A method for pre-warning of high-voltage transmission line galloping without spacer without interval in night vision scene

Abstract

The application discloses a kind of galloping early warning methods of high-voltage transmission line without spacer bar under night vision scene, comprising the following steps: S1, install high-pixel monitoring snapshot integrated machine on line tower, collect line sample picture set under night scene;S2, use dark channel prior theory to carry out night vision image enhancement to high-voltage transmission line;S3, using geometric rotation, scaling and other methods to expand sample set, label data set, construct night line detection and segmentation data set;S 4, using YOLOv5 target detection model to quickly identify high-voltage transmission line in each frame image;S5, add a segmentation head to YOLOv5 to carry out semantic segmentation to high-voltage transmission line;S6, use morphological method to carry out line boundary extraction and tracking;S7, oval trajectory detection is carried out to the boundary tracked, combined with different early warning indexes, realize night high-voltage transmission line without spacer bar galloping early warning.The application can effectively early warn the occurrence of high-voltage transmission line galloping event under night vision scene.

Description

Technical Field

[0001] This invention relates to the fields of computer vision and power transmission line monitoring, and in particular to a method for early warning of galloping of high-voltage power transmission lines without spacers in night vision scenarios. Background Technology

[0002] Transmission line galloping refers to low-frequency, large-amplitude self-excited vibrations generated by wind excitation when a line is eccentrically covered with ice under certain natural conditions. Once a galloping accident occurs, it can cause the high-voltage transmission line itself to break, towers to be damaged, etc., seriously threatening the safe operation of high-voltage transmission lines and affecting the stable operation of the power grid.

[0003] Current methods for monitoring line galloping risks primarily use line spacers as detection feature points. Image processing technology is used to monitor the center coordinates of these feature points and track their displacement trajectories in the horizontal and vertical directions to achieve line galloping monitoring. However, in some special scenarios, such as high-voltage transmission lines in open-pit coal mines, spacers are generally absent, making galloping monitoring methods using fixed spacers as feature points no longer applicable. Furthermore, while image and video camera technologies are used for high-voltage transmission line galloping monitoring, they also face the challenge of low line outline clarity in night vision conditions. Summary of the Invention

[0004] Considering the current lack of methods for monitoring the galloping of high-voltage transmission lines without spacers, and the poor performance of galloping monitoring in night vision scenarios, the technical problem to be solved by this invention is to provide a method that can enhance the dark channel of line images in night vision scenarios, effectively improving the clarity of the line in night vision. Based on this, the invention uses the YOLOv5 model for image detection and segmentation, employs morphological methods to extract and track line boundaries, performs elliptical trajectory detection on the boundaries, and combines various early warning indicators to achieve a method for early warning of galloping of high-voltage transmission lines without spacers in night vision scenarios.

[0005] The technical solution adopted in this invention is: a method for early warning of galloping of high-voltage transmission lines without spacers in night vision scenarios, characterized by the following steps:

[0006] S1. Install high-pixel monitoring and capture integrated machines on power supply line towers in the power supply area that needs to be monitored to collect sample image sets of the lines in nighttime scenes;

[0007] S2. Night vision image enhancement of high-voltage transmission lines using dark channel prior theory;

[0008] S3. Expand the sample set using geometric rotation, scaling, MOSAIC data augmentation, etc., label the dataset, and construct a nighttime line detection and segmentation dataset;

[0009] S4. Use the YOLOv5 target detection model to quickly identify high-voltage transmission lines in each frame of the image;

[0010] S5. Add a segmentation header to YOLOv5 to perform semantic segmentation of high-voltage transmission lines;

[0011] S6. Use morphological erosion and boundary tracing methods to extract and track line boundaries;

[0012] S7. Detect elliptical trajectories at the tracking boundaries and combine them with different early warning indicators to achieve nighttime early warning of galloping of high-voltage transmission lines without spacers; specifically:

[0013] In step S7, elliptical trajectory detection is performed on the tracked boundary. When the arc segments of the line boundary are extracted, invalid arc segments are removed, and elliptical fitting and verification are performed on the valid arc segments. The specific steps are as follows:

[0014] S711: Extract the line boundary arc segments and use the boundary pixel number threshold condition to remove invalid arc segments with a small number of pixels;

[0015] S712: Based on the gradient direction of the pixels on the arc segment, the effective arc segments are divided into two categories: arc segments with positive derivative values ​​are denoted as D=+1, and arc segments with negative derivative values ​​are denoted as D=-1;

[0016] S713: Based on concavity and convexity, effective arc segments are divided into two categories: concave arcs and convex arcs. The division is made by the relative position of the straight line L connecting the start and end points of the arc segment and the pixel points on the arc. The discrimination formula is as follows: If f(x,y)>0 always exists, it is a convex arc; if f(x,y)<0 always exists, it is a concave arc; if both f(x,y)>0 and f(x,y)<0 exist, the arc segment is discarded.

[0017] Where (xL, yL) and (xR, yR) are the coordinates of the beginning and end points of the arc segment, respectively.

[0018] S714: Combining steps S712 and S713, the arc segments can be divided into four categories (convex, D=+1), (convex, D=-1), (concave, D=-1), and (concave, D=+1) according to the grayscale direction and concavity / convexity of the pixels; select 5 data points from the arc segments (convex, D=+1) and (concave, D=+1), or (convex, D=-1) and (concave, D=-1) to fit the ellipse equation and solve for the ellipse equation parameters;

[0019] S715: Pixel point evaluation is performed using the obtained elliptical model, and the set of interior points that conform to the elliptical model is determined by the geometric distance norm 2.

[0020] S716: If the number of interior points obtained is greater than the number of interior points obtained in the previous step, then update the optimal interior point set and the optimal ellipse model.

[0021] S717: Determine whether the optimal set of interior points has reached the expected number. If so, output the optimal ellipse model and directly execute step S719.

[0022] S718: Repeat S714-S717 to complete the arc segment ellipse fitting;

[0023] S719: If the ratio of the number of pixels belonging to the current ellipse to the ellipse's circumference is less than a set threshold (0.45), then the ellipse is considered a false ellipse.

[0024] In step S7, to determine whether the verified trajectory might form an ellipse, the above three dancing warning indicators and the probability of forming an ellipse are combined to perform the following steps:

[0025] S721: If the verified trajectory cannot form an ellipse, the warning level is no action; if it can form an ellipse, calculate the probability P of forming an ellipse and proceed to S722.

[0026] S722: If the probability P is less than 25%, the warning level is also no action; otherwise, proceed to step S723.

[0027] S723: If the probability P is less than 50%, the warning level is Level II; otherwise, the warning level is Level I.

[0028] Once the relevant steps are completed, on-site maintenance personnel can ultimately determine whether line galloping has occurred at night and the level of threat posed by it, so as to take corresponding countermeasures.

[0029] Preferably, the installation location of the high-pixel monitoring and capture integrated machine in step S1 should maintain a suitable distance from the high-voltage transmission line, and the camera's viewing angle and focal length should remain fixed, capturing images according to the set equal time intervals.

[0030] Preferably, in step S2, night vision image enhancement of high-voltage transmission lines is performed using dark channel prior theory. The specific process includes: first, dividing the image into small grids of uniform height; then, calculating the minimum values ​​of the three RGB color channels in each grid. According to dark channel prior theory, for most local areas of an outdoor fog-free image (excluding the sky area), there exist some pixels (dark pixels) with very low values ​​in at least one color channel, approaching 0. This can be expressed by the formula:

[0031] J dark →0, J c (y) represents any color channel of J, J dark For the dark channel, Ω(x) represents the window at pixel X.

[0032] Then, atmospheric transmittance is calculated based on illumination intensity and dark channel image, and the final image is restored:

[0033] Substituting the last two equations into the first equation, we get:

[0034] A c Indicates atmospheric light intensity, Transmittance, This is the enhanced image.

[0035] Preferably, the sample set to be expanded in step S3 comes partly from the acquired night vision high-voltage transmission line images and partly from the images enhanced based on dark channel priors. For the annotation of the detection and segmentation dataset, three image target type labels are selected: line, tower, and tree.

[0036] Preferably, after establishing the YOLOv5 object detection network in step S4, the model is trained using the detection dataset constructed in step S3 with different training parameters.

[0037] Preferably, the segmentation head added in step S5 uses the lowest-level feature map of FPN (FPN). W / 8, H Using (8, 256) as input, perform three nearest neighbor upsampling operations, and finally output ( W , H ,2) Feature map; In the transmission line segmentation header, the transmission line is the foreground and the others are the background.

[0038] Preferably, in step S6, for high-voltage transmission lines without spacers, the changes in line morphology are used as the basis for galloping analysis, and morphological methods are used to extract and track the line boundaries. The specific steps are as follows:

[0039] 1) Boundary Extraction

[0040] S61: Scan the segmented image line by line using a 3x3 convolution kernel;

[0041] S62: If all 8 neighbors of a black point are black points or all 8 neighbors of a white point are white points, then the point is an internal point, and the pixels outside the white internal point are eroded.

[0042] S63: Subtract the eroded image from the original image, retaining the boundary pixels to achieve line boundary extraction;

[0043] 2) Boundary tracking

[0044] S64: After the line boundary is extracted, select a pixel on the target line boundary, and scan the pixel and its neighborhood with a 3*3 convolution kernel starting from the pixel.

[0045] S65: If the pixel is an internal point, then its neighboring pixel will be set as the starting point in a clockwise (or counterclockwise) order;

[0046] S66: Repeat this process until all other pixels on the line boundary are found, and then repeat S64~S66 again to achieve tracking of the entire line boundary.

[0047] The beneficial effects of this invention are:

[0048] (1) The present invention solves the problem that the target clarity in the monitoring image is low in night vision scene, resulting in low detection accuracy and false detection and missed detection;

[0049] (2) The present invention adopts an algorithm that combines morphological theory and elliptical trajectory detection to realize line galloping early warning, which solves the problem of lack of galloping detection methods for high-voltage transmission lines without spacers in special scenarios.

[0050] (3) This invention combines expert domain knowledge with big data AI, providing better ideas for the implementation of artificial intelligence.

[0051] Overall, this invention can enhance the dark channel of line images in night vision scenarios, effectively improving the clarity of night vision lines. Based on this, the image is detected and segmented using the YOLOv5 model, and morphological methods are used to extract and track the line boundaries. Elliptical trajectory detection is performed on the boundaries, and combined with various early warning indicators, the invention can achieve early warning of galloping of high-voltage transmission lines without spacers in night vision scenarios. Attached Figure Description

[0052] Figure 1 This is a flowchart illustrating an embodiment of the present invention: a method for early warning of galloping of high-voltage transmission lines without spacers in night vision scenarios.

[0053] Figure 2 This is a comparison of night vision image data enhancement results obtained from an embodiment of the present invention, which provides an early warning method for galloping of high-voltage transmission lines without spacers in night vision scenarios.

[0054] Figure 3 This is a schematic diagram of the MOSAIC data enhancement method used in the present invention, which is a method for early warning of galloping of high-voltage transmission lines without spacers in night vision scenarios.

[0055] Figure 4 This invention relates to the YOLOv5 network model framework used in a night vision-based early warning method for high-voltage transmission lines without spacers.

[0056] Figure 5This is a schematic diagram of the high-voltage transmission line identification results according to an embodiment of the method for early warning of galloping high-voltage transmission lines without spacers in night vision scenarios of the present invention;

[0057] Figure 6 This is a schematic diagram of the high-voltage transmission line segmentation result according to an embodiment of the method for early warning of galloping of high-voltage transmission lines without spacers in night vision scenarios of the present invention;

[0058] Figure 7 This is a schematic diagram of morphological line boundary extraction and tracking, based on an embodiment of a method for early warning of galloping high-voltage transmission lines without spacers in night vision scenarios according to the present invention.

[0059] Figure 8 This is a schematic diagram of the line detection and segmentation results based on the YOLOv5 model in an embodiment of the method for early warning of galloping of high-voltage transmission lines without spacers in night vision scenarios according to the present invention;

[0060] Figure 9 This is a schematic diagram of the ellipse detection result based on arc segment extraction, from an embodiment of the present invention, which provides an early warning method for galloping of high-voltage transmission lines without spacers in night vision scenarios. Detailed Implementation

[0061] The invention will now be further described with reference to the accompanying drawings.

[0062] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly described below with reference to the accompanying drawings of the embodiments. These embodiments are only some, not all, of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the protection scope of the present invention.

[0063] like Figure 1 As shown in the figure, an embodiment of the present invention provides a method for early warning of galloping of high-voltage transmission lines without spacers in night vision scenarios, comprising the following steps:

[0064] S1. Install high-pixel monitoring and capture integrated machines on power supply line towers in the power supply area that needs to be monitored to collect sample image sets of the lines in nighttime scenes;

[0065] In a specific embodiment, the installation location of the high-pixel monitoring and capture integrated machine in step S1 should maintain a suitable distance from the high-voltage transmission line, and the camera's angle and focal length should remain fixed, capturing images at a set time interval of every ten minutes.

[0066] S2. Night vision image enhancement of high-voltage transmission lines using dark channel prior theory;

[0067] In a specific embodiment, step S2 utilizes the dark channel prior theory to enhance the night view image of the high-voltage transmission line, including the following steps:

[0068] First, the image is divided into small grids of uniform height. Within each grid, the minimum value of each of the three RGB color channels is calculated. Based on the dark channel prior theory, for most local areas of a fog-free outdoor image (excluding the sky region), there exist some pixels (dark pixels) with very low values ​​in at least one color channel, approaching 0. This can be expressed by the formula:

[0069] The color channel represented is the dark channel, and the window at pixel X is represented.

[0070] Then, atmospheric transmittance is calculated based on illumination intensity and dark channel image, and the final image is restored:

[0071]

[0072] Substituting the last two equations into the first equation, we get:

[0073]

[0074] Indicates atmospheric light intensity, Transmittance, This is the enhanced image.

[0075] S3. Expand the sample set using geometric rotation, scaling, MOSAIC data augmentation, etc., label the dataset, and construct a nighttime line detection and segmentation dataset;

[0076] In a specific embodiment, the sample set to be expanded in step S3 is partly from the collected night vision high-voltage transmission line images and partly from the images enhanced based on the dark channel prior; for the annotation of the detection and segmentation dataset, three types of image target type labels are selected: line, tower, and tree.

[0077] S4. Use the YOLOv5 target detection model to quickly identify high-voltage transmission lines in each frame of the image;

[0078] In a specific embodiment, after establishing the YOLOv5 object detection network in step S4, the detection dataset constructed in step S3 is used to train the model with different training parameters, as shown in Table 1. In this embodiment of the invention, the line recognition effect is best when the epochs are 300, the Bacth-size is 8, and the threshold is 0.55. The recognition results are as follows: Figure 5 As shown.

[0079] Table 1 Training parameter settings

[0080]

[0081] S5. Add a segmentation header to YOLOv5 to perform semantic segmentation of high-voltage transmission lines;

[0082] In a specific embodiment, the segmentation head added in step S5 uses the lowest-level feature map of FPN ( W / 8, H Using (8, 256) as input, perform three nearest neighbor upsampling operations, and finally output ( W , H 2) Feature map; In the transmission line segmentation header, the transmission line is the foreground and the others are the background. The segmentation result is as follows: Figure 6 As shown.

[0083] S6. Use morphological methods to extract and track line boundaries;

[0084] In a specific embodiment, the morphological method used for line boundary extraction and tracking in step S6 is as follows:

[0085] 1) Boundary Extraction

[0086] S61: Scan the segmented image line by line using a 3x3 convolution kernel;

[0087] S62: If all 8 neighbors of a black dot (pixel 0) are black dots or all 8 neighbors of a white dot (pixel 255) are white dots, then the dot is an interior dot, and the pixels outside the white interior dot are eroded (set to the background color).

[0088] S63: Subtract the eroded image from the original image, retaining the boundary pixels to achieve line boundary extraction;

[0089] 2) Boundary tracking

[0090] S64: After the line boundary is extracted, select a pixel on the target line boundary, and scan the pixel and its neighborhood with a 3*3 convolution kernel starting from the pixel.

[0091] S65: If the pixel is an internal point, then its neighboring pixel will be set as the starting point in a clockwise (or counterclockwise) order;

[0092] S66: Repeat this process until all other pixels on the line boundary are found. Then repeat S64~S66 again to track the entire line boundary. Extract and track the results as follows: Figure 7 As shown.

[0093] S7. Perform elliptical trajectory detection on the tracked boundary and combine different early warning indicators to achieve early warning of galloping of high-voltage transmission lines without gap bars at night.

[0094] In a specific embodiment, in step S7, elliptical trajectory detection is performed on the tracked boundary. When the arc segments of the line boundary are extracted, invalid arc segments are removed, and elliptical fitting and verification are performed on the valid arc segments. The steps are as follows:

[0095] S711: Extract the line boundary arc segments and use the boundary pixel number threshold condition to remove invalid arc segments with a small number of pixels;

[0096] S712: Based on the gradient direction of pixels on the arc segment, effective arc segments are divided into two categories. Arc segments where the derivative of a pixel is positive are denoted as... D =+1, the arc segment whose derivative at the pixel point is negative is denoted as . D =-1;

[0097] S713: Based on their concavity or convexity, effective arc segments are divided into two categories: concave arcs and convex arcs. These are determined by a straight line connecting the beginning and end points of the arc segment. L The relative positions of pixels on the arc are used to divide the data, and the discrimination formula is as follows; if always... f ( x, y If )>0, then it is a convex arc; if it always holds true... f ( x, y If ) < 0, then it is a concave arc; if there is already f ( x, y )>0 and there is f ( x, y If ) < 0, then this arc segment is removed.

[0098]

[0099] in,( x L , y L )and( x R , y R ) are the coordinates of the beginning and end points of the arc segment, respectively.

[0100] S714: Combining steps S712 and S713, the arc segment can be divided into (convex, non-convex) based on the pixel grayscale direction and concavity / convexity. D =+1), (convex, D =-1), (concave, D =-1), (concave, D =+1) four categories; from arc segment (convex, D =+1) and (concave, D =+1), or (convex, D =-1) and (concave, D Five data points are selected from =-1) to fit the ellipse equation and solve for the ellipse equation parameters;

[0101] S715: Pixel point evaluation is performed using the obtained elliptical model, and the set of interior points that conform to the elliptical model is determined by the geometric distance norm 2.

[0102] S716: If the number of interior points obtained is greater than the number of interior points obtained in the previous step, then update the optimal interior point set and the optimal ellipse model.

[0103] S717: Determine whether the optimal set of interior points has reached the expected number. If so, output the optimal ellipse model and directly execute step S719.

[0104] S718: Repeat S714-S717 to complete the arc segment ellipse fitting;

[0105] S719: If the ratio of the number of pixels belonging to the current ellipse to the ellipse circumference is less than a set threshold (0.45), then the ellipse is considered a false ellipse.

[0106] In a specific embodiment, to determine whether the verified trajectory might form an ellipse, the above three dancing warning indicators and the probability of forming an ellipse are combined to perform the following steps:

[0107] S721: If the verified trajectory cannot form an ellipse, the warning level is no action; if it can form an ellipse, the probability of forming an ellipse is calculated. P Continue executing S722;

[0108] S722: If probability P If the percentage is less than 25%, the warning level is also no action; otherwise, proceed to step S723.

[0109] S723: If probability P If the rate is less than 50%, the warning level is Level II; otherwise, the warning level is Level I.

[0110] Once the relevant steps are completed, on-site maintenance personnel can ultimately determine whether line galloping has occurred at night and the level of threat posed by it, so as to take corresponding countermeasures.

[0111] In this embodiment of the invention, the ellipse detection result based on arc segment extraction is as follows: Figure 9 As shown, without spacers as observation points for judging line galloping, expert knowledge can be used to determine if a line galloping accident has occurred if the line's motion forms an ellipse. Lines 1 and 2 have a relatively high probability of forming elliptical motion, Line 3 has the potential to form elliptical motion, while Line 4 does not have the potential to form an elliptical motion trajectory.

[0112] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0113] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for early warning of galloping of high-voltage transmission lines without spacers in night vision scenarios, characterized in that: Includes the following steps: S1. Install high-pixel monitoring and capture integrated machines on power supply line towers in the power supply area that needs to be monitored to collect sample image sets of the lines in nighttime scenes; S2. Night vision image enhancement of high-voltage transmission lines using dark channel prior theory; S3. Expand the sample set using geometric rotation, scaling, and MOSAIC data augmentation, label the dataset, and construct a nighttime line detection and segmentation dataset; S4. Use the YOLOv5 target detection model to quickly identify high-voltage transmission lines in each frame of the image; S5. Add a segmentation header to YOLOv5 to perform semantic segmentation of high-voltage transmission lines; S6. Use morphological erosion and boundary tracing methods to extract and track line boundaries; S7. Detect elliptical trajectories at the tracking boundaries and combine them with different early warning indicators to achieve nighttime early warning of galloping of high-voltage transmission lines without spacers; specifically: In step S7, elliptical trajectory detection is performed on the tracked boundary. When the arc segments of the line boundary are extracted, invalid arc segments are removed, and elliptical fitting and verification are performed on the valid arc segments. The specific steps are as follows: S711: Extract the line boundary arc segments and use the boundary pixel number threshold condition to remove invalid arc segments with a small number of pixels; S712: Based on the gradient direction of the pixels on the arc segment, the effective arc segments are divided into two categories: arc segments with positive derivative values ​​are denoted as D=+1, and arc segments with negative derivative values ​​are denoted as D=-1; S713: Based on concavity and convexity, effective arc segments are divided into two categories: concave arcs and convex arcs. The division is based on the relative position of the straight line L connecting the start and end points of the arc segment and the pixels on the arc. The discrimination formula is as follows: If f(x,y)>0 always holds, it is a convex arc; if f(x,y)<0 always holds, it is a concave arc; if both f(x,y)>0 and f(x,y)<0 exist, the arc segment is discarded. Where (xL, yL) and (xR, yR) are the coordinates of the beginning and end points of the arc segment, respectively; S714: Combining steps S712 and S713, the arc segments are divided into four categories (convex, D=+1), (convex, D=-1), (concave, D=-1), and (concave, D=+1) according to the grayscale direction and concavity / convexity of the pixels; 5 data points are selected from the arc segments (convex, D=+1) and (concave, D=+1), or (convex, D=-1) and (concave, D=-1) to fit the ellipse equation and solve for the ellipse equation parameters; S715: Pixel point evaluation is performed using the obtained elliptical model, and the set of interior points that conform to the elliptical model is determined by the geometric distance norm 2. S716: If the number of interior points obtained is greater than the number of interior points obtained in the previous step, then update the optimal interior point set and the optimal ellipse model. S717: Determine whether the optimal set of interior points has reached the expected number. If so, output the optimal ellipse model and directly execute step S719. S718: Repeat S714-S717 to complete the arc segment ellipse fitting; S719: If the ratio of the number of pixels belonging to the current ellipse to the ellipse's circumference is less than a set threshold of 0.45, then the ellipse is considered a false ellipse. In step S7, to determine whether the verified trajectory might form an ellipse, the above three dancing warning indicators and the probability of forming an ellipse are combined to perform the following steps: S721: If the verified trajectory cannot form an ellipse, the warning level is no action; if it can form an ellipse, calculate the probability P of forming an ellipse and proceed to S722. S722: If the probability P is less than 25%, the warning level is also no action; otherwise, proceed to step S723. S723: If the probability P is less than 50%, the warning level is Level II; otherwise, the warning level is Level I. After the relevant steps are completed, the on-site maintenance personnel will make a final judgment on whether line galloping has occurred at night and the level of threat of the galloping, so as to take corresponding measures.

2. The method for early warning of galloping of high-voltage transmission lines without spacers in night vision scenarios according to claim 1, characterized in that: In step S1, the installation location of the high-pixel monitoring and capture device should maintain a suitable distance from the high-voltage transmission line, and the camera's angle and focal length should remain fixed, capturing images according to the set equal time intervals.

3. The method for early warning of galloping of high-voltage transmission lines without spacers in night vision scenarios according to claim 1, characterized in that: In step S2, night vision image enhancement of high-voltage transmission lines is performed using dark channel prior theory. The specific process includes: first, dividing the image into small grids of uniform height; then, calculating the minimum value of each of the three color channels (RGB) in each grid. According to dark channel prior theory, for most local areas of an outdoor fog-free image, there are some pixels with very low values ​​in at least one color channel, approaching 0. This can be expressed by the formula: express Any color channel, It is a dark passage. Represents the window at pixel X; Then, atmospheric transmittance is calculated based on illumination intensity and dark channel image, and the final image is restored: Substituting the last two equations into the first equation, we get: Indicates atmospheric light intensity, Transmittance, This is the enhanced image.

4. The method for early warning of galloping of high-voltage transmission lines without spacers in night vision scenarios according to claim 1, characterized in that: In step S3, the sample set that needs to be expanded comes partly from the collected night vision high-voltage transmission line images and partly from the images enhanced based on the dark channel prior.

5. The method for early warning of galloping of high-voltage transmission lines without spacers in night vision scenarios according to claim 1, characterized in that: In step S3, for the annotation of the detection and segmentation dataset, three types of image target type labels are selected: lines, towers, and trees.

6. The method for early warning of galloping of high-voltage transmission lines without spacers in night vision scenarios according to claim 1, characterized in that: In step S4, after establishing the YOLOv5 object detection network, the model is trained using the detection dataset constructed in step S3 and different training parameters are set.

7. The method for early warning of galloping of high-voltage transmission lines without spacers in night vision scenarios according to claim 1, characterized in that: In step S5, the added segmentation head uses the lowest-level feature map of FPN ( W / 8, H Using (8, 256) as input, perform three nearest neighbor upsampling operations, and finally output ( W , H Feature map of ,2); In the transmission line segmentation header, the transmission line is the foreground and everything else is the background.

8. The method for early warning of galloping of high-voltage transmission lines without spacers in night vision scenarios according to claim 1, characterized in that: In step S6, for high-voltage transmission lines without spacers, the changes in line morphology are used as the basis for galloping analysis. Morphological methods are used to extract and track the line boundaries. The specific steps are as follows: 1) Boundary Extraction S61: Scan the segmented image line by line using a 3x3 convolution kernel; S62: If all 8 neighbors of a black point are black points or all 8 neighbors of a white point are white points, then the point is an internal point, and the pixels outside the white internal point are eroded. S63: Subtract the eroded image from the original image, retaining the boundary pixels to achieve line boundary extraction; 2) Boundary tracking S64: After the line boundary is extracted, select a pixel on the target line boundary, and scan the pixel and its neighborhood with a 3*3 convolution kernel starting from the pixel. S65: If the pixel is an internal point, then one of its neighboring pixels will be set as the starting point in a clockwise order; S66: Repeat this process until all other pixels on the line boundary are found, and then repeat S64~S66 again to achieve tracking of the entire line boundary.

Images