Visual measurement method for sag of ice-coated conductor

Abstract

The invention discloses a power transmission line iced conductor sag vision measurement method based on deep learning and computer vision. According to the method, a power transmission line image is collected through a robot airborne camera, a conductor pixel region is accurately extracted by using a Deep-PIDNet semantic segmentation network which introduces cavity convolution and a deep-Bag fusion module, inverse perspective transformation is realized in combination with a homography matrix to eliminate image distortion, and a parabola model is fitted through an RANSAC algorithm to accurately calculate a sag value. The problems that a traditional method depends on multiple sensors and is low in precision and poor in environmental adaptability are solved, it is verified that the measurement error is small (smaller than or equal to + / -5%), and the method has the advantages of being low in cost and high in robustness and is suitable for automatic monitoring of severe cold icing scenes.

Description

Technical Field

[0001] The present invention belongs to the technical field of icing monitoring and intelligent detection of transmission lines, and specifically relates to a high-precision calculation method for the sag of ice-covered conductors based on visual measurement and deep learning. Background Art

[0002] As the lifeblood of the national economy, the power system often faces the risk of line breakage and tower collapse caused by icing in extremely low temperature environments. To ensure the safe operation of the power grid, vision-based transmission line icing monitoring and sag measurement have become a current research hotspot. Related technologies are mainly divided into two directions.

[0003] 1. Vision-based transmission line ice detection technology:

[0004] The current detection system includes three mainstream methods: binocular stereo vision, monocular vision, and sag vision measurement. Binocular vision technology uses drones to obtain stereo images and quantify ice thickness using 3D reconstruction algorithms. Monocular vision technology utilizes two technical approaches: traditional image processing and deep learning. For example, the Nusantika team uses a modified Canny operator combined with morphological filtering to detect ice edges. For the latter, Hao Yanpeng's team developed an improved Mask R-CNN network that integrates multi-scale features and is capable of intelligently classifying ice-covered tracks in complex backgrounds. It should be noted that the first two methods primarily provide qualitative analysis results and are currently still used as auxiliary monitoring methods.

[0005] 2. Conductor sag measurement method based on inverse perspective transformation:

[0006] Currently, the mainstream approach in this field uses drones equipped with visual sensors or lidar to collect line data, and then conducts 3D modeling and analysis using techniques such as inverse perspective transformation. Typical research includes: Huang Xinbo's team uses a multi-camera system to capture conductor feature points and calculate sag parameters through coordinate transformation. While this approach eliminates the need for sensor installation, it suffers from high hardware costs; Lu Qinxin's team constructs a cube grid model based on multi-angle drone aerial photography. While this improves detection efficiency, it is limited by the sparsity of conductor feature points, which affects the matching accuracy of the bundle adjustment method; Qian Jianguo's team innovatively uses deep learning to locate the spatial coordinates of the anti-vibration hammer to fit a parabolic function. While this solution is cost-effective, it is not applicable to special lines without anti-vibration hammers. Summary of the Invention

[0007] In order to solve the above problems, the present invention provides a method for visually measuring the sag of ice-covered conductors. The present invention mainly realizes high efficiency, low cost and high robustness of sag measurement, which is convenient for the calculation and research of ice load.

[0008] The technical solution proposed in the present invention is: a method for visually measuring the sag of an ice-covered conductor, comprising:

[0009] ① Use the image acquisition equipment equipped with the de-icing robot to capture the ice coverage of the transmission line in real time, and then transmit the obtained image data to the information processing system;

[0010] ② The algorithm uses the optimized Deep-PIDNet architecture to achieve semantic segmentation. By introducing the D-BasicBlock structural unit and the DeepBag feature fusion mechanism, it is specifically designed for the characteristics of transmission line images. It ultimately outputs the pixel-level segmentation results of the transmission line and generates the corresponding mask image.

[0011] ③ Using the pre-obtained homography transformation matrix, prepare to process the wire mask image;

[0012] ④ Use the homography matrix to perform reverse perspective transformation on the wire mask image, eliminate background interference in the mask image through noise reduction, perform wire connectivity operation and extract sparse feature point sets;

[0013] ⑤ Based on the discrete feature points of the transmission line, RANSAC is used to perform parabola fitting, analyze its geometric characteristics and construct a conductor model, and finally solve the sag value of the conductor.

[0014] Specifically, step 2 is implemented as follows: The D-BasicBlock module improves upon the BasicBlock architecture by introducing dilated convolutions, effectively expanding the receptive field of the convolution operation while maintaining the same number of parameters. This design enhances the network's ability to extract contextual features at different scales, making it particularly suitable for capturing long-range spatial dependencies in transmission line images, significantly improving segmentation accuracy.

[0015] The DeepBag fusion module effectively integrates visual information from different levels by integrating the deep features of the three branches and combining them with the shallow features of the P and D branches (P1, P2, D1, and D2). This design not only preserves the underlying details of the input data but also maintains the fine structural representation of distant wires. By complementing these multi-level features, it significantly improves the integrity and accuracy of small object detection.

[0016] Furthermore, the specific implementation of step ③ is as follows: a homography matrix is used to quantify the perspective deformation relationship between two images with different viewing angles.

[0017] Let the homography matrix be H, and its expression is:

[0018]

[0019] h 00 ,h 01 ,…h 22 are the unknown coefficients of the homography matrix;

[0020] It can be deduced that the homography matrix is only related to the rotation matrix, translation vector and camera internal parameters;

[0021] Through Zhang Zhengyou calibration method, after obtaining the intrinsic and extrinsic parameters of the camera, the homography matrix can be derived and the expression of its specific coefficients can be determined:

[0022]

[0023] The inverse perspective transformation process can be expressed as:

[0024]

[0025] Where S is the scale ratio;

[0026] According to the homography matrix, the inverse perspective transformation image is obtained, and the two-dimensional shape of the wire in its world plane is mapped from the original image.

[0027] Furthermore, the specific implementation of step ④ is as follows: first, the noise interference in the wire mask image is filtered out, then the inverse perspective transformation is performed through the homography matrix, and then the wire threading operation is performed on the transformed image to finally obtain the sparse feature point set of the wire.

[0028] Furthermore, the specific implementation of step ⑤ is as follows: When performing parabola fitting on the feature points based on the RANSAC algorithm, a plane rectangular coordinate system is established with the suspension point on the left side of the transmission line as the coordinate origin, where the horizontal right direction is the positive direction of the x-axis and the vertical downward direction is the positive direction of the y-axis. In this coordinate system, the parabola mathematical model of the transmission line can be expressed as follows:

[0029] y=a0+a1x+a2x 2

[0030] Where:

[0031] y is the vertical coordinate of a point on the transmission line;

[0032] x is the horizontal coordinate of a point on the transmission line;

[0033] a0, a1, and a2 are unknown coefficients;

[0034] Given the gear spacing L and the hanging point heights h1 and h2

[0035] The coordinates of the two hanging points can be expressed as point A (0, 0) and point B (L, |h1-h2|). The line connecting the two hanging points of the transmission line pole can be expressed by the formula:

[0036]

[0037] The sag f at the midpoint of the gear can be obtained by calculation:

[0038]

[0039] The beneficial effects of the present invention are as follows: through the improved Deep-PIDNet semantic segmentation network, combined with the D-BasicBlock module and the DeepBag fusion module, the long-distance dependencies and detailed features of the transmission line are effectively captured, the segmentation accuracy is significantly improved (MIoU reaches 83.33%), and the anti-interference ability is strong, which is suitable for ice-covered conductor detection in complex backgrounds; only the robot's onboard camera is required to collect images, and there is no need to install additional devices on the conductor, which reduces the implementation cost and operational risk, and is particularly suitable for severe cold environments; the inverse perspective transformation of the homography matrix is used to eliminate perspective distortion, and the parabola model is fitted with the RANSAC algorithm to accurately calculate the sag value, and the measurement error is controlled within ±5%, meeting engineering requirements; the method has no obvious restrictions on the type of icing (snow, ice), and its wide applicability has been verified by laboratory and simulation data, providing a reliable basis for icing load analysis; the method provides an efficient, low-cost and high-precision solution for transmission line safety monitoring, and has significant engineering application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 This is a system flow chart of the present invention

[0041] Figure 2 A comparison diagram of BasicBlock (top) and D-BasicBlock (bottom);

[0042] Figure 3 Schematic diagram of the comparison between the original Bag and DeepBag;

[0043] Figure 4 Schematic diagram of transmission line sag; DETAILED DESCRIPTION

[0044] Example: A visual measurement method for ice-covered conductor sag. For operation instructions, please refer to Figure 1 , the de-icing robot's onboard camera is used to obtain real-time images of ice-covered transmission lines, and the image information is transmitted to the processor. The processor uses the improved Deep-PIDNet semantic segmentation network to segment the transmission line pixels in the image according to the presentation characteristics of the transmission line and form a conductor mask map. Based on the conductor mask map, the pre-calculated homography matrix is used to perform inverse perspective transformation on the conductor mask map, and the interference noise in the processed conductor mask map is filtered out. The conductor is threaded to obtain the sparse feature points of the conductor. Based on the sparse feature points of the conductor, the morphological characteristics of the transmission line are analyzed, and the RANSAC curve fitting algorithm is used to perform parabolic curve fitting on the feature points to model the conductor and obtain the sag.

[0045] The core steps are as follows:

[0046] 1. PIDNet semantic segmentation network:

[0047] D-BasicBlock module

[0048] Transmission lines are long, continuous lines with large spans in images. Based on these characteristics, the D-BasicBlock module was proposed. This module improves upon the BasicBlock architecture by introducing dilated convolutions, effectively expanding the receptive field of the convolution operation while maintaining the same number of parameters. This design enhances the network's ability to extract contextual features at different scales, making it particularly suitable for capturing long-range spatial dependencies in transmission line images, significantly improving segmentation accuracy.

[0049] DeepBag fusion module

[0050] In images, transmission lines vary greatly in scale, and distant conductors occupy a very small proportion. After multiple convolutions, they are easily overwhelmed by background information, leading to missed detections. However, their edge features stand out against the surrounding background. Therefore, to address this issue and incorporate the characteristics of transmission lines, this paper proposes a new DeepBag fusion module based on the original network's Bag fusion module.

[0051] The DeepBag fusion module effectively integrates visual information from different levels by integrating the deep features of the three branches and combining the shallow features of the P and D branches (P1, P2, D1, and D2). This design not only preserves the underlying details of the input data but also maintains the fine structural representation of long-distance wires. Through the complementary nature of multi-level features, it significantly improves the integrity and accuracy of small object detection.

[0052] 2. Inverse perspective transformation

[0053] Performing an inverse perspective transformation on the wire mask image according to the homography matrix, filtering out interference noise in the processed wire mask image, performing a threading operation on the wires, and obtaining sparse feature points of the wires;

[0054] That

[0055] The homography matrix is used to describe the perspective transformation relationship between two images. Let the homography matrix be H, and its expression is:

[0056]

[0057] h 00 ,h 01 ,…h 22 are the unknown coefficients of the homography matrix;

[0058] It can be deduced that the homography matrix is only related to the rotation matrix, translation vector and camera internal parameters;

[0059] Through Zhang Zhengyou calibration method, after obtaining the intrinsic and extrinsic parameters of the camera, the specific coefficients of the homography matrix can be calculated as follows:

[0060]

[0061] The inverse perspective transformation process can be expressed as:

[0062]

[0063] Where S is the scale ratio;

[0064] According to the homography matrix, the inverse perspective transformation image is obtained, and the two-dimensional shape of the wire in its world plane is mapped from the original image.

[0065] 3. Fitting parabolic curve to model the wire

[0066] The RANSAC curve fitting algorithm is used to fit the characteristic points. The left end hanging point of the transmission line is taken as the origin, the right direction is the positive x-axis, and the downward direction is the positive y-axis. The world coordinate system of the plane where the transmission line is located is used. The parabola model function of the transmission line is expressed as follows:

[0067] y=a0+a1x+a2x 2

[0068] Where y is the vertical coordinate of a point on the transmission line, x is the horizontal coordinate of a point on the transmission line, and a0, a1, and a2 are unknown coefficients.

[0069] 4. Calculate sag

[0070] Please refer to Figure 4 Analysis of the calculation of sag in actual situations:

[0071] Given the gear spacing L and the hanging point heights h1 and h2

[0072] The coordinates of the two hanging points can be expressed as point A (0, 0) and point B (L, |h1-h2|). The line connecting the two hanging points of the transmission line pole can be expressed by the formula:

[0073]

[0074] The sag f at the midpoint of the gear can be obtained by calculation:

[0075]

Claims

1. A method for visually measuring sag of ice-covered conductors, characterized in that: include: The de-icing robot uses its onboard camera to capture real-time images of ice-covered transmission lines and transmits the image information to the processor. The processor uses an improved Deep-PIDNet semantic segmentation network: it adds a D-BasicBlock module and a DeepBag fusion module. Based on the characteristics of the transmission lines, it segments the pixels of the transmission lines in the image and forms a conductor mask. Based on the wire mask map, a pre-calculated homography matrix is used; Performing an inverse perspective transformation on the wire mask image according to the homography matrix, filtering out interference noise in the processed wire mask image, performing a threading operation on the wires, and obtaining sparse feature points of the wires; Based on the sparse feature points of the conductor, the morphological characteristics of the transmission line are analyzed, and the RANSAC curve fitting algorithm is used to perform parabolic curve fitting on the feature points to model the conductor and obtain the sag; The visual sag measurement method for ice-covered conductors according to claim 1 is characterized in that the D-BasicBlock module is improved based on the BasicBlock structure and effectively expands the receptive field of the convolution operation while keeping the number of parameters unchanged by introducing the dilated convolution technology. This design enhances the network's ability to extract contextual features of different scales and is particularly suitable for capturing long-distance spatial dependency features in transmission line images, thereby significantly improving segmentation accuracy. The visual measurement method for sag of ice-covered conductors according to claim 1 is characterized in that the DeepBag fusion module effectively integrates the deep features of the three branches and combines the shallow features of the P branch and the D branch (P1, P2, D1, D2), thereby effectively fusing visual information at different levels. This design not only retains the underlying details of the input data, but also maintains the fine structural representation of long-distance conductors, and significantly improves the integrity and accuracy of small target detection through multi-level feature complementarity.

2. The method for visually measuring the sag of an ice-covered conductor according to claim 1, characterized in that: The homography matrix is used to describe the perspective transformation relationship between two images; Let the homography matrix be H, and its expression is: h 00 ,h 01 ,…h 22 are the unknown coefficients of the homography matrix; It can be deduced that the homography matrix is only related to the rotation matrix, translation vector and camera internal parameters; Through Zhang Zhengyou calibration method, after obtaining the intrinsic and extrinsic parameters of the camera, the specific coefficients of the homography matrix can be calculated as follows:

3. The method for visually measuring the sag of an ice-covered conductor according to claim 1, characterized in that: The inverse perspective transformation process can be expressed as: Where S is the scale ratio; According to the homography matrix, the inverse perspective transformation image is obtained, and the two-dimensional shape of the wire in its world plane is mapped from the original image.

4. The method for visually measuring sag of ice-covered conductors according to claim 1, characterized in that: The interference noise in the wire mask image that has been screened out is threaded through the image using the homography matrix and inverse perspective transformation to obtain sparse feature points of the wires.

5. The method for visually measuring sag of ice-covered conductors according to claim 1, characterized in that: The RANSAC curve fitting algorithm is used to fit the characteristic points. The left end hanging point of the transmission line is taken as the origin, the right direction is the positive x-axis, and the downward direction is the positive y-axis, which is used as the world coordinate system of the plane where the transmission line is located. The parabola model function of the transmission line is expressed as follows: y=a0+a1x+a2x 2 Where: y is the vertical coordinate of a point on the transmission line; x is the horizontal coordinate of a point on the transmission line; a0, a1, and a2 are unknown coefficients; Given the gear spacing L and the hanging point heights h1 and h2; The coordinates of the two hanging points can be expressed as point A (0, 0) and point B (L, |h1-h2|). The line connecting the two hanging points of the transmission line pole can be expressed by the formula: The sag f at the midpoint of the gear can be obtained by calculation:

Images