A winter power distribution area snow cover state detection method based on unmanned aerial vehicle inspection

Abstract

The application belongs to the technical field of power transmission line channel operation and maintenance, and particularly relates to a winter power distribution area snow cover state detection method based on unmanned aerial vehicle inspection. Test training data collection and classification database establishment; test training data preprocessing, extraction of snow cover indexes in rectangular power supply area images, extraction based on image internal edge density characteristic indexes; snow cover index and snow cover state association matching, snow cover state recognition model training based on a support vector machine, and snow cover state recognition based on the snow cover state recognition model. The winter power distribution area snow cover state detection method based on unmanned aerial vehicle inspection of the application analyzes image data of different regions under different snow cover states, constructs snow cover index characteristics by using color and contour elements in the image data, constructs the influence weight by associating and matching the snow cover index and the snow cover state, optimizes the input of the support vector machine, and further improves the accuracy of snow cover state recognition.

Description

Technical Field

[0001] This application belongs to the field of power transmission line channel operation and maintenance technology, and in particular relates to a method for detecting snow cover status in winter power distribution areas based on drone inspection. Background Technology

[0002] Winter snow cover poses challenges to the operation and maintenance of power grids, especially given the increasing frequency of extreme weather events in recent years, which have significantly impacted people's normal work, life, and studies. Since snow cover often generates associated hazards such as icing as temperatures change, it is crucial to promptly analyze and assess the snow cover status within the power grid area after widespread snowfall or rain in winter. This allows for proactive risk assessment of potential snow disasters and subsequent maintenance tasks. With the development of drone control technology, winter snow cover inspections are increasingly being replaced by drone-based operations. Therefore, the efficiency of using drone inspection data to quickly and effectively analyze and assess the snow cover status of distribution areas directly impacts the efficiency of related tasks. Summary of the Invention

[0003] The purpose of this application is to provide a method for detecting snow cover status in winter power distribution areas based on UAV inspection images as raw data, which has good status recognition capabilities, low data processing difficulty, and is easy to apply.

[0004] To achieve the above objectives, this application adopts the following technical solution.

[0005] A method for detecting snow cover status in winter power distribution areas based on drone inspection includes the following steps:

[0006] Step 1: Data collection for testing and training, and establishment of a classification database, specifically including:

[0007] Collect relevant snow-covered image data from existing data as raw data, classify them according to different landforms and different zoning functions, and establish test and training databases corresponding to each type.

[0008] Step 2: Preprocessing of test training data, specifically including:

[0009] 2.1 Power supply area interception

[0010] Specifically, this means: acquiring source data images, drawing circular power supply area images with the center of each power distribution area in the image as the center and the distance between the user farthest from the power distribution station and the power distribution station as the radius, drawing rectangular power supply area images with the circular power supply area images as the circumcircle, so that at least one vertex of the rectangular power supply area image is located at the farthest power user.

[0011] 2.2 Pixel Adjustment

[0012] Specifically, this means: acquiring all rectangular power supply area images, and adjusting all rectangular power supply area images to a uniform pixel height or width while maintaining the original aspect ratio;

[0013] 2.3 Status Labeling

[0014] Specifically, this refers to using existing image annotation programs and either manual or machine recognition methods to classify and annotate all rectangular power supply area images according to the snow cover index F(H,S,T); the snow cover state w is divided into: low snow cover, moderate snow cover, high snow cover, and extremely high snow cover; among which... in This refers to the historical snow cover thickness since the current power supply area was formed; This refers to the historical ambient temperature since the current power supply area was formed;

[0015] Step 3: Extraction of snow cover indicators from the images of each rectangular power supply area, specifically including:

[0016] 3.1 Extraction of GSV, a dual-channel feature index based on the HSV color space

[0017] Specifically, it refers to: using the range of channel values ​​as the horizontal axis and dividing it into R intervals, drawing a histogram based on the number of pixels in the corresponding value interval, and establishing a dual-channel feature index GSV based on the HSV color space;

[0018] in Q S (x) represents the number of pixels in the S channel with a saturation value equal to x. This represents the number of pixels in channel S distributed within the interval [0, i]. Q is the total number of pixels in the S channel; where Q V (x) represents the number of pixels in the V channel with a brightness value equal to x. Let V be the number of pixels distributed in the interval [j, R]. is the total number of pixels in the V channel, where i and j are threshold parameters that control the evaluation of the dual-channel feature index GSV;

[0019] 3.2 Extraction of Image Intrinsic Edge Density Features

[0020] Specifically, this refers to: based on the secondary establishment of the image's internal edge density feature index EDGE, and... Q edge Q represents the number of pixels in the rectangular power supply area image that can detect edges; sum The total number of pixels in the rectangular power supply area image; where, in each algorithm, an edge determination threshold parameter k is involved to determine whether a pixel belongs to an edge;

[0021] Step 4: Matching snow cover indicators with snow cover status, specifically including:

[0022] Analysis of variance (ANOVA) is used to evaluate the influence of snow cover index attributes on snow cover status. For each snow cover index, its influence on snow cover status can be expressed as a correlation coefficient. in It is the sum of squares of the deviations between states; w represents the state classification; n r The number of rectangular power supply area images belonging to state w; Let be the mean of the feature indexes of the rectangular power supply area image belonging to state w. x represents the mean of the feature indices for all rectangular power supply area images. iw Let be the feature index value of the i-th sample within the rectangular power supply area image group belonging to state w; the specific calculation process is as follows:

[0023] 4.1 Obtain all rectangular power supply area images in each category database. Under different values ​​of threshold parameters i, j and edge determination threshold parameter k, calculate the dual-channel feature index GSV and the internal edge density feature index EDGE respectively.

[0024] 4.2 Calculate the influence (μ) of the dual-channel characteristic index GSV and the internal edge density characteristic index EDGE on the snow cover state under the aforementioned different value conditions. GSV and μ EDGE ;

[0025] 4.3 Based on empirical judgment, μ is selected. GSV and μ EDGE Values ​​with a correlation coefficient of 0.15 or higher are used as the criteria for judging relevance, and values ​​with a correlation coefficient of less than 0.15 are deleted.

[0026] 4.4 Based on actual computing resources and accuracy requirements, select effective values ​​according to the correlation coefficient from high to low.

[0027] 4.5 Based on the aforementioned valid value scenarios, several dual-channel feature indices GSV and internal edge density feature indices EDGE are obtained respectively;

[0028] Step 5: Training the snow-covered state recognition model based on support vector machines, specifically including:

[0029] After obtaining the snow cover feature index based on the aforementioned step four, use it as the input of the support vector machine and the snow cover state as the output to train the support vector machine and obtain the snow cover state recognition model.

[0030] Step 6: Identify the snow cover state based on the snow cover state recognition model, specifically including:

[0031] Based on the snow cover status recognition model obtained in step five above, the inspection image to be analyzed is used as input to obtain the snow cover status recognition result.

[0032] To further improve and supplement the aforementioned method for detecting snow cover status in winter power distribution areas based on UAV inspection, in step two, in order to improve the accuracy of identification, the source data should be annotated multiple times, and based on the results of each round of hierarchical annotation of the rectangular power supply area images, the images of each rectangular power supply area should be finally annotated according to the probability of the hierarchical results appearing at the highest level.

[0033] To further improve and supplement the aforementioned method for detecting snow cover status in winter power distribution areas based on drone inspection, in step two, under the premise of calculating the snow cover index F(H,S,T), the snow cover status corresponding to different snow cover indices in the historical data of the current power supply area is empirically determined, and specific snow cover index values ​​corresponding to low snow cover, medium snow cover, high snow cover, and extremely high snow cover are specified for different power supply areas.

[0034] To further improve and supplement the aforementioned method for detecting snow cover status in winter power distribution areas based on drone inspection, in step one, the terrain type includes at least mountainous areas, plains, and hills; and the functional zoning type includes at least rural areas, urban areas, industrial areas, and agricultural areas.

[0035] Its beneficial effects are as follows:

[0036] The method for detecting snow cover status in power distribution areas during winter based on UAV inspection in this application analyzes image data of different areas under different snow cover statuses, constructs snow cover index features using elements such as color and contour in the image data, constructs the influence weights by matching the association between snow cover indexes and snow cover status, optimizes the input of support vector machine, and uses a mature support vector machine training model for training and classification, thereby improving the accuracy of snow cover status recognition. Attached Figure Description

[0037] Figure 1 This is a schematic diagram of the power supply area interception method. Detailed Implementation

[0038] The present application will be described in detail below with reference to specific embodiments.

[0039] The method for detecting snow cover in power distribution areas during winter based on UAV inspection in this application mainly utilizes UAV inspection image recognition combined with support vector machine training model and feature extraction scheme to identify and detect the snow cover status of power distribution areas in winter. This method is used to quickly analyze and identify the snow cover status of power distribution areas in winter snowy weather, so as to provide support for power grid maintenance and emergency repair operations and fault prediction.

[0040] It mainly includes the following steps

[0041] Step 1: Data Collection for Testing and Training and Establishment of Classification Database

[0042] Before implementation, this application needs to collect sufficient training data to extract the necessary features of the snow-covered power distribution area in winter so as to enable automatic identification and classification. Due to the vast land area of ​​China, the regional topography and weather conditions vary greatly, and the snowfall and snowfall distribution characteristics in winter are different. This will lead to significant differences in the features contained in the images of snow-covered areas in different regions. In order to complete the monitoring and data processing more accurately and efficiently, the data should be classified according to different landforms and different zoning functions. Then, the corresponding snow-covered image data should be collected from the existing data as the raw data, and test training databases corresponding to each type should be established.

[0043] Based on the actual situation, the landform types include at least mountainous areas, plains, and hills; the zoning functional types include at least rural areas, urban areas, industrial areas, and agricultural areas.

[0044] Step 1: Preprocessing of test and training data

[0045] A power distribution area can be viewed as a power supply network composed of multiple substations. Each substation is independent, and its power supply range radiates outward from the substation as the core. The substations are combined to form a whole. Due to the influence of supply and demand, the power supply range is generally limited to living and working areas. Therefore, the corresponding power supply areas are arranged in patches or strips. When performing snow cover detection, only the data within the corresponding power supply area needs to be considered. Therefore, after obtaining the aforementioned test training databases of various types, appropriate preprocessing can be performed, namely, background segmentation according to the characteristics of the corresponding area to remove irrelevant elements from the original image and improve the efficiency of subsequent data processing. Specifically, this refers to:

[0046] 2.1 Power supply area interception

[0047] like Figure 1 As shown, the source data image is obtained, and a circular power supply area image is drawn with the center of each power distribution area in the image as the center and the distance between the user farthest from the power distribution station and the power distribution station as the radius. A rectangular power supply area image is drawn with the circular power supply area image as the circumcircle, so that at least one vertex of the rectangular power supply area image is located at the farthest power user.

[0048] 2.2 Pixel Adjustment

[0049] Acquire all rectangular power supply area images, and adjust all rectangular power supply area images to a uniform pixel height or width while maintaining the original aspect ratio;

[0050] 2.3 Status Labeling

[0051] After the aforementioned processing, the images in various training databases are unified into rectangular power supply area images with the same pixel height or pixel width. To obtain the corresponding features for different snow cover levels, all rectangular power supply area images need to be hierarchically labeled; specifically:

[0052] Using existing image annotation programs, all rectangular power supply area images are classified and labeled according to the snow cover index F(H,S,T) using manual or machine recognition methods; the snow cover status is classified as: low snow cover, moderate snow cover, high snow cover, and extremely high snow cover.

[0053] The snow cover index is used to characterize the impact of the current snow cover status on the normal power supply of the accessory area. The higher the index value, the greater the adverse impact on the normal power supply of the distribution area. Considering the comprehensive impact on the stability of the distribution structure, maintenance difficulty, and power load during the actual operation of the distribution network, this application constructs a snow cover index F(H,S,T) based on the current snow cover thickness H, snow cover area ratio S, and ambient temperature T. The calculation method is as follows:

[0054] in This refers to the historical snow cover thickness since the current power supply area was formed; This refers to the historical ambient temperature since the current power supply area was formed;

[0055] Because of the differences in infrastructure and geographical conditions in different power supply areas, the snow resistance and disaster control capabilities of different power supply areas under the same snow cover conditions are not consistent. Therefore, based on the calculated snow cover index F(H,S,T), the specific snow cover index values ​​corresponding to low snow cover, moderate snow cover, high snow cover and extremely high snow cover are specified for different power supply areas according to empirical judgment of the snow cover state corresponding to different snow cover in historical data of the current power supply area.

[0056] In particular, to improve the accuracy of recognition, the source data should be annotated in multiple rounds, and based on the results of each round of hierarchical annotation of the rectangular power supply area image, the final hierarchical annotation of each rectangular power supply area image should be performed according to the probability of the hierarchical result appearing.

[0057] Step 2: Extraction of snow cover indicators from images of each rectangular power supply area

[0058] 3.1 Extraction of GSV, a dual-channel feature index based on the HSV color space

[0059] Due to the differences in visible light reflection and scattering between snow-covered and non-snow-covered areas, snow-covered areas are generally snow-white, and the higher the snow thickness, the more obvious the color. Non-snow-covered areas are other dark colors, and the thinner the snow thickness, the more obvious the color. The color attributes in the HSV color space of the two images are significantly different. Through analysis, it was found that the pixel values ​​of the S channel (saturation) and V channel (brightness) in the HSV color space of the rectangular power supply area image under non-snow-covered state are significantly different. Using the range of channel values ​​as the horizontal axis and dividing it into R intervals, a histogram was drawn with the number of pixels in the corresponding value interval. It can be found that the distribution of high and low values ​​of the histograms of the two channels in the snow-covered and non-snow-covered areas shows opposite trends. Therefore, this application first establishes a dual-channel feature index GSV based on the HSV color space.

[0060]

[0061] Q S (x) represents the number of pixels in the S channel with a saturation value equal to x. This represents the number of pixels in channel S distributed within the interval [0, i]. Q is the total number of pixels in the S channel; where Q V (x) represents the number of pixels in the V channel with a brightness value equal to x. Let V be the number of pixels distributed in the interval [j, R]. is the total number of pixels in the V channel, where i and j are threshold parameters that control the evaluation of the dual-channel feature index GSV;

[0062] The dual-channel feature index GSV characterizes the probability and visibility of white in a rectangular power supply area image, i.e., the snow thickness. The higher the dual-channel feature index GSV value, the higher the combined value of the snow area and thickness in the rectangular power supply area image, and the more severe the snow coverage probability and snow condition.

[0063] 3.2 Extraction of Image Intrinsic Edge Density Features

[0064] The aforementioned dual-channel feature index GSV in the HSV color space mainly collects the features of image brightness and color changes. However, due to the influence of factors such as the drone inspection perspective and real-time weather lighting, the dual-channel feature index GSV of the HSV color space for rectangular power supply area images in the same area may differ at different acquisition times and angles. Considering that when the snow cover is stable, the surface of the snow-covered area is smooth due to the snow cover, reducing the number of detectable edges in the image, while the non-snow-covered area has a large number of edge structures due to the presence of various structures, and these edge structures are generally not affected by lighting and color changes. At the same time, edge detection algorithms are already very mature and stable, including the Canny algorithm, the Roberts algorithm, etc. Therefore, this application establishes the internal edge density feature index EDGE of the image based on this. Q edge Q represents the number of pixels in the rectangular power supply area image that can detect edges; sum The total number of pixels in the rectangular power supply area image; where, in each algorithm, an edge determination threshold parameter k is involved to determine whether a pixel belongs to an edge;

[0065] The image internal edge density feature index EDGE also characterizes the snow coverage area in the snow-covered area of ​​the rectangular power supply area image. The larger the image internal edge density feature index EDGE value, the larger the snow coverage area and snow depth.

[0066] Step 3: Correlation and Matching of Snow Cover Indicators and Snow Cover Status

[0067] In actual data analysis, since the snow cover index in the aforementioned rectangular power supply area image is a continuous variable, while the snow cover state is a discrete variable, to more accurately and effectively correlate the snow cover characteristics with the snow cover state, and to more quickly and accurately determine the snow cover state of a specific rectangular power supply area image, the analysis of variance is used to analyze the degree of influence of the snow cover index attributes on the snow cover state. Therefore, for each snow cover index, its influence on the snow cover state can be expressed as a correlation coefficient.

[0068] in It is the sum of squares of the deviations between states; w represents the state classification; n r The number of rectangular power supply area images belonging to state w; Let be the mean of the feature indexes of the rectangular power supply area image belonging to state w. x represents the mean of the feature indices for all rectangular power supply area images. iw is the feature index value of the i-th sample within the rectangular power supply area image group belonging to state w;

[0069] The specific calculation process is as follows:

[0070] 4.1 Obtain all rectangular power supply area images in each category database. Under different values ​​of threshold parameters i, j and edge determination threshold parameter k, calculate the dual-channel feature index GSV and the internal edge density feature index EDGE respectively.

[0071] 4.2 Calculate the influence (μ) of the dual-channel characteristic index GSV and the internal edge density characteristic index EDGE on the snow cover state under the aforementioned different value conditions. GSV and μ EDGE ;

[0072] 4.3 Based on empirical judgment, μ is selected. GSV and μ EDGE Values ​​with a correlation coefficient of 0.15 or higher are used as the criteria for judging relevance, and values ​​with a correlation coefficient of less than 0.15 are deleted.

[0073] 4.4 Based on actual computing resources and accuracy requirements, select effective values ​​according to the correlation coefficient from high to low.

[0074] 4.5 Based on the aforementioned valid value scenarios, several dual-channel feature indices GSV and internal edge density feature indices EDGE are obtained respectively;

[0075] Step 4: Training the snow-covered state recognition model based on support vector machines

[0076] After obtaining several snow cover feature indicators based on the aforementioned step four, these indicators are used as input to a support vector machine (SVM) with the snow cover state as the output. The SVM is then trained to obtain a snow cover state recognition model.

[0077] Step 5: Identify snow cover status based on the snow cover status recognition model.

[0078] Based on the snow cover status recognition model obtained in step five above, the inspection image to be analyzed is used as input to obtain the snow cover status recognition result.

[0079] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit the scope of protection of this application. Although this application has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of this application without departing from the substance and scope of the technical solutions of this application.

Claims

1. A method for detecting snow cover status in winter power distribution areas based on unmanned aerial vehicle (UAV) inspection, characterized in that, Includes the following steps: Step 1: Collect test training data and establish a classification database, which specifically includes: Collect relevant snow-covered image data from existing data as raw data, classify them according to different landforms and different zoning functions, and establish test and training databases corresponding to each type. Step 2: Preprocessing of test training data, specifically including: 2.1 Power supply area extraction Specifically, this refers to: acquiring source data images, drawing circular power supply area images with the center of each power distribution area in the image as the center and the distance between the farthest user and the power distribution station as the radius, and drawing rectangular power supply area images with the circular power supply area images as the circumcircle, such that at least one vertex of the rectangular power supply area image is located at the farthest power user; 2.2 Pixel Adjustment Specifically, this refers to: acquiring all rectangular power supply area images, and adjusting all rectangular power supply area images to a uniform pixel height or width while maintaining their original aspect ratio; 2.3 Status Labeling Specifically, this refers to using existing image annotation programs and either manual or machine recognition methods to categorize all rectangular power supply area images according to the snow cover index. The snow cover status is classified into four levels: low snow cover, moderate snow cover, high snow cover, and extremely high snow cover. ;in This refers to the historical snow cover thickness since the current power supply area was formed; This refers to the historical ambient temperature since the current power supply area was formed; Step 3: Extraction of snow cover indicators from images of each rectangular power supply area, specifically including: 3.1 Dual-channel feature indicators based on HSV color space. Extraction Specifically, this refers to: using the channel value range as the horizontal axis and dividing it into R intervals, plotting a histogram based on the number of pixels within each corresponding value interval, and establishing a dual-channel feature index based on the HSV color space. ; in ; This represents the number of pixels in the S channel whose saturation value is equal to x. This represents the number of pixels in channel S distributed within the interval [0, i]. The total number of pixels in the S channel; where This represents the number of pixels in the V channel whose brightness value equals x. Let V be the number of pixels distributed in the interval [j, R]. The total number of pixels in the V channel, where i and j are the control parameters for the dual-channel features. 3.2 Extraction of image internal edge density feature index; The threshold parameter for evaluation; Specifically, this refers to: establishing an internal edge density feature index for the image based on this method. and ;in This represents the number of pixels representing the detectable edges in the rectangular power supply area image. The total number of pixels in the rectangular power supply area image; where, in each algorithm, an edge determination threshold parameter k is involved to determine whether a pixel belongs to an edge; Step 4: Matching snow cover indicators with snow cover status, specifically including: Analysis of variance (ANOVA) is used to evaluate the influence of snow cover index attributes on snow cover status. For each snow cover index, its influence on snow cover status can be expressed as a correlation coefficient. ;in It is the sum of squares of the deviations between states; ; r represents the state classification; The number of rectangular power supply area images belonging to state r; Let be the mean of the feature indexes of the rectangular power supply area image belonging to state r. The mean of the feature indices for all rectangular power supply area images; Let i be the feature index value of the i-th sample in the rectangular power supply area image group belonging to state r; the specific calculation process is as follows: 4.1 Obtain all rectangular power supply area images in each classification database, and calculate the dual-channel feature index under different values ​​of threshold parameters i, j and edge determination threshold parameter k. and internal edge density characteristic index 4.2 Calculate the dual-channel characteristic index under the aforementioned different value cases. and internal edge density characteristic index The extent of the impact on snow cover status and 4.3 Based on experience, select and 4.4 Values ​​with a correlation coefficient of 0.15 or higher are used as the criterion for determining relevance, and values ​​with a correlation coefficient below 0.15 are deleted; 4.5 Based on actual computing resources and accuracy requirements, effective values ​​are selected in descending order of correlation coefficient; 4.6 Based on the aforementioned effective values, several dual-channel feature indicators are obtained. and internal edge density characteristic index ; Step 5: Training the snow-covered state recognition model based on support vector machines, specifically including: After obtaining the snow cover feature index based on the aforementioned step four, use it as the input of the support vector machine and the snow cover state as the output to train the support vector machine and obtain the snow cover state recognition model. Step 6: Identify the snow cover state based on the snow cover state recognition model, specifically including: Based on the snow cover status recognition model obtained in step five above, the inspection image to be analyzed is used as input to obtain the snow cover status recognition result.

2. The method for detecting snow cover status in winter power distribution areas based on UAV inspection according to claim 1, characterized in that, In step two, to improve the accuracy of recognition, the source data should be annotated in multiple rounds, and based on the results of each round of hierarchical annotation of the rectangular power supply area image, the final hierarchical annotation of each rectangular power supply area image should be performed according to the probability of the hierarchical result appearing.

3. The method for detecting snow cover status in winter power distribution areas based on UAV inspection according to claim 1, characterized in that, In step two, after calculating the snow cover index... Under the premise of this, based on the historical data of the current power supply area, the specific snow cover index values ​​corresponding to different snow cover indices are determined empirically, and specific snow cover index values ​​are specified for different power supply areas under low snow cover, moderate snow cover, high snow cover, and extremely high snow cover.

4. The method for detecting snow cover status in winter power distribution areas based on UAV inspection according to claim 1, characterized in that, In step one, the landform types include at least mountainous areas, plains, and hills; the zoning functions include at least rural areas, urban areas, industrial areas, and agricultural areas.

Images