Power transmission line icing thickness identification method and system based on improved hough transform

Abstract

The present application relates to the technical field of power transmission line monitoring, in particular to a power transmission line icing thickness identification method and system based on improved Hough transformation. The method collects RGB images of the power transmission line by a camera and stores the reference image; according to the global light intensity, the RGB channel weighting coefficient of the RGB image is adaptively determined, the weighted gray image is synthesized, the binary edge image is generated through filtering denoising, Sobel operator gradient calculation, non-maximum suppression and double threshold edge detection; the binary edge image is set with multiple groups of scale parameters from coarse to fine for Gaussian convolution, the skeleton line and bilateral edge line are extracted through three-dimensional accumulator voting, angle coverage verification and false alarm number verification; the local tangent fitting is performed by traversing the skeleton line pixel points, the local pixel width is calculated along the normal direction, the actual icing thickness is converted combined with the reference image, and the coordinate deviation is corrected based on the quantization offset mean value, and the uneven icing thickness distribution along the line is output.

Description

Technical Field

[0001] This invention relates to the field of transmission line monitoring technology, specifically to a method and system for identifying icing thickness on transmission lines based on an improved Hough transform. Background Technology

[0002] Icing on transmission lines is a significant factor threatening the safe and stable operation of power systems. Excessive ice accumulation can lead to serious accidents such as conductor breakage, tower collapse, and line tripping, posing a major threat to power grid safety. Therefore, real-time monitoring and accurate identification of ice thickness on transmission lines is of significant engineering importance.

[0003] Existing methods for detecting icing thickness mainly fall into three categories: manual inspection, direct sensor measurement, and image recognition. Manual inspection is labor-intensive and inefficient, making it difficult to meet the real-time monitoring needs of large-scale lines. While direct sensor measurement offers high accuracy, it is costly to install and maintain, and fails to reflect the uneven distribution of icing along the line. Image recognition methods, with their advantages of being non-contact, low-cost, and easy to deploy remotely, have received widespread attention in recent years. However, existing image recognition methods still face several challenges when processing images of iced transmission lines: firstly, icing causes irregular deformation and local breaks at the line edges, making it difficult for traditional edge detection methods to accurately extract the complete line contour; secondly, changes in illumination, background noise, and pixel discretization errors significantly affect edge extraction accuracy, leading to large deviations in icing thickness measurement results; furthermore, most existing methods can only provide an overall estimate of icing thickness, failing to reflect the local distribution characteristics of uneven icing along the line. Summary of the Invention

[0004] This invention provides a method and system for identifying the icing thickness of transmission lines based on an improved Hough transform, enabling high-precision, non-contact, automated measurement of uneven icing thickness.

[0005] To achieve the above objectives, the present invention provides the following technical solution: This invention relates to a method for identifying icing thickness on transmission lines based on an improved Hough transform, comprising: S100: Captures RGB images of power transmission lines using a camera and stores a baseline image of the line in an ice-free state; S200: The RGB channel weighting coefficients of the RGB image are adaptively determined according to the global illumination intensity, and a grayscale image is synthesized by weighting; after filtering and denoising, Sobel operator gradient calculation and non-maximum suppression, and double threshold edge detection, a binary edge image is generated; S300: Gaussian convolution is performed on the binary edge image with multiple sets of scale parameters from coarse to fine to generate a multi-scale smooth image; the edge point set of each scale smooth image is extracted, and candidate detection results are selected by voting using a three-dimensional accumulator; the candidate detection results are verified by angle coverage and false alarm count to obtain the effective detection results at each scale; the scales are fused sequentially from coarse to fine, and the fine-scale results are retained when overlapping to obtain the skeleton line and double-sided edge line of the transmission line; S400: Traverse the pixels on the skeleton line, perform local tangent fitting on each pixel, and generate a normal line based on the slope of the tangent line as the measurement direction of the ice thickness; solve for the intersection of the normal line and the two-sided edge lines, and calculate the local pixel width; combine the reference pixel width of the reference image with the actual diameter of the line, and convert the local pixel width into the actual ice thickness; correct the lateral coordinate deviation of the intersection point based on the mean value of edge extraction quantization based on the reference image, and recalculate and output the uneven ice thickness distribution along the line.

[0006] As a preferred embodiment of the present invention, step S100 specifically includes: Image acquisition is performed using two modes: timed triggering and temperature / humidity threshold triggering. After receiving the image, the server checks the image clarity, line integrity, and unobstructed conditions. If the image does not meet the standards, it sends a re-acquisition command to re-acquire the image. If the image still does not meet the standards, it sends an alarm signal to the maintenance terminal.

[0007] As a preferred embodiment of the present invention, in S200, the step of adaptively determining the RGB channel weighting coefficients includes: Calculate the global illumination intensity of an RGB image; The light intensity is divided into three intervals: low light, normal light and strong light. A set of RGB channel weighting coefficients is used for each light interval. A grayscale image is synthesized by weighting the RGB three channels according to the determined weighting coefficients.

[0008] As a preferred embodiment of the present invention, in S200, the step of dual-threshold edge detection includes: Calculate the mean and standard deviation of the gradient magnitude of the gradient image, and determine the high threshold and low threshold by linear combination of the two. Pixels with gradient magnitudes exceeding a high threshold are identified as strong edge pixels; Pixels with gradient magnitudes below the low threshold are directly discarded. Pixels whose gradient magnitude is between the high threshold and the low threshold are retained as weak edge pixels if there are strong edge pixels in their 3×3 neighborhood; otherwise, they are discarded. Binarize the pixels with strong edges and pixels with weak edges to generate a binary edge image.

[0009] As a preferred embodiment of the present invention, in S300, the step of setting multiple sets of scale parameters from coarse to fine includes: Adaptive boundary constraints are applied to the scale parameters based on the pixel width of the lines in the reference image. Set the minimum scale as the lower bound of the scale parameter set; Set a maximum scale, the value of which shall not exceed the product of the line pixel width and the safety constraint coefficient; Intermediate scale parameters are generated between the minimum and maximum scales using a linear or geometric progression strategy, thus constructing a complete multi-scale parameter set.

[0010] As a preferred embodiment of the present invention, in S300, the step of screening candidate detection results by voting using a three-dimensional accumulator includes: A three-dimensional accumulator is constructed based on the line's geometric parameters, with the line's center x-coordinate, y-coordinate, and radius as dimensions, and all initial values ​​are set to 0. Iterate through each edge point in the edge point set, iterate through all possible radius values ​​according to the standard equation of the line, solve for the corresponding line center coordinates, and accumulate votes for the corresponding positions of the three-dimensional accumulator. Local maxima filtering is performed on the 3D accumulator, and parameter groups whose cumulative values ​​exceed a set threshold are extracted as candidate detection results.

[0011] As a preferred embodiment of the present invention, in S300, the steps of angle coverage verification and false alarm count verification include: For each candidate detection result, calculate the polar angle corresponding to the boundary pixel, construct an angle histogram, and if the proportion of non-empty intervals exceeds the set threshold, then pass the angle coverage verification. The false alarm number (NFA) of the candidate detection results is calculated based on the number of boundary pixels that meet the conditions under gradient alignment, combined with the total number of pixels in the image and the line degrees of freedom; if the NFA is less than 1, it is verified by the false alarm number. Meanwhile, candidate detection results verified by angle coverage and false alarm count are used as valid detection results.

[0012] As a preferred embodiment of the present invention, in S400, the step of solving for the intersection of the normal and the bilateral edge lines includes: Substitute the pixel coordinates of the two edge lines one by one into the normal equation and set the distance threshold; If the Euclidean distance from an edge pixel to the normal is less than the distance threshold, it is determined to be a candidate intersection point; If a single edge line has multiple candidate intersection points, the candidate intersection point with the closest Euclidean distance to the skeleton line pixel is selected as the final intersection point; If a single edge line has only one candidate intersection point, then that point is directly taken as the final intersection point.

[0013] As a preferred embodiment of the present invention, in S400, the step of correcting the lateral coordinate deviation of the intersection point includes: Keep the vertical coordinates of the intersection points of the normal and the two edge lines unchanged; Based on the lateral coordinates of the intersection point of the normal line and the two-sided edge line of the edge extraction mean of the reference image, the quantization offset mean is obtained by statistical analysis of the edge extraction deviation of the reference image and is a fixed calibration value. The local pixel width is recalculated based on the corrected intersection coordinates, and the corrected actual icing thickness is calculated by combining the reference pixel width of the reference image with the actual diameter of the line.

[0014] This invention also proposes a transmission line icing thickness identification system based on an improved Hough transform, the system comprising: Image acquisition module: Acquires RGB images of the power transmission line through a camera and stores a baseline image under icing-free conditions; Image preprocessing module: adaptively determines the RGB channel weighting coefficients of the RGB image based on global illumination intensity, and synthesizes a grayscale image by weighting; after filtering and denoising, Sobel operator gradient calculation and non-maximum suppression, and double threshold edge detection, a binary edge image is generated; Line detection module: Gaussian convolution is performed on the binary edge image with multiple sets of scale parameters from coarse to fine to generate a multi-scale smooth image; the edge point set of each scale smooth image is extracted, and candidate detection results are selected by voting through a three-dimensional accumulator; the candidate detection results are verified by angle coverage and false alarm count to obtain the effective detection results at each scale; the results are fused sequentially from coarse to fine scale, and the fine scale results are retained when overlapping to obtain the skeleton line and double-sided edge lines of the transmission line; Thickness calculation module: It iterates through the pixels on the skeleton line, performs local tangent fitting on each pixel, and generates a normal line based on the tangent slope as the measurement direction of the ice thickness; it solves for the intersection of the normal line and the two-sided edge lines, and calculates the local pixel width; it combines the reference pixel width of the reference image with the actual diameter of the line, and converts the local pixel width into the actual ice thickness; it corrects the lateral coordinate deviation of the intersection point based on the mean value of edge extraction quantization based on the reference image, and recalculates and outputs the uneven ice thickness distribution along the line.

[0015] The beneficial effects of this invention are: 1. This invention proposes an adaptive RGB channel weighted grayscale method based on global illumination intensity. It adopts corresponding weighting coefficients for three regions: low illumination, normal illumination, and strong illumination, which effectively balances the impact of complex illumination conditions on image quality, significantly improves the grayscale difference between lines and background, and lays a high-quality image foundation for subsequent edge extraction.

[0016] 2. This invention employs an improved Hough transform method combining multi-scale Gaussian convolution with three-dimensional accumulator voting. It filters false detection results through a dual verification mechanism of angle coverage verification and false alarm count, and merges the results sequentially from coarse to fine scales while retaining fine-scale results during overlap. This effectively solves the problem of line edge breakage caused by uneven icing and achieves accurate extraction of skeleton lines and double-sided edge lines.

[0017] 3. This invention uses a method that combines local tangent fitting and normal measurement to calculate ice thickness. It introduces a coordinate compensation mechanism based on the quantized offset mean of the reference image to correct the edge extraction system error, thereby achieving refined quantification of the uneven ice thickness along the transmission line and significantly improving the accuracy and reliability of ice thickness measurement. Attached Figure Description

[0018] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart illustrating the method for identifying icing thickness on transmission lines based on the improved Hough transform of this invention. Figure 2 This is a schematic diagram of the image preprocessing process after acquisition according to the present invention; Figure 3 This is a schematic diagram of the multi-scale Hough transform structure of the present invention; Figure 4 This is a schematic diagram of the structure of the transmission line icing thickness identification system based on the improved Hough transform of the present invention. Detailed Implementation

[0019] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0020] Example 1: As Figure 1 As shown, the present invention provides a method for identifying the icing thickness of transmission lines based on an improved Hough transform, comprising: S100: Captures RGB images of power transmission lines using a camera and stores a baseline image of the line in an ice-free state; Furthermore, step S100 specifically includes: Image acquisition is performed using two modes: timed triggering and temperature / humidity threshold triggering. After receiving the image, the server checks the image clarity, line integrity, and unobstructed conditions. If the image does not meet the standards, it sends a re-acquisition command to re-acquire the image. If the image still does not meet the standards, it sends an alarm signal to the maintenance terminal.

[0021] Specifically, an unobstructed installation point is selected below the crossarm of the transmission tower to fix a remote high-definition camera, ensuring that the camera's field of view covers the entire transmission line segment. The camera uses RGB color imaging mode and transmits the collected RGB images to the remote monitoring center server in real time. The server stores the images according to the line number, collection time, and status identification rules, and records the environmental parameters at the time of collection simultaneously.

[0022] Image acquisition employs two modes: timed triggering and temperature / humidity threshold triggering. The timed triggering mode automatically acquires RGB images at fixed intervals; the temperature / humidity threshold triggering mode uses temperature and humidity sensor data as a basis and automatically triggers acquisition when environmental conditions meet the icing risk criteria. Preferably, acquisition is triggered when the temperature is not higher than 0℃ and the humidity is not lower than 80%RH.

[0023] When the transmission line is free of ice, multiple RGB images are acquired using a timed trigger mode. The frame with the highest clarity, intact line edges, and no obstructions is selected and stored as the reference image on the monitoring center server for subsequent ice thickness calculations. When the temperature and humidity threshold trigger conditions are met or the timed trigger time is reached, the camera automatically acquires a single RGB image while maintaining the same shooting parameters as the reference image.

[0024] After receiving the RGB image, the server automatically verifies the image clarity, line integrity, and unobstructed conditions. If the RGB image does not meet the standards, the server sends a re-acquisition command to the control module to re-acquire the RGB image; if it still does not meet the standards after re-acquisition, an alarm signal is sent to the maintenance terminal.

[0025] S200: The RGB channel weighting coefficients of the RGB image are adaptively determined according to the global illumination intensity, and a grayscale image is synthesized by weighting; after filtering and denoising, Sobel operator gradient calculation and non-maximum suppression, and double threshold edge detection, a binary edge image is generated; Specifically, the RGB images acquired by the S100 are sequentially subjected to adaptive weighted grayscale conversion, filtering and denoising, gradient calculation and non-maximum suppression, and dual-threshold edge detection to generate a binary edge image. The process is as follows: Figure 2 As shown.

[0026] Further, in S200, the step of adaptively determining the RGB channel weighting coefficients includes: Calculate the global illumination intensity of an RGB image; The light intensity is divided into three intervals: low light, normal light and strong light. A set of RGB channel weighting coefficients is used for each light interval. A grayscale image is synthesized by weighting the RGB three channels according to the determined weighting coefficients.

[0027] Specifically, the three-channel pixel data of the RGB image is read, and the image pixel matrix is ​​set as follows. The grayscale values ​​of the red, green, and blue channels of any pixel are respectively Image pixel height is The image pixel width is Calculate the global illumination intensity of an RGB image. : ; in, For global illumination intensity, The row coordinates of the pixel. The column coordinates of the pixel. Coordinates The grayscale values ​​of the red, green, and blue channels of a pixel. These represent the image's pixel height and width, respectively.

[0028] Based on global illumination intensity The lighting conditions were divided into three zones: low light, normal light, and strong light. For each lighting zone, a corresponding set of RGB channel weighting coefficients was applied. The weighting coefficients for the red, green, and blue channels were denoted as follows: , , Preferably, when A value no greater than 80 indicates low light intensity. Take 0.35, Take 0.55, Take 0.10; when A value greater than 80 and not greater than 180 is considered normal illumination. Take 0.299, Take 0.587, Take 0.114; when A value greater than 180 indicates strong light. Take 0.25, Take 0.60, Set the weighting factor to 0.15. Based on the determined weighting coefficients, weight the RGB three channels to synthesize a grayscale image using the following formula to obtain the coordinates. grayscale value of the pixel : ; in, coordinates The grayscale value of the pixel at that location. , , These are the weighting coefficients for the red, green, and blue channels, respectively. , , Coordinates The grayscale values ​​of the red, green, and blue channels of a pixel.

[0029] Gaussian filtering was applied to the grayscale image for noise reduction using a 3×3 Gaussian filter template. To smooth a grayscale image, iterate through each pixel and calculate the weighted average of that pixel and its 3×3 neighborhood pixels to obtain the coordinates. The grayscale value of the pixel after Gaussian filtering : ; in, coordinates The grayscale value of the pixel after Gaussian filtering. Gaussian filter template In position The weighting coefficient at the location, , This is the neighborhood offset. , coordinates The grayscale value of the pixel.

[0030] The horizontal gradient of each pixel is calculated using the Sobel operator. with vertical gradient The calculation formula is: ; ; in, coordinates The horizontal gradient of the pixel. coordinates The vertical gradient of a pixel. For the horizontal Sobel operator template at position The coefficient at the location, For the vertical Sobel operator template at position The coefficient at the location, coordinates The grayscale value of the pixel after Gaussian filtering. , This is the neighborhood offset.

[0031] Then calculate the gradient magnitude. With gradient direction : ; ; in, coordinates The gradient magnitude of the pixel. coordinates The gradient direction of the pixel. coordinates The horizontal gradient of the pixel. coordinates The vertical gradient at each pixel is calculated, and the gradient magnitude is normalized to generate a gradient image.

[0032] Non-maximum suppression is applied to the gradient image to adjust the gradient direction. The range of values ​​is from Adjusted to To obtain the standardized gradient direction : ; in, coordinates The normalized gradient direction of the pixel. coordinates The gradient direction of the pixel.

[0033] Will Divided into 4 sector regions, corresponding to 4 reference directions: The reference direction is 0° when the gradient falls within (0°, 22.5°) and (157.5°, 180°); 45° when it falls within (22.5°, 67.5°); 90° when it falls within (67.5°, 112.5°); and 135° when it falls within (112.5°, 157.5°). Two neighboring pixels are selected along the corresponding reference direction for each pixel. If the gradient magnitude of the current pixel... If the gradient magnitude is greater than that of two neighboring pixels, the pixel is retained; otherwise, its grayscale value is set to 0 to complete non-maximum suppression and generate the gradient image after edge reduction.

[0034] Further, in S200, the dual-threshold edge detection step includes: Perform double-threshold edge detection on the edge-trimmed gradient image. Calculate the mean gradient magnitude of all pixels in the gradient image. with standard deviation The high threshold is determined by a linear combination of the two. With low threshold Preferably, , Among them, 1.2 and 0.4 are the preferred coefficients. This represents the mean of the gradient magnitudes across all pixels in the gradient image. Let $\frac{ ... For high threshold, Low threshold. Gradient magnitude. Higher than The pixel is determined to be a strong edge pixel; gradient magnitude Below Pixels that are directly removed; gradient magnitude Between and For pixels between two points, if a strong edge pixel exists in its 3×3 neighborhood, it is retained as a weak edge pixel; otherwise, it is discarded. The strong and weak edge pixels are binarized, with the grayscale value of the strong edge pixel and the retained weak edge pixel set to 255, and the grayscale value of the remaining pixels set to 0, generating a binary edge image.

[0035] S300: Gaussian convolution is performed on the binary edge image with multiple sets of scale parameters from coarse to fine to generate a multi-scale smooth image; the edge point set of each scale smooth image is extracted, and candidate detection results are selected by voting using a three-dimensional accumulator; the candidate detection results are verified by angle coverage and false alarm count to obtain the effective detection results at each scale; the scales are fused sequentially from coarse to fine, and the fine-scale results are retained when overlapping to obtain the skeleton line and double-sided edge line of the transmission line; Specifically, using the binary edge image generated by S200 as input, the process sequentially performs multi-scale spatial construction and scale parameter setting, multi-scale Gaussian convolution and edge point extraction, 3D accumulator voting and candidate result screening, double verification, and multi-scale fusion to finally extract the skeleton line and bilateral edge lines of the transmission line. The specific process is as follows: Figure 3 As shown.

[0036] Furthermore, in S300, the step of setting multiple sets of scale parameters from coarse to fine includes: Set scale parameter group The scale control parameters satisfy Based on the pixel width of the lines in the reference image. Adaptive boundary constraints are applied to the scale parameters. A minimum scale is set. Preferably, as a lower bound of the scale parameter set. This is used to preserve high-frequency details at the edges and for precise positioning. Set the maximum scale. Its value does not exceed the line pixel width. With safety constraint factor The product of, i.e. ,in For safety constraint factors, preferably, This is used to prevent excessive smoothing from causing the upper and lower edges of the transmission line to merge. and Intermediate scale parameters are generated using linear or geometric progression strategies to construct a complete multi-scale parameter set.

[0037] Specifically, the binary edge image output by S200 As input, based on each scale parameter Multi-scale smoothed images are generated through Gaussian convolution operations. Gaussian convolution kernel. The expression is: ; in, For the first Gaussian convolution kernels at various scales , The coordinate variables are the coordinates of the Gaussian kernel function. For the first Each scale control parameter. Binary edge image. With Gaussian convolution kernel Perform two-dimensional spatial convolution to obtain the first... Smooth grayscale image at each scale : ; in, For the first Smooth grayscale images at various scales This is a binary edge image. For the first Gaussian convolution kernels at various scales This is a two-dimensional spatial convolution operator. The larger the value, the smoother the image, and the better it connects discrete fracture edges caused by uneven icing at a coarse scale and suppresses noise interference. For smoothed grayscale images at each scale... By setting a grayscale threshold to extract local maxima, the edge point set is re-extracted to obtain the first... Edge point set at each scale .

[0038] Furthermore, in S300, the step of filtering candidate detection results through voting using a three-dimensional accumulator includes: Based on the line's geometric parameters, a system is constructed with the line center x-coordinate as the reference point. y-axis and radius 3D accumulator for dimension All initial values ​​are set to 0. The parameter value range is limited to... , , ,in The original image width, The height of the original image. , These are the lower and upper bounds of the radius value, respectively. Preferably, Take 5 pixels. Pick Pixel.

[0039] Traversing the edge point set Each edge point in Iterate through all possible radius values. According to the standard equation of the line Solve for the corresponding center coordinates of the line. And accumulate votes at the corresponding positions of the three-dimensional accumulator: ; in, For three-dimensional accumulators in parameter groups The cumulative value at that point, For edge point set The Middle The coordinates of the edge points , These are the x and y coordinates of the line center, respectively. The radius is the line radius.

[0040] For three-dimensional accumulators Perform local maximum filtering and set the accumulator filtering threshold. : ; in, Filter thresholds for accumulators The accumulator threshold coefficient, The maximum accumulated value in the three-dimensional accumulator, preferably, Extracting the desired result The parameter set constitutes the candidate detection result set. .

[0041] Furthermore, in S300, the steps of angle coverage verification and false alarm count verification include: For candidate detection result set Each candidate result in Calculate the polar angle corresponding to its boundary pixels. Construct an angle histogram containing 36 intervals. If the proportion of non-empty intervals in the histogram exceeds the angle coverage verification threshold... If the candidate result is verified by angle coverage, then... The preferred angular coverage verification threshold .

[0042] Calculate the number of false alarms for each candidate result. : ; in, The number of false alarms The number of tests for the gradient alignment angle threshold is preferably [number]. Take 5. This represents the total number of pixels in the original image. For the degree of freedom of the line, preferably Take 3. This represents the total number of pixels at the candidate line boundary. The number of pixels required to satisfy the gradient alignment condition. The gradient alignment angle threshold is defined as the angle between the gradient alignment angle and the line normal vector being less than 100°. And gradient magnitude The number of pixels, of which , This represents the probability of single-pixel gradient alignment. The coefficients are binomial coefficients. To sum the variables, we need to iterate through them, taking the values ​​from... arrive integer values, for The probability that each pixel simultaneously satisfies the gradient alignment condition. Candidate results verified by angle coverage and false alarm count are then used as valid detection results at this scale, forming the valid detection result set. .

[0043] The effective detection results from each scale are fused sequentially in descending order of scale parameter. The largest scale is used as the starting point. The corresponding valid detection result set is used as the initial fusion result set. The effective detection results sets at each scale are collected sequentially. Fusion to .right Each test result Calculate its relationship with All test results distance : ; in, The distance between the two test results. , , for The x-coordinate, y-coordinate, and radius of the candidate results. , , for The x-coordinate, y-coordinate, and radius of the detection result. If And it does not satisfy the tangency condition, that is, the candidate result is not tangent to the given condition. The overlap ratio of the target results is less than And the distance between overlapping points is less than Then join in ;like If the tangency condition is met, then the detection results at smaller scales are retained, and from... Remove the original test results ,in The fusion distance threshold is preferably... Take 4.0 pixels. The threshold for the proportion of overlapping points is preferably... Take 15%, The threshold for the distance between overlapping points, preferably Take 2.0 pixels. After blending, This is the final result of the multi-scale Hough transform for line detection, from... Extract the center coordinates of each detection result By connecting the center coordinates in spatial order, the skeleton line of the transmission line is obtained through fitting. Based on the skeleton line, the radius parameters of each detection result are combined. The two edge lines of the transmission line are obtained by shifting them to both sides respectively.

[0044] S400: Traverse the pixels on the skeleton line, perform local tangent fitting on each pixel, and generate a normal line based on the slope of the tangent line as the measurement direction of the ice thickness; solve for the intersection of the normal line and the two-sided edge lines, and calculate the local pixel width; combine the reference pixel width of the reference image with the actual diameter of the line, and convert the local pixel width into the actual ice thickness; correct the lateral coordinate deviation of the intersection point based on the mean value of edge extraction quantization based on the reference image, and recalculate and output the uneven ice thickness distribution along the line.

[0045] Specifically, it iterates through every pixel on the skeleton line extracted by S300. Select adjacent vertical lines along the extension direction of the skeleton line. 100 pixels, together forming a group containing A set of local sliding windows for each pixel ,in The number of pixels on one side of the skeleton line in the sliding window. A set of local sliding windows The total number of pixels in the set. The least squares method is used to analyze the set. A straight line is fitted to all pixels in the matrix, and the equation of the fitted local tangent line is given by... The local tangent slope is solved by minimizing the sum of squared errors from each pixel to the fitted line. With intercept : ; ; in, The slope of the local tangent. This is the local tangent intercept. A set of local sliding windows The total number of pixels in the image. For set The Middle The coordinates of each pixel.

[0046] Based on the local tangent slope obtained from the solution , generated pixels The normal line, whose direction is the direction in which the ice thickness is measured at that local location. The normal equation is determined in three cases based on the direction of the tangent: when the slope of the local tangent... When the value is a non-zero finite value, the slope of the normal is The equation of the normal is When the tangent is horizontal, that is... When the normal is a vertical line, its equation is: When the tangent is vertical, that is... When the normal is horizontal, the equation is: .

[0047] Furthermore, in S400, the step of solving for the intersection of the normal and the two edge lines includes: The two edge lines extracted by S300 are denoted as the upper edge lines. and the bottom edge line Both edge lines are composed of continuous sets of pixels. For the upper edge line... pixel set and the bottom edge line pixel set The pixel coordinates were substituted one by one into the corresponding normal equation for verification, and a distance threshold was set. , The threshold for determining the Euclidean distance from an edge pixel to the normal is preferably set as follows: Take 1 to 2 pixels. If the Euclidean distance from an edge pixel to the normal is less than... If a pixel is found to be a candidate intersection point, then that pixel is considered a candidate intersection point. If a single edge line has multiple candidate intersection points, then all candidate intersection points and skeleton line pixels are compared. The Euclidean distance is used to select the closest candidate intersection point as the final intersection point; if a single edge line has only one candidate intersection point, that point is directly used as the final intersection point. The normal and upper edge line are obtained according to this rule. final intersection and the lower edge line final intersection .

[0048] With the final intersection point and Based on the baseline, the Euclidean distance formula is used to calculate the first... Local pixel width at each skeleton line pixel point : ; in, For the first The local pixel width at each skeleton line pixel point For the normal and the upper edge line final intersection coordinates Normal and lower edge line final intersection The coordinates are determined. All pixels along the skeleton line are traversed, and local tangent fitting, normal intersection point location, and pixel width calculation are performed point-by-point to obtain the set of local pixel widths along the icing transmission line. ,in This represents the total number of pixels on the skeleton line.

[0049] A baseline image, acquired using the same acquisition parameters as the icing image, is retrieved. The baseline image undergoes the same preprocessing and edge extraction operations as the S200 image, and the baseline pixel width of the transmission line in the non-icing state is calculated. And obtain the actual diameter of the transmission line. Regarding the first Local pixel width at each skeleton line pixel point The actual icing thickness at that location was calculated. : ; in, For the first The actual icing thickness at each skeleton line pixel. For the first The local pixel width at each skeleton line pixel point The base pixel width, This represents the actual diameter of the line. It is the set of local pixel widths. By converting each value individually, the set of actual icing thicknesses at various local locations along the route is obtained. .

[0050] Further, in S400, the step of correcting the lateral coordinate deviation of the intersection point includes: Due to factors such as image noise, edge extraction accuracy, and pixel discretization, directly calculated edge intersection points are prone to slight offsets. Therefore, a coordinate compensation algorithm based on the quantized offset mean is introduced to correct the intersection point coordinates. This preserves the intersection points... and vertical coordinate , The quantization offset mean is extracted based on the edges of the reference image, remaining unchanged. Correct the lateral coordinates of the intersection points, where The values ​​are obtained from the edge extraction deviation statistics of the reference image and are fixed calibration values. The corrected intersection point coordinates are as follows: and .

[0051] Since the same offset correction is applied to the lateral coordinates of both the upper and lower edge lines After correction, the horizontal distance between the two intersection points remains unchanged, therefore the corrected local pixel width remains unchanged. The calculation results are consistent with those before the correction: ; in, Extract the quantized offset mean for the edge. and These are the corrected coordinates of the intersection points of the upper and lower edges, respectively. Then, the corrected... Substituting into the actual icing thickness conversion formula, we obtain the actual icing thickness after coordinate compensation. It completes the output of the uneven ice thickness distribution along the line.

[0052] This embodiment achieves automated identification and precise quantification of icing thickness on transmission lines through a complete process from S100 to S400. Under complex lighting conditions and uneven icing scenarios, this embodiment effectively overcomes the shortcomings of traditional methods in terms of icing edge breakage, background noise interference, and pixel discretization errors, achieving high-precision non-contact measurement of icing thickness. This provides reliable data support for icing monitoring and maintenance decisions on transmission lines. Example

[0053] like Figure 4 As shown, this embodiment provides a transmission line icing thickness identification system based on an improved Hough transform. The system includes: Image acquisition module: Acquires RGB images of the power transmission line through a camera and stores a baseline image under icing-free conditions; Image preprocessing module: adaptively determines the RGB channel weighting coefficients of the RGB image based on global illumination intensity, and synthesizes a grayscale image by weighting; after filtering and denoising, Sobel operator gradient calculation and non-maximum suppression, and double threshold edge detection, a binary edge image is generated; Line detection module: Gaussian convolution is performed on the binary edge image with multiple sets of scale parameters from coarse to fine to generate a multi-scale smooth image; the edge point set of each scale smooth image is extracted, and candidate detection results are selected by voting through a three-dimensional accumulator; the candidate detection results are verified by angle coverage and false alarm count to obtain the effective detection results at each scale; the results are fused sequentially from coarse to fine scale, and the fine scale results are retained when overlapping to obtain the skeleton line and double-sided edge lines of the transmission line; Thickness calculation module: It iterates through the pixels on the skeleton line, performs local tangent fitting on each pixel, and generates a normal line based on the tangent slope as the measurement direction of the ice thickness; it solves for the intersection of the normal line and the two-sided edge lines, and calculates the local pixel width; it combines the reference pixel width of the reference image with the actual diameter of the line, and converts the local pixel width into the actual ice thickness; it corrects the lateral coordinate deviation of the intersection point based on the mean value of edge extraction quantization based on the reference image, and recalculates and outputs the uneven ice thickness distribution along the line.

[0054] It should be noted that the transmission line icing thickness identification system based on improved Hough transform provided in this embodiment of the invention is used to execute all the process steps of the transmission line icing thickness identification method based on improved Hough transform in the above embodiment 1. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0055] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for identifying icing thickness on transmission lines based on improved Hough transform, characterized in that, include: S100: Captures RGB images of power transmission lines using a camera and stores a baseline image of the line in an ice-free state; S200: The RGB channel weighting coefficients of the RGB image are adaptively determined according to the global illumination intensity, and a grayscale image is synthesized by weighting; after filtering and denoising, Sobel operator gradient calculation and non-maximum suppression, and double threshold edge detection, a binary edge image is generated; S300: Gaussian convolution is performed on the binary edge image with multiple sets of scale parameters from coarse to fine to generate a multi-scale smooth image; the edge point set of each scale smooth image is extracted, and candidate detection results are selected by voting using a three-dimensional accumulator; the candidate detection results are verified by angle coverage and false alarm count to obtain the effective detection results at each scale; the scales are fused sequentially from coarse to fine, and the fine-scale results are retained when overlapping to obtain the skeleton line and double-sided edge line of the transmission line; S400: Traverse the pixels on the skeleton line, perform local tangent fitting on each pixel, and generate a normal line based on the slope of the tangent line as the measurement direction of the ice thickness; solve for the intersection of the normal line and the two-sided edge lines, and calculate the local pixel width; combine the reference pixel width of the reference image with the actual diameter of the line, and convert the local pixel width into the actual ice thickness; correct the lateral coordinate deviation of the intersection point based on the mean value of edge extraction quantization based on the reference image, and recalculate and output the uneven ice thickness distribution along the line.

2. The method for identifying icing thickness of transmission lines based on improved Hough transform according to claim 1, characterized in that, The steps in S100 specifically include: Image acquisition is performed using two modes: timed triggering and temperature / humidity threshold triggering. After receiving the image, the server checks the image clarity, line integrity, and unobstructed conditions. If the image does not meet the standards, it sends a re-acquisition command to re-acquire the image. If the image still does not meet the standards, it sends an alarm signal to the maintenance terminal.

3. The method for identifying icing thickness of transmission lines based on improved Hough transform according to claim 1, characterized in that, In S200, the step of adaptively determining the RGB channel weighting coefficients includes: Calculate the global illumination intensity of an RGB image; The light intensity is divided into three intervals: low light, normal light and strong light. A set of RGB channel weighting coefficients is used for each light interval. A grayscale image is synthesized by weighting the RGB three channels according to the determined weighting coefficients.

4. The method for identifying icing thickness of transmission lines based on improved Hough transform according to claim 1, characterized in that, In S200, the dual-threshold edge detection step includes: Calculate the mean and standard deviation of the gradient magnitude of the gradient image, and determine the high threshold and low threshold by linear combination of the two. Pixels with gradient magnitudes exceeding a high threshold are identified as strong edge pixels; Pixels with gradient magnitudes below the low threshold are directly discarded. Pixels whose gradient magnitude is between the high threshold and the low threshold are retained as weak edge pixels if there are strong edge pixels in their 3×3 neighborhood; otherwise, they are discarded. Binarize the pixels with strong edges and pixels with weak edges to generate a binary edge image.

5. The method for identifying icing thickness of transmission lines based on improved Hough transform according to claim 1, characterized in that, In S300, the step of setting multiple sets of scale parameters from coarse to fine includes: Adaptive boundary constraints are applied to the scale parameters based on the pixel width of the lines in the reference image. Set the minimum scale as the lower bound of the scale parameter set; Set a maximum scale, the value of which shall not exceed the product of the line pixel width and the safety constraint coefficient; Intermediate scale parameters are generated between the minimum and maximum scales using a linear or geometric progression strategy, thus constructing a complete multi-scale parameter set.

6. The method for identifying icing thickness of transmission lines based on improved Hough transform according to claim 1, characterized in that, In S300, the step of filtering candidate detection results by voting using a three-dimensional accumulator includes: A three-dimensional accumulator is constructed based on the line's geometric parameters, with the line's center x-coordinate, y-coordinate, and radius as dimensions, and all initial values ​​are set to 0. Iterate through each edge point in the edge point set, iterate through all possible radius values ​​according to the standard equation of the line, solve for the corresponding line center coordinates, and accumulate votes for the corresponding positions of the three-dimensional accumulator. Local maxima filtering is performed on the 3D accumulator, and parameter groups whose cumulative values ​​exceed a set threshold are extracted as candidate detection results.

7. The method for identifying icing thickness of transmission lines based on improved Hough transform according to claim 1, characterized in that, In S300, the steps for angle coverage verification and false alarm count verification include: For each candidate detection result, calculate the polar angle corresponding to the boundary pixel, construct an angle histogram, and if the proportion of non-empty intervals exceeds the set threshold, then pass the angle coverage verification. The false alarm number (NFA) of the candidate detection results is calculated based on the number of boundary pixels that meet the conditions under gradient alignment, combined with the total number of pixels in the image and the line degrees of freedom; if the NFA is less than 1, it is verified by the false alarm number. Meanwhile, candidate detection results verified by angle coverage and false alarm count are used as valid detection results.

8. The method for identifying icing thickness of transmission lines based on improved Hough transform according to claim 1, characterized in that, In S400, the step of finding the intersection point of the normal and the two side edge lines includes: Substitute the pixel coordinates of the two edge lines one by one into the normal equation and set the distance threshold; If the Euclidean distance from an edge pixel to the normal is less than the distance threshold, it is determined to be a candidate intersection point; If a single edge line has multiple candidate intersection points, the candidate intersection point with the closest Euclidean distance to the skeleton line pixel is selected as the final intersection point; If a single edge line has only one candidate intersection point, then that point is directly taken as the final intersection point.

9. The method for identifying icing thickness of transmission lines based on improved Hough transform according to claim 1, characterized in that, In S400, the step of correcting the lateral coordinate deviation of the intersection point includes: Keep the vertical coordinates of the intersection points of the normal and the two edge lines unchanged; Based on the lateral coordinates of the intersection point of the normal line and the two-sided edge line of the edge extraction mean of the reference image, the quantization offset mean is obtained by statistical analysis of the edge extraction deviation of the reference image and is a fixed calibration value. The local pixel width is recalculated based on the corrected intersection coordinates, and the corrected actual icing thickness is calculated by combining the reference pixel width of the reference image with the actual diameter of the line.

10. A transmission line icing thickness identification system based on improved Hough transform, characterized in that, The system is used to execute the transmission line icing thickness identification method based on the improved Hough transform as described in any one of claims 1-9, and the system comprises: Image acquisition module: Acquires RGB images of the power transmission line through a camera and stores a baseline image under icing-free conditions; Image preprocessing module: adaptively determines the RGB channel weighting coefficients of the RGB image based on global illumination intensity, and synthesizes a grayscale image by weighting; after filtering and denoising, Sobel operator gradient calculation and non-maximum suppression, and double threshold edge detection, a binary edge image is generated; Line detection module: Gaussian convolution is performed on the binary edge image with multiple sets of scale parameters from coarse to fine to generate a multi-scale smooth image; the edge point set of each scale smooth image is extracted, and candidate detection results are selected by voting through a three-dimensional accumulator; the candidate detection results are verified by angle coverage and false alarm count to obtain the effective detection results at each scale; the results are fused sequentially from coarse to fine scale, and the fine scale results are retained when overlapping to obtain the skeleton line and double-sided edge lines of the transmission line; Thickness calculation module: It iterates through the pixels on the skeleton line, performs local tangent fitting on each pixel, and generates a normal line based on the tangent slope as the measurement direction of the ice thickness; it solves for the intersection of the normal line and the two-sided edge lines, and calculates the local pixel width; it combines the reference pixel width of the reference image with the actual diameter of the line, and converts the local pixel width into the actual ice thickness; it corrects the lateral coordinate deviation of the intersection point based on the mean value of edge extraction quantization based on the reference image, and recalculates and outputs the uneven ice thickness distribution along the line.

Images