily, the preset stability threshold is set according to the stability grade of vegetation growth, and the evergreen vegetation is 2.0, the deciduous vegetation is 2.5, and the herbaceous vegetation is 3.0. If k≤threshold, it indicates that the current residual is still within the historical normal fluctuation range, and the growth trend has not changed significantly, and the consistency analysis result is that the model can be continued; if k>threshold, it is determined that the growth trend deviates abnormally, and the model needs to be rebuilt, and the parameter adjustment direction is determined by analyzing the residual sign-positive residual indicates that the current growth rate is faster than the historical model, and the growth rate parameter needs to be increased; negative residual needs to decrease the related parameter, and the adjustment amplitude is positively correlated with k value, which is specifically the base parameter value multiplied by (1+k×0.1) or (1-k×0.1), to ensure that the model quickly adapts to the new growth dynamics.

[0090] Specifically, the reference growth model is dynamically adjusted based on the consistency analysis result to obtain a real-time growth model adapted to the current growth stage, comprising:

[0091] When the consistency analysis result is that the model can be continued, the current effective accumulated temperature corresponding to the current collection date is obtained.

[0092] The current morphological parameter and the current effective accumulated temperature are taken as new data points, which are merged with the historical morphological parameter sequence and the corresponding historical effective accumulated temperature data sequence of the independent vegetation unit to obtain an updated morphological parameter sequence and an effective accumulated temperature sequence.

[0093] Based on the updated morphological parameter sequence and effective accumulated temperature sequence, the same fitting method as constructing the reference growth model is adopted to re-perform growth curve fitting to obtain the real-time growth model;

[0094] When the consistency analysis result is that the model needs to be reconstructed, the current effective accumulated temperature corresponding to the current collection date is obtained;

[0095] The current morphological parameter and the current effective accumulated temperature are taken as independent data points, and a new set of model parameters is initialized in combination with the adjustment direction and amplitude of the model parameters;

[0096] Based on the new model parameters, a temporary initial growth model is constructed;

[0097] The historical morphological parameter sequence and the corresponding historical effective accumulated temperature data sequence of the independent vegetation unit are taken as a verification data set to verify the temporary initial growth model;

[0098] If the verification is passed, the temporary initial growth model is taken as the real-time growth model used for prediction in the current inspection period, and the reference growth model of the independent vegetation unit is marked to be updated after subsequent data accumulation;

[0099] If the verification is not passed, the original reference growth model is maintained as the real-time growth model, and a model abnormality reminder is generated.

[0100] In the embodiments of the application, first, when the consistency analysis result is that the model can be continued, the current effective accumulated temperature is collected in real time by the environmental sensor carried by the unmanned aerial vehicle, and is calculated according to the formula ≥10℃ accumulated temperature = Σ (daily mean temperature-10℃) in combination with the daily mean temperature data of the local weather station, wherein the daily mean temperature is greater than 10℃. The current morphological parameter and the current effective accumulated temperature form new data points (T, P), which are merged with (T1, P1), (T2, P2)…(Tn, Pn) of the independent vegetation unit in the historical database to form an updated time sequence data set.

[0101] Secondly, the same Logistic or Gompertz curve fitting method as constructing the reference model is adopted, the Levenberg-Marquardt algorithm is used for nonlinear regression on the merged data, the model parameters are re-solved, and the new curve obtained is the real-time growth model, the expression of which remains the same as that of the reference model, but the parameter value is dynamically optimized with the new data.

[0102] Further, when the consistency analysis result is that the model needs to be rebuilt, in addition to obtaining the current effective accumulated temperature, the spectral reflectance characteristics of the current vegetation, such as the NDVI value, are also extracted as an auxiliary adjustment basis. Based on the parameter adjustment direction and amplitude determined by the foregoing residual analysis, for example, for the Logistic model P(t) = K / (1+exp(a-bt)), if the growth rate parameter b needs to be increased, then the initial new parameter b2 = b1 x(1+k x 0.1), where k is the standard deviation multiple, b2 is the initial new parameter, and b1 is the initial old parameter.

[0103] Specifically, a temporary model is initialized with new parameters, that is, a growth curve model with complete prediction function is quickly formed using the newly initialized parameters. The historical morphological parameter sequence and the corresponding effective accumulated temperature are used as a verification data set, and the verification data set accounts for 30% of the total amount of historical data. The verification passes if the R2, that is, the coefficient of determination, of the model prediction value and the verification data is greater than or equal to 0.85.

[0104] Further, the mean absolute percentage error MAPE of the model prediction value and the historical actual value is calculated. If MAPE≤15%, the verification passes, that is, if the prediction error of the temporary initial growth model for the verification data set is lower than the preset error tolerance, the temporary model is upgraded to the real-time growth model, and the unit reference model is marked in the system to be retrained in the next growth period in combination with more data; if MAPE>15%, that is, if the prediction error is higher than or equal to the error tolerance, the temporary model is abandoned, and the original reference model is used, and an abnormality reminder including the vegetation ID, the current residual, and the parameter adjustment attempt record is pushed to the background management system, prompting the technical personnel to manually review and troubleshoot whether there is a sensor failure, data labeling error, or external factors such as extreme weather influence.

[0105] S50: Obtain prediction effective accumulated temperature data in a future preset warning period, and input the prediction effective accumulated temperature data into each real-time growth model to predict the potential growth state profile of each vegetation unit at the end of the warning period.

[0106] In the embodiment of the application, the prediction effective accumulated temperature data in the future preset warning period is obtained. First, the future 1-3 month weather forecast data published by the meteorological department is called to extract the daily average temperature sequence, and the prediction effective accumulated temperature is obtained by accumulation according to the ≥10℃ accumulated temperature calculation formula.

[0107] The early warning period is dynamically set according to the growth speed of the vegetation, and the fast-growing grass is 1 month, the shrub is 2 months, and the tree is 3 months. The predicted effective accumulated temperature data are input into a real-time growth model, the current morphological parameters are taken as initial values, the predicted height, crown width and volume parameters at the end of the early warning period are obtained through model integral operation, and the predicted parameters are converted into corresponding spatial geometric contours, i.e., potential growth state contours, based on the contour extraction algorithm established in S30. The contours include boundary coordinates, area, perimeter and other spatial attributes, and provide basic data for subsequent safety distance evaluation of the power line.

[0108] Specifically, step S50 in the method includes:

[0109] obtaining the start and end dates of a future preset early warning period, and obtaining daily predicted average temperature data in the future preset early warning period;

[0110] based on the daily predicted average temperature data and a preset growth base temperature, daily calculation and accumulation are performed to obtain the cumulative predicted effective accumulated temperature in the future preset early warning period;

[0111] for the real-time growth model of each vegetation unit, the cumulative predicted effective accumulated temperature is taken as input, and the predicted morphological parameters at the end of the early warning period are calculated through the real-time growth model;

[0112] based on the predicted morphological parameters, the potential growth state contour representing the size and contour of the vegetation unit at the end of the early warning period is generated.

[0113] In the embodiment of the application, first, the future daily predicted average temperature data is obtained from the authoritative meteorological data platform interface according to the start and end dates of the preset early warning period, the data granularity is accurate to 0.1℃, and a confidence label is attached, and the prediction data with a confidence degree greater than or equal to 80% is preferentially selected; if some date data is missing or the confidence degree is insufficient, the time series interpolation method is used to fill in the historical climate characteristics to ensure the continuity of the accumulated temperature calculation.

[0114] Secondly, the growth base temperature is set according to the type of vegetation, wherein the growth base temperature of trees and shrubs is 10℃, the growth base temperature of herbaceous plants is 5℃, and the growth base temperature of special economic crops is configured separately by querying a species characteristic database, for example, the growth base temperature of tea trees is set to 12℃. Based on this, the daily effective accumulated temperature contribution value is calculated. When the daily average temperature is higher than the growth base temperature, the contribution value is the difference between the daily average temperature and the base temperature; when the daily average temperature is lower than or equal to the base temperature, the contribution value is 0, and all daily contribution values in the early warning period are accumulated to obtain the cumulative predicted effective accumulated temperature.

[0115] Exemplarily, if the early warning period is 10 days, the 10-day average temperature is 15℃, 16℃, 14℃, 12℃, 18℃, 17℃, 19℃, 13℃, 15℃, 16℃, respectively, the growth base temperature is 10℃, then the daily contribution value is 5℃, 6℃, 4℃, 2℃, 8℃, 7℃, 9℃, 3℃, 5℃, 6℃, respectively, and the cumulative effective accumulated temperature for 10 days is 5+6+4+2+8+7+9+3+5+6=55℃.

[0116] Again, for the real-time growth model of each vegetation unit, the above cumulative predicted effective accumulated temperature is substituted into the model equation, such as the Gompertz model P(t)=K×exp(-exp(r×(t0-t))), where K is an environmental capacity parameter, r is a growth rate parameter, and t0 is a growth inflection point time. Through the parameter mapping relationship in the model, the predicted height, crown diameter, and three-dimensional volume and other morphological parameters at the end of the early warning period are calculated.

[0117] Subsequently, a contour generation engine is called to take the spatial coordinates of the current vegetation unit as the origin, determine the vertical dimension parameter according to the predicted height, construct an elliptical or circular boundary in the horizontal direction in combination with the crown radius, generate a three-dimensional mesh model containing multiple layers of details through a triangulation algorithm, and finally output the potential growth state contour containing geometric size data and growth trend vectors, such as height growth rate and crown expansion direction, as dynamic attributes, providing multi-dimensional reference basis for subsequent safety distance assessment.

[0118] Among them, based on the predicted morphological parameters, the potential growth state contour representing the size and contour of the vegetation unit at the end of the early warning period is generated, including:

[0119] According to the height and crown width prediction values in the predicted morphological parameters, a geometric outward expansion model is constructed with the current position of the current independent vegetation unit as the bottom edge center;

[0120] The edges of the circumscribed polygon are proportionally expanded to match the crown width prediction value in the predicted morphological parameters, to generate the planar projection contour at the end of the early warning period;

[0121] In combination with the height prediction value in the predicted morphological parameters, the planar projection contour is stretched along the vertical direction to generate a three-dimensional envelope as the potential growth state contour, wherein the generation of the three-dimensional envelope is constrained according to the typical height and crown width ratio relationship of the independent vegetation unit analyzed from the historical morphological parameter sequence.

[0122] In the embodiments of the present application, first, the center position of the bottom edge is determined based on the current geographic coordinates of the independent vegetation unit as the reference point. According to the predicted crown width parameter, an initial contour in the horizontal direction is constructed by using an elliptical fitting method. For trees and shrubs, the crown width usually presents an approximate elliptical distribution, and the long axis direction is determined according to the main direction of vegetation expansion in the historical image. If there is no significant directional feature in the historical data, a circular contour is used by default. For herbaceous vegetation, a circular planar model is used due to the growth density, and the radius is 1 / 2 of the predicted crown width.

[0123] Secondly, the constructed initial contour boundary is then expanded outward by a preset safety redundancy. The redundancy is dynamically adjusted according to the vegetation type, and the redundancy for trees is 5%, for shrubs is 8%, and for herbaceous is 10%, to cope with possible errors in model prediction. The vertex coordinates of the expanded polygon are calculated by the polar coordinate conversion method to ensure smooth and continuous contour boundaries.

[0124] Subsequently, in the vertical dimension, the side slope of the three-dimensional envelope is constrained by using the predicted height as the reference and combining the height-crown width ratio coefficient extracted from the historical morphological parameter sequence of the vegetation unit. For example, the typical high crown ratio of trees is 2.5-3.0, and the high crown ratio of shrubs is 1.0-1.5, to avoid the appearance of abnormal contours that do not conform to biological characteristics.

[0125] For example, when the predicted height is 5 meters and the crown width is 2 meters, if the average historical high crown ratio of the tree is 2.8, the side slope should be controlled between 5 / 2=2.5 and 3.0, and the proportion constraint is realized by adjusting the inclination angle of the contour side. The final generated three-dimensional envelope includes vertex coordinate set, triangle patch index, normal vector, and other three-dimensional model basic data, as well as statistical parameters such as contour surface area, volume, and height quantile.

[0126] S60: Based on the three-dimensional space model and the potential growth state contour, the minimum predicted spatial distance between each vegetation unit and the nearby conductor is calculated. If the minimum predicted spatial distance is less than a preset safety distance threshold, an encroachment risk warning information for the corresponding vegetation unit is generated.

[0127] In the embodiments of the present application, the three-dimensional space model includes three-dimensional coordinate data of the power line, including sag of the conductor, position of the tower, and other information, and the potential growth state contour provides the spatial geometric shape of the vegetation at the end of the warning period.

[0128] When calculating the minimum predicted spatial distance, first, the three-dimensional model of the power line and the potential growth state contour of the vegetation are imported into the same spatial coordinate system, and a spatial distance calculation algorithm is used to traverse the distance between all vertices on the surface of the vegetation contour and the conductor, and find the minimum value.

[0129] For the conductor, the sag variation of different points is considered, the three-dimensional coordinates of each point are calculated through the conductor mechanics model, and the accuracy of distance calculation is ensured. The preset safety distance threshold is set comprehensively according to the voltage grade of the power line, the type of tower and the local meteorological conditions, for example, the safety distance threshold of 110kV line is 5 meters, the safety distance threshold of 220kV line is 6 meters, etc.

[0130] Further, if the calculated minimum predicted spatial distance is less than the set safety distance threshold, a risk of invasion warning information is triggered, which contains the specific position of the vegetation unit, the predicted invasion time, the current distance from the conductor and the potential risk level, etc. so that the inspection personnel can take measures such as pruning and cutting in time to ensure the safe operation of the power line.

[0131] Specifically, step S60 in the method comprises:

[0132] extracting a three-dimensional coordinate sequence of the conductor segment adjacent to the potential growth state contour space position from the three-dimensional space model;

[0133] discretely sampling the three-dimensional envelope surface of the potential growth state contour to obtain three-dimensional coordinates of a plurality of surface sampling points, wherein the number of surface sampling points is determined according to the crown diameter;

[0134] respectively calculating the Euclidean distance from each surface sampling point to all segments in the three-dimensional coordinate sequence of the conductor segment, and taking the minimum value as the minimum distance from the current sampling point to the conductor;

[0135] traversing all surface sampling points, and taking the global minimum value in all minimum distances as the minimum predicted spatial distance of the current vegetation unit from the nearby conductor;

[0136] comparing the minimum predicted spatial distance with a preset safety distance threshold, and if it is less than the safety distance threshold, determining that there is a risk of invasion, and generating a warning information containing the identification of the corresponding vegetation unit, the minimum predicted spatial distance and the time at the end of the warning period.

[0137] In the embodiment of the application, first, based on the topological relationship of the three-dimensional space model, the conductor segment within the preset search radius from the vegetation potential growth state contour is quickly located through the space indexing technology. The search radius is dynamically set according to the voltage grade, 30 meters for 500kV line, 20 meters for 220kV line, and 15 meters for 110kV and below, to reduce the amount of invalid calculation. When extracting the conductor segment, the three-dimensional coordinates of key nodes such as tower hanging points and midpoints of tension segments need to be included, and the coordinates are arranged in sequence to form a continuous coordinate sequence according to the line direction.

[0138] Secondly, the surface of the three-dimensional envelope of the vegetation potential growth state profile is discretely sampled. The number of sampling points is proportional to the crown diameter. The number of surface sampling points is determined according to the crown diameter, which corresponds to one of the morphological parameters described above. The number of surface sampling points is linearly set according to the crown diameter, and the sampling density is 5-8 / m. The calculation formula is: the number of sampling points = ceil(crown diameter x 6), wherein ceil is the ceiling function.

[0139] Exemplarily, when the crown diameter is 5 meters, the number of sampling points is 30; when the crown diameter is 10 meters, the number of sampling points is 60; and when the crown diameter is ≤1 meter, the number of sampling points is ≥10, ensuring the minimum sampling accuracy.

[0140] Thirdly, the minimum distance from each surface sampling point to the traverse line segment is calculated. For a single sampling point (Xp, Yp, Zp) and a traverse line segment AB (A point coordinates (Xa, Ya, Za), B point coordinates (Xb, Yb, Zb)), the shortest distance from the space point to the line segment is solved by the vector projection method. First, calculate the vector AP=(Xp-Xa, Yp-Ya, Zp-Za) and the vector AB=(Xb-Xa, Yb-Ya, Zb-Za), and then calculate the projection parameter t=(AP·AB) / (AB·AB). If t≤0, the shortest distance is the modulus of AP; if t≥1, the shortest distance is the modulus of BP; otherwise, the shortest distance is the modulus of AP-t·AB. All line segments in the traverse line segment sequence are traversed, and the minimum value of the calculation result is taken as the minimum distance of the current sampling point.

[0141] Then, the global minimum value is selected from the minimum distances of all sampling points as the minimum predicted spatial distance between the vegetation unit and the traverse. To improve the calculation efficiency, the divide-and-conquer strategy is used to group and calculate the sampling points in parallel, each group containing 10-20 sampling points. After the calculation in the group is completed, the local minimum values are summarized to determine the global minimum value. At the same time, the sampling point coordinates and the traverse line segment number corresponding to the minimum distance are recorded to provide a position reference for subsequent risk analysis.

[0142] Finally, the minimum predicted spatial distance is compared with the preset safety distance threshold. The safety distance threshold is determined according to the “110kV-750kV Overhead Transmission Line Design Specification” (GB50545), and is modified considering voltage level, meteorological conditions such as ice thickness, wind speed and terrain factors.

[0143] Exemplarily, the basic safety distance of 110kV line in plain areas is 4 meters, and the modified value is 4.4 meters in mountainous areas, and the modified value is 4.62 meters when the ice thickness exceeds 10mm. If the minimum predicted spatial distance is less than the modified threshold, a warning information is generated, including the vegetation unit ID, the line name, the time at the end of the warning period, the current distance value, the threshold and the risk level.

[0144] Further, the risk level is divided into slight / medium / severe, and is divided according to the proportion of the distance from the threshold value: <5% is slight, 5%-15% is medium, and >15% is severe.

[0145] Illustratively, the early warning information is pushed to the inspection management platform through an encryption protocol, and a short message is sent to the responsible operation and maintenance personnel, and the content format is: "[Invasion limit early warning] Line XX-XX# tower interval, vegetation ID: XXX, predicted distance 3.2 meters < threshold value 4.0 meters, risk level: medium, please check and handle before X month X day".

[0146] In summary, compared with the prior art, the present application combines the real-time morphological data of vegetation obtained by unmanned aerial vehicle inspection with the effective accumulated temperature of meteorological prediction, constructs a dynamic growth model to predict the potential growth profile of vegetation, and realizes risk early warning through three-dimensional space distance calculation, effectively solving the problems of insufficient prediction of vegetation growth trend and lag of safety distance evaluation in traditional inspection. This method improves the foresight and accuracy of vegetation risk control of power line channel through multi-source data fusion and fine modeling, provides a basis for intelligent inspection decision-making, especially in complex terrain and climate conditions, can reduce the cost of manual inspection, shorten the risk response time, and ensure the safe and stable operation of the power transmission line.

[0147] In summary, the embodiments of the present application have at least the following technical effects:

[0148] The embodiment of the application provides a power line inspection method based on a UAV, first, time sequence remote sensing images of a power line corridor in the past multiple growth periods are acquired, historical accumulated temperature data is combined, a reference growth model reflecting the variation law of morphological parameters of each independent vegetation unit with effective accumulated temperature is constructed, and historical trend modeling of vegetation growth characteristics is realized. Secondly, the current period inspection image is collected by using the optical load of the UAV, the morphological parameters of each current independent vegetation unit are extracted through preprocessing, feature recognition, instance segmentation and other steps, and data support is provided for growth state analysis. Thirdly, the current morphological parameters and the historical growth sequence are time sequence aligned, consistency analysis is completed through relative deviation calculation and goodness of fit test, the reference growth model is dynamically adjusted to adapt to the current growth stage, and the timeliness and accuracy of model prediction are ensured. On this basis, the predicted effective accumulated temperature of the future early warning period is acquired, the potential growth state profile of the vegetation is obtained by inputting the real-time growth model, the three-dimensional space model of the power line is combined, the minimum predicted spatial distance between the vegetation and the conductor is calculated through discretization sampling, the invasion limit risk warning is generated after comparison with the safety threshold, and the prospective evaluation and early warning of the vegetation growth risk of the power line corridor are realized. Through the above technical scheme, the dynamic prediction model of the application fuses the historical growth law and real-time monitoring data, combines the spatial layout information of the power line, provides a basis for the power line operation and maintenance department to formulate a differentiated obstacle removal plan, effectively reduces the line failure probability caused by vegetation growth, and improves the safety and economy of the power grid operation.

[0149] It should be noted that the above sequence of the embodiments of the application is only for description, and does not represent the advantages and disadvantages of the embodiments. And the above describes specific embodiments of the present application. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are possible or can be advantageous.

[0150] The above only describes the preferred embodiments of the application, and does not limit the application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the application shall be included in the protection scope of the application.

[0151] The present application and the drawings are only exemplary descriptions of the application, and are considered to cover any and all modifications, changes, combinations or equivalents within the scope of the application. Obviously, those skilled in the art can make various modifications and changes to the application without departing from the scope of the application. Thus, if these modifications and changes of the application belong to the scope of the application and its equivalents, the application intends to include these modifications and changes.

Claims

1. A method for unmanned aerial vehicle based power line inspection, the method comprising: The method comprises: acquiring time-series remote sensing images of a target power line corridor in multiple historical growth periods, and constructing a baseline growth model of each independent vegetation unit in the corridor based on the time-series remote sensing images; acquiring a three-dimensional space model of the target power line corridor, wherein the three-dimensional space model at least contains spatial position information of conductors and towers; collecting current period inspection images of the target power line corridor by an optical load carried by a UAV, and identifying and segmenting each current independent vegetation unit from the current period inspection images to acquire current morphological parameters of each current independent vegetation unit; performing time-series alignment and consistency analysis on the current morphological parameters and historical growth sequences of the corresponding independent vegetation unit in the time-series remote sensing images, and dynamically adjusting the baseline growth model based on the consistency analysis result to obtain a real-time growth model adapted to the current growth stage; wherein the time-series alignment and consistency analysis on the current morphological parameters and the historical growth sequences of the corresponding independent vegetation unit in the time-series remote sensing images comprises: for each current independent vegetation unit, acquiring a historical morphological parameter sequence corresponding to the independent vegetation unit; performing time-series alignment on the current morphological parameters and a plurality of historical contemporaneous morphological parameters in the historical morphological parameter sequence which are in the same growth season as the current acquisition date; calculating the relative deviation between the current morphological parameters and the average of the plurality of historical contemporaneous morphological parameters, and determining whether the relative deviation exceeds a preset allowable deviation range; if the relative deviation exceeds the allowable deviation range, performing overall goodness-of-fit test on the current morphological parameters and the historical morphological parameter sequence, and taking the test result as the consistency analysis result, which comprises: performing synchronous fitting of the current morphological parameters and the historical morphological parameter sequence based on the baseline growth model to obtain fitting residuals corresponding to the current growth stage; comparing the fitting residuals with fitting residual distribution of the historical morphological parameter sequence in the corresponding growth stage, and calculating the standard deviation multiple of the fitting residuals deviating from the mean of the residual distribution; if the standard deviation multiple is less than or equal to a preset stability threshold, it is determined that the growth trend is consistent, and the consistency analysis result is that the model can be continued; if the standard deviation multiple is greater than the stability threshold, it is determined that the growth trend is inconsistent, and the consistency analysis result is that the model needs to be rebuilt, and the adjustment direction and amplitude of the model parameters are determined based on the fitting residuals; acquiring predicted effective accumulated temperature data in a preset warning period in the future, and inputting the predicted effective accumulated temperature data into each real-time growth model to predict the potential growth state profile of each vegetation unit at the end of the warning period; based on the three-dimensional space model and the potential growth state profile, calculating the minimum predicted spatial distance between each vegetation unit and the nearby conductor, and if the minimum predicted spatial distance is less than a preset safety distance threshold, generating an encroachment risk warning information for the corresponding vegetation unit. 2.The unmanned aerial vehicle based power line inspection method of claim 1, wherein, The construction step of the baseline growth model comprises: extracting a historical morphological parameter sequence of each independent vegetation unit in the time-series remote sensing images in multiple historical growth periods; acquire a historical accumulated temperature data sequence corresponding to the plurality of historical growth cycles; for each independent vegetation unit, based on the historical morphological parameter sequence of the independent vegetation unit and the corresponding historical accumulated temperature data sequence, a growth curve model representing the change of the morphological parameter of the independent vegetation unit with accumulated temperature is fitted to obtain a reference growth model of the independent vegetation unit. 3.The UAV-based power line inspection method of claim 1, wherein, Collect the current period inspection image of the target power line corridor by the optical load carried by the unmanned aerial vehicle, and identify and segment each current independent vegetation unit from the current period inspection image to obtain the current morphological parameter of each current independent vegetation unit, including: Preprocessing the current period inspection image to enhance the contrast features of the vegetation area and the background; Based on the preset vegetation spectral feature library and spatial texture feature library, identify and preliminarily screen the potential vegetation area from the preprocessed image; Instance segmentation is performed on the potential vegetation area, and each current independent vegetation unit is separated and identified according to the independence and spatial continuity of the geometric contour. For each segmented current independent vegetation unit, calculate the corresponding circumscribed polygon, and extract the current morphological parameter representing the size and contour of the current independent vegetation unit based on the circumscribed polygon, wherein the current morphological parameter includes at least one of vegetation height, crown diameter, and projected area. 4.The UAV-based power line inspection method of claim 3, wherein, Instance segmentation is performed on the potential vegetation area, and each current independent vegetation unit is separated and identified according to the independence and spatial continuity of the geometric contour. Binaryzation is performed on the potential vegetation area to obtain an initial segmentation map; Boundary tracking is performed on the connected domains in the initial segmentation map to obtain the geometric contours of each connected domain; Based on the preset contour integrity threshold, repair the geometric contours with concave points or breakpoints; Calculate the area, perimeter and minimum circumscribed rectangle of each repaired geometric contour, and determine the independence of the geometric contour based on the difference in the area-perimeter ratio and the non-overlapping of the minimum circumscribed rectangle. Based on the spatial Euclidean distance between the geometric contours, merge adjacent contours with a distance less than a preset merging threshold to finally separate and identify each current independent vegetation unit.

5. The unmanned aerial vehicle based power line inspection method of claim 1, wherein, Based on the consistency analysis result, dynamically adjust the reference growth model to obtain a real-time growth model adapted to the current growth stage, including: When the consistency analysis result is that the model can be continued, acquire the current accumulated temperature corresponding to the current collection date; Take the current morphological parameter and the current accumulated temperature as new data points, combine them with the historical morphological parameter sequence of the independent vegetation unit and the corresponding historical accumulated temperature data sequence to obtain an updated morphological parameter sequence and an accumulated temperature sequence; Based on the updated morphological parameter sequence and accumulated temperature sequence, use the same fitting method as constructing the reference growth model to re-fit the growth curve to obtain the real-time growth model; When the consistency analysis result is that the model needs to be reconstructed, acquire the current accumulated temperature corresponding to the current collection date; The current morphological parameters and the current effective accumulated temperature are combined as independent data points, and the adjustment direction and amplitude of the model parameters are initialized to obtain a new set of model parameters; Based on the new model parameters, a temporary initial growth model is constructed; The historical morphological parameter sequence and the corresponding historical effective accumulated temperature data sequence of the independent vegetation unit are used as a verification data set to verify the temporary initial growth model; If the verification is passed, the temporary initial growth model is used as the real-time growth model for prediction in the current inspection period, and the reference growth model of the independent vegetation unit is marked for updating after subsequent data accumulation; If the verification is not passed, the original reference growth model is maintained as the real-time growth model, and a model abnormality warning is generated.

6. The unmanned aerial vehicle based power line inspection method of claim 1, wherein, Predicted effective accumulated temperature data in a future preset warning period is obtained, and the predicted effective accumulated temperature data is input into each real-time growth model to predict the potential growth state profile of each vegetation unit at the end of the warning period, including: The start and end dates of the future preset warning period are obtained, and daily predicted average temperature data in the future preset warning period is obtained; Based on the daily predicted average temperature data and a preset growth base point temperature, the cumulative predicted effective accumulated temperature in the future preset warning period is calculated and accumulated daily; For the real-time growth model of each vegetation unit, the cumulative predicted effective accumulated temperature is input to calculate the predicted morphological parameters at the end of the warning period through the real-time growth model; Based on the predicted morphological parameters, the potential growth state profile representing the size and profile of the vegetation unit at the end of the warning period is generated.

7. The unmanned aerial vehicle based power line inspection method of claim 6, wherein, Based on the predicted morphological parameters, the potential growth state profile representing the size and profile of the vegetation unit at the end of the warning period is generated, including: According to the height and crown width prediction values in the predicted morphological parameters, a geometric expansion model is constructed with the current position of the independent vegetation unit as the bottom center; The edges of the circumscribed polygon are expanded in proportion to match the crown width prediction value in the predicted morphological parameters to generate a planar projection profile at the end of the warning period; Combined with the height prediction value in the predicted morphological parameters, the planar projection profile is stretched in the vertical direction to generate a three-dimensional envelope as the potential growth state profile, wherein the generation of the three-dimensional envelope is constrained according to the typical height and crown width ratio relationship of the independent vegetation unit analyzed from the historical morphological parameter sequence. 8.The UAV-based power line inspection method of claim 1, wherein, Based on the three-dimensional space model and the potential growth state profile, the minimum predicted spatial distance between each vegetation unit and the nearby guide line is calculated, and if the minimum predicted spatial distance is less than a preset safety distance threshold, an encroachment risk warning information for the corresponding vegetation unit is generated, including: The three-dimensional coordinate sequence of the guide line segment adjacent to the spatial position of the potential growth state profile is extracted from the three-dimensional space model; The surface of the three-dimensional envelope of the potential growth state profile is discretely sampled to obtain the three-dimensional coordinates of a plurality of surface sampling points, wherein the number of surface sampling points is determined according to the crown width diameter; Calculate the Euclidean distance from each of the surface sampling points to all line segments in the three-dimensional coordinate sequence of the conductor line segment, and take the minimum value as the minimum distance between the current sampling point and the conductor line; Traverse all surface sampling points, and take the global minimum value in all minimum distances as the minimum predicted spatial distance between the current vegetation unit and the nearby conductor line; Compare the minimum predicted spatial distance with the preset safety distance threshold value, and if it is smaller, determine that there is an encroachment risk, and generate warning information containing the corresponding vegetation unit identifier, minimum predicted spatial distance, and warning period end time.

Images