Method, device and equipment for detecting quality of conductor of power transmission line

Abstract

This invention discloses a method, apparatus, and equipment for inspecting the quality of transmission line conductors, relating to the field of transmission line inspection technology. The method includes the following steps: based on multispectral inspection data of the transmission line, identifying the conductor outline and discretizing the conductor into several observation areas; calculating the surface heat loss and axial conduction heat of each observation area based on the surface temperature, ambient temperature, and thermal properties of the conductor, inverting the internal heating power of each observation area, and filtering abnormal areas using a preset judgment threshold; extracting the inter-frame displacement sequence of the multispectral inspection data relative to a preset background, calculating the cross-modal phase correlation of the inter-frame displacement sequence, and then correcting the coordinates of the abnormal areas; extracting the appearance defect features of the corrected abnormal areas, and generating quality inspection results based on the internal heating power. This invention addresses the technical problems of misjudging internal defects in conductors and the difficulty in accurately aligning multimodal data in existing technologies.

Description

Technical Field

[0001] This invention relates to the field of power transmission line testing technology, and more specifically, to methods, apparatus and equipment for testing the quality of power transmission line conductors. Background Technology

[0002] Existing solutions for detecting defects in power transmission and distribution line conductors mostly employ drones equipped with multimodal sensors for inspection, in order to meet the need for dynamic monitoring of line conditions in complex field environments.

[0003] For example, patent application CN120219390A discloses an automated intelligent defect analysis method based on UAV image acquisition for power transmission and distribution lines. This method acquires infrared thermal imaging time-series data and multispectral imaging data from a UAV, calculates the temperature gradient direction on the conductor surface to extract the hot spot diffusion path, and simultaneously locates the boundary of the surface oxide region based on multispectral data. Finally, it performs spatial topological matching between the hot spot migration direction and the oxide boundary to analyze defect correlation.

[0004] However, the aforementioned existing technologies still have some technical problems in practical applications: On the one hand, existing technologies mostly rely on the spatial distribution characteristics and dynamic tracking of the temperature gradient on the conductor surface. When the transmission line is in a complex meteorological environment, the surface heat migration distribution is easily affected by external wind load disturbances or uneven local convective heat dissipation conditions, resulting in a deviation between the surface observation characteristics and the actual internal heat source state, making it difficult to directly and quantitatively characterize the actual heating intensity of hidden defects inside the conductor; On the other hand, in the cross-modal feature fusion stage, existing technologies mainly rely on topological matching based on the extracted spatial contour trajectory or morphological distribution. For slender and weakly textured overhead conductor targets, when they undergo irregular micro-movements due to environmental influences during inspection, dynamic misalignment at the bottom pixel level is easily generated between multispectral images, resulting in the inability to accurately align the subsequently extracted appearance defect features with the local thermodynamic features. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a method for detecting the quality of transmission line conductors. This method uses the inversion of internal heating power as a basis for anomaly screening and utilizes the cross-modal phase correlation of inter-frame displacement in multispectral data for coordinate correction. This addresses the technical problems of misjudging internal defects of conductors due to environmental interference and the difficulty in accurately aligning multimodal data caused by conductor micro-motion in existing UAV inspections.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A method for quality inspection of transmission line conductors includes the following steps: Based on multispectral inspection data of transmission lines, the outline of the conductor is identified and the conductor is discretized into several observation areas; Based on the surface temperature, ambient temperature, and thermal properties of the conductor in each observation area, the surface heat loss and axial heat conduction in each observation area are calculated, the internal heating power of each observation area is obtained by inversion, and abnormal areas are screened in combination with the preset judgment threshold. After extracting the inter-frame displacement sequence of the multispectral inspection data relative to a preset background and calculating the cross-modal phase correlation of the inter-frame displacement sequence, coordinate correction is performed on the abnormal region. Extract the appearance defect features of the abnormal area after correction, and combine them with the internal heating power to generate quality inspection results.

[0007] In a preferred embodiment, the multispectral inspection data includes visible light data and infrared data; the identification of the conductor contour includes: extracting a first axial trajectory and a second axial trajectory representing the direction of the conductor based on the visible light data and the infrared data respectively; and extracting the connected domain boundaries of the corresponding conductor regions based on the first axial trajectory and the second axial trajectory respectively to obtain the visible light conductor contour and the infrared conductor contour.

[0008] In a preferred embodiment, the steps for generating the first axial trajectory and the second axial trajectory include: extracting candidate edge point sets from visible light data and infrared data in the initial frame, respectively; and performing curve fitting on the candidate edge point sets based on the spatial geometric features of the overhead power transmission line to obtain the first axial trajectory and the second axial trajectory.

[0009] In a preferred embodiment, the formulas for calculating the surface heat loss and axial heat conduction are as follows: , , in, For heat loss on the surface, The heat transfer coefficient of the conductor surface. The outer surface area of ​​the observation area. For the first infrared mode in multispectral inspection data Surface temperature of each observation area Ambient temperature; For axial heat conduction, Thermal conductivity, Let be the cross-sectional area of ​​the conductor. The discretization step size of the observation area. , These are the first infrared modes. The, the Surface temperature of the observation area.

[0010] In a preferred embodiment, the screening of abnormal regions includes: calculating a local relative change rate based on the difference between the heating power inside the target observation area and the heating power inside the adjacent observation area; and comparing the local relative change rate with a preset judgment threshold to screen abnormal regions.

[0011] In a preferred embodiment, the coordinate correction of the abnormal region includes: extracting the inter-frame displacement sequences corresponding to visible light data and infrared data in the multispectral inspection data; calculating the cross-power spectrum of the visible light inter-frame displacement sequence and the infrared inter-frame displacement sequence within the same observation area, and extracting the offset parameter corresponding to the phase correlation peak as a displacement compensation parameter; and correcting the coordinates of the abnormal region based on the displacement compensation parameter.

[0012] In a preferred embodiment, the step of extracting the inter-frame displacement sequence corresponding to the visible light data and infrared data in the multispectral inspection data includes: extracting the absolute displacement sequence of each observation area in the visible light data and infrared data respectively; and subtracting the absolute displacement sequence from the synchronously extracted preset background displacement sequence to obtain the inter-frame displacement sequence corresponding to the visible light data and infrared data.

[0013] In a preferred embodiment, generating the quality inspection result includes: converting the local relative change rate of the abnormal region into an anomaly score based on a pre-built nonlinear mapping relationship; mapping the corrected coordinates of the abnormal region to the visible light data of the multispectral inspection data, and extracting the appearance defect features of the corresponding region; inputting the anomaly score and appearance defect features into a trained defect classification model to obtain a quality inspection result including defect type and risk level.

[0014] A transmission line conductor quality inspection device includes the following modules: a contour recognition module, used to identify the conductor contour and discretize the conductor into several observation areas based on multispectral inspection data of the transmission line; a heat generation analysis module, used to calculate the surface heat loss and axial conduction heat of each observation area based on the surface temperature, ambient temperature, and thermal property parameters of the conductor, invert the internal heating power of each observation area, and filter abnormal areas in combination with a preset judgment threshold; a coordinate correction module, used to extract the inter-frame displacement sequence of the multispectral inspection data relative to a preset background, calculate the cross-modal phase correlation of the inter-frame displacement sequence, and then perform coordinate correction on the abnormal areas; and a quality inspection module, used to extract the appearance defect features of the corrected abnormal areas and generate quality inspection results in combination with the internal heating power.

[0015] An electronic device includes: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of any of the methods.

[0016] The technical effects and advantages of the transmission line conductor quality inspection method of the present invention are as follows: 1. This invention calculates the surface heat loss and axial conduction heat of each observation area, inverts the internal heating power of each observation area, and uses this to screen abnormal areas. Under the framework of local thermodynamic equilibrium, it realizes the physical mapping from surface thermal appearance to the actual internal heating intensity, effectively decouples the interference of complex field meteorological environment on the two-dimensional temperature field of conductor surface, reduces the risk of defect misjudgment caused by external environmental disturbance, and improves the quantitative analysis accuracy and robustness of transmission line conductor quality inspection.

[0017] 2. This invention extracts the inter-frame displacement sequence of multispectral inspection data relative to a preset background and calculates the cross-modal phase correlation of the inter-frame displacement sequence to correct the coordinates of abnormal areas. It transforms the dynamic physical motion of overhead conductors caused by wind deflection during inspection into effective temporal constraint features for cross-modal registration. This overcomes the problem of static matching misalignment caused by feature sparsity in the spatial domain for slender and weakly textured conductor targets, and achieves accurate spatial alignment of the underlying multimodal data. This further improves the accuracy and intelligence level of transmission line detection in complex natural environments. Attached Figure Description

[0018] Figure 1 This is a schematic flowchart of a method for detecting the quality of transmission line conductors provided in an embodiment of the present invention.

[0019] Figure 2 A three-dimensional surface diagram of the cross-modal phase correlation peak provided in an embodiment of the present invention.

[0020] Figure 3 This is a schematic diagram of the confusion matrix of the defect classification model provided in an embodiment of the present invention.

[0021] Figure 4 This is a schematic diagram of a power transmission line conductor quality testing device provided in an embodiment of the present invention.

[0022] Figure 5 This is a structural block diagram of an exemplary electronic device provided for implementing embodiments of the present disclosure. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0024] Example 1, Figure 1The present invention provides a method for quality inspection of transmission line conductors, comprising the following steps: Step 1: Based on the multispectral inspection data of the transmission line, identify the conductor outline and discretize the conductor into several observation areas.

[0025] 1.1 Acquire and preprocess multispectral inspection data of transmission lines, including: During the drone's flight along a pre-set inspection route, multispectral inspection data, including visible light and infrared video streams, is collected in real time using onboard heterogeneous sensors, while simultaneously recording the ambient temperature. To ensure data time synchronization, a unified timestamp and GNSS position coordinates are injected into the collected multispectral image frames to construct multimodal aligned data, which is then stored in a local cache.

[0026] After data synchronization is completed, the initial frames of the multispectral inspection data are preprocessed: for visible light data, a Gaussian filter with a preset kernel scale (e.g., 5×5 pixels) is used to perform spatial smoothing to suppress environmental speckle noise, and nonlinear mapping (e.g., using Gamma correction, setting the Gamma value to 0.45) is used to enhance the image brightness range to improve contrast; for infrared data, a two-dimensional median filter is used to perform spatial domain noise reduction to smooth sensor hotspot noise and reduce instantaneous temperature fluctuation interference.

[0027] 1.2 Identify the conductor profile, including: 1.2.1 Extract candidate edge point sets from the visible light data and infrared data in the initial frame respectively. The Canny edge detection algorithm is adopted, taking the initial frames of preprocessed visible light and infrared data as input. By calculating the magnitude and direction of the global pixel grayscale gradient of the image and performing non-maximum suppression to remove false edges, the algorithm then uses pre-set high and low thresholds for edge extraction to connect edges. That is, strong edge points with gradient magnitudes greater than the high threshold are retained, and non-edge points with gradient magnitudes less than the low threshold are removed. For weak edge points with gradient magnitudes between the high and low thresholds, they are retained only if they are connected to strong edge points. Finally, a candidate edge point set is formed for the visible light and infrared data. Since the image pixel grayscale value is a dimensionless relative brightness level (usually ranging from 0 to 255), and considering the difference in imaging contrast between visible light and infrared images, different high and low thresholds are set for visible light and infrared data respectively. For example, it is recommended that the high and low thresholds for visible light data be set to 150 and 50 respectively, and for infrared data, which has more blurred edges, the high and low thresholds are recommended to be set to 100 and 30 respectively.

[0028] 1.2.2 Curve fitting is performed on the candidate edge point set to obtain the first axial trajectory and the second axial trajectory representing the direction of the conductor, as follows: Considering the drooping spatial geometry of overhead transmission lines due to their own weight and span, and the geometric distortion that occurs when a physical catenary in three-dimensional space is projected onto the focal plane of a two-dimensional camera through perspective, the system constructs a parabolic equation reflecting the projection shape of the conductor in the two-dimensional image plane to ensure the real-time convergence and robustness of the algorithm. The specific expression of the parabolic equation is as follows: (1) in, and These represent the horizontal and vertical pixel coordinates of the conductor pixel in the two-dimensional image plane, respectively. The coefficient of the quadratic term; The coefficient of the linear term; This is a constant term.

[0029] A nonlinear least squares method is used to fit the parameters of the parabolic equation. Specifically, a loss function is established based on the sum of squares of the orthogonal physical distances from each pixel in the candidate edge point set to the curve corresponding to the equation. The Levenberg-Marquardt (LM) algorithm is used to combine the coordinate data of the candidate edge point set with the initial parameters of a quadratic polynomial (specifically, extracting the endpoint with the smallest x-coordinate, the endpoint with the largest x-coordinate, and the midpoint with the middle x-coordinate from the candidate edge point set, substituting the coordinates of these three feature pixels into the quadratic polynomial approximation equation to form a system of linear equations, which can then be analytically solved to obtain the coefficients A and B). Using the initial estimated value of C as input, a dynamic damping factor is inherently introduced on the main diagonal of the approximate Hessian matrix (the initial damping factor can be set to 0.01). In each iteration, the value of the damping factor is automatically adjusted according to the decrease of the loss function. Taking advantage of the control characteristics of the algorithm tending to gradient descent when the value of the damping factor is too large and smoothly transitioning to Gauss-Newton method when the value is too small, the solution process can adaptively switch between global robust search and local fast convergence, thereby stably and quickly iteratively updating the undetermined parameters of the equation (i.e., the coefficients of the quadratic term, the coefficients of the linear term, and the constant term).

[0030] After the parameter iterative solution process of the above algorithm converges, the target parameters obtained by solving based on the visible light candidate point set and the infrared candidate point set are substituted back into the equation, so as to fit the first axial trajectory and the second axial trajectory corresponding to the visible light image and the infrared image respectively.

[0031] 1.2.3 Using the first axial trajectory and the second axial trajectory as references, the connected domain boundaries of the corresponding conductor regions are extracted to obtain the visible light conductor profile and the infrared conductor profile, as follows: Using the first axial trajectory and the second axial trajectory as the center baselines, a search neighborhood is constructed by extending a preset pixel width (e.g., ±20 pixels) to both sides of the curve normal. Within the search neighborhood, pixels with grayscale changes greater than a preset contour change threshold are extracted pixel by pixel along the normal direction as boundary points. The boundary points are connected to form the boundary of the connected domain, thereby obtaining the visible light guide contour and the infrared guide contour respectively.

[0032] It should be noted that the preset contour change threshold is a dynamic gray-level jump threshold adaptively calculated within a local search window based on the Otsu method. Specifically, taking the gray-level image of the local search window (e.g., 15×15 pixels) where the current pixel is located as input, the candidate gray levels from 0 to 255 are traversed, and the inter-class variance of the background pixel class and the target pixel class divided by the candidate gray level is calculated respectively. The candidate gray level that maximizes the inter-class variance is selected as the dynamic gray-level jump threshold specifically for the local search window.

[0033] 1.3 The traverse is discretized into several observation regions, as follows: Extract the first and second connected domains of the visible light and infrared guide wires, respectively. Using attitude information such as yaw, pitch, and roll angles obtained from the UAV platform and the intrinsic and extrinsic parameter matrices of the multispectral camera, and combining the first and second axial trajectories fitted in step 1.2.2, establish a longitudinal projection plane in three-dimensional space based on the actual direction of the guide wire. Subsequently, calculate the homography matrix from each camera coordinate system to the projection plane of the longitudinal profile, and perform perspective projection transformation based on the homography matrix on the first and second connected domains of the guide wires, so that the pixel spacing in the projection plane and the step size in the real physical space are approximately proportionally linearly mapped.

[0034] After completing the projection mapping, the first and second wire connected regions after projection are equidistantly divided along the trajectory direction at a preset fixed pixel step size (e.g., 50 pixels, which is equivalent to 0.5 meters in physical space under the known mapping transformation relationship), to obtain a series of visible light observation regions corresponding to the visible light data and a series of infrared observation regions corresponding to the infrared data.

[0035] This step uses candidate edge point sets extracted from visible light and infrared data to fit and generate a first axial trajectory and a second axial trajectory, which are then used to constrain the search domain of the connected domain boundary. This effectively eliminates interference from complex objective backgrounds, improves the recognition accuracy of visible light and infrared wire contours, and provides a reliable spatial benchmark for discretization and thermal balance calculation.

[0036] Step 2: Based on the surface temperature, ambient temperature, and thermal properties of the conductor in each observation area, calculate the surface heat loss and axial heat conduction in each observation area, invert the internal heating power of each observation area, and filter out abnormal areas using a preset judgment threshold, including: 2.1 Calculate the surface heat loss and axial heat conduction in each observation area, and then invert the internal heating power of each observation area as follows: For each infrared observation region formed after discretization, the infrared radiation grayscale values ​​of all pixels in the corresponding connected domain of the observation region are extracted, and the infrared radiation grayscale values ​​of the pixels are inverted into the true physical temperature based on the physical temperature measurement formula, which is as follows: (2) in, For the first The actual physical temperature of each pixel, in degrees Celsius (°C); The ambient temperature is expressed in degrees Celsius (°C). For the first The infrared radiation grayscale value of each pixel; The average infrared radiation grayscale value of the environmental background area; The preset emissivity parameter for the conductor surface is used, for example, the empirical value of 0.85 for alumina surface. The pre-calibrated temperature conversion factor for the infrared camera hardware is, for example, 0.04℃ per gray level.

[0037] The arithmetic mean of the true physical temperatures of all pixels within each infrared observation area is calculated and taken as the surface temperature of that observation area. Based on the surface temperature of the observation area, the internal heat generation power is inverted. Since the thermodynamic behavior of the conductor element (i.e., the observation area) in an actual operating power transmission network follows the law of conservation of steady-state energy, the internal heat generation power is numerically equal to the sum of surface heat loss to the external environment and axial heat loss to adjacent observation areas. The formulas for calculating the surface heat loss and axial heat loss are as follows: (3) (4) in, Heat loss from the surface is measured in watts (W). The heat transfer coefficient of the conductor surface is used to comprehensively characterize the convective and radiative heat dissipation capacity of the conductor surface. Its value is obtained by calling the system's preset standard empirical constant for a light wind environment (e.g., 12W / (m²·K)). The outer surface area of ​​the observation region is equal to the perimeter of the conductor cross-section and the discretization step size of the observation region. The product of , in square meters (m²). For the infrared mode, the first The surface temperature of each observation area is expressed in degrees Celsius (°C). For axial heat conduction, the unit is watt (W). This represents the cross-sectional area of ​​the conductor, expressed in square meters (m²). The discretization step size of the observation area is specifically the actual physical step size equivalent to the fixed pixel step size when the connected domain of the conductor is divided into equal intervals in step 1, and the unit is meters (m). , These are the first infrared modes. The, the The surface temperature of each observation area is expressed in degrees Celsius (°C). The thermal conductivity is obtained by analyzing the basic line ledger to obtain the target conductor model, extracting the theoretical solid thermal conductivity of the corresponding metal material, and multiplying it by a preset stranding gap correction factor (recommended value range is 0.65 to 0.75).

[0038] 2.2 Filtering out abnormal regions, as detailed below: Based on the difference between the internal heating power of the target observation area and the internal heating power of adjacent observation areas, the local relative change rate is calculated. Specifically, taking the target observation area as the center, a predetermined number of observation areas (e.g., 5 in front and 5 behind) are selected along the axial direction of the conductor to form a local spatial sliding window. The median or mean of the internal heating power of all observation areas within the local spatial sliding window is calculated as the local reference power of the target observation area. The difference between the internal heating power of the target observation area and the local reference power is calculated, and the difference is divided by the local reference power to obtain the local relative change rate.

[0039] By comparing the local relative change rate with a preset judgment threshold (recommended value range of 15% to 20%), if the local relative change rate of the target observation area is greater than the preset judgment threshold, it is screened as an abnormal area.

[0040] This step inverts the internal heating power based on the surface temperature, ambient temperature, and thermal properties of the conductor in each observation area, and calculates the local relative rate of change based on the difference between the internal heating power of the target observation area and the internal heating power of adjacent observation areas. This effectively isolates background interference caused by uneven wind cooling and load fluctuations, and improves the identification confidence and spatial positioning accuracy of local heating defects in the conductor (such as scattered strands and broken strands).

[0041] Step 3: Extract the inter-frame displacement sequence of the multispectral inspection data relative to the preset background, calculate the cross-modal phase correlation of the inter-frame displacement sequence, and then perform coordinate correction on the abnormal areas.

[0042] 3.1 Extract the inter-frame displacement sequences corresponding to the visible light data and infrared data in the multispectral inspection data, including: 3.1.1 Extract the absolute displacement sequence of each observation area from the visible light and infrared data. The continuous image frame sequences of visible light and infrared light are used as inputs, respectively, and the Lucas-Kanade (LK) optical flow algorithm is used for feature tracking. Specifically, the algorithm is based on the assumption that the local pixel grayscale remains unchanged between adjacent frames and the spatial consistency constraint. It obtains pixel-level motion vectors by solving the basic optical flow equations. Then, the motion vectors are used to continuously track the spatial pixel coordinate changes of feature points in the observation area of ​​the conductor in the visible light data and infrared data over time, thereby obtaining the absolute displacement sequence corresponding to each observation area.

[0043] Meanwhile, since the data alignment and perspective projection transformation based on GNSS and attitude information in step 1 mainly eliminates low-frequency macroscopic attitude distortion, it is difficult to completely filter out the high-frequency mechanical micro-oscillations of the UAV gimbal. In order to avoid non-coplanar background geometric tearing caused by perspective projection, the system uses the same LK optical flow algorithm to obtain a preset background displacement sequence in the visible light and infrared original image sequences before performing the perspective projection transformation. Specifically: First, in the initial frame of each original image sequence, a non-conductor background region is delineated using the inverse mask of the corresponding conductor connected component. Within this non-conductor background region, the gray-level gradient of each pixel in two mutually perpendicular directions is calculated. Pixels whose gray-level gradient magnitudes in both directions are greater than a preset background gradient threshold (e.g., 40) are extracted and used as the initial pixel coordinates of stable background feature points (such as fixed towers). Subsequently, using the initial pixel coordinates as the tracking starting point, the motion vector of the feature point between adjacent frames is calculated using the LK optical flow algorithm, and this motion vector is superimposed on the initial pixel coordinates to track the original pixel coordinates of the background feature point in adjacent consecutive frames. Finally, using the homography matrix obtained in step 1, the original pixel coordinates of the background feature point in adjacent frames are mapped to the projection plane of the longitudinal section of the conductor, and the mapped coordinates of the same source feature points are subtracted to generate the preset background displacement sequence corresponding to the visible light data and infrared data.

[0044] In practice, the feature point extraction and LK optical flow algorithm tracking of the non-conductor background region can be quickly achieved by calling open-source computer vision code libraries in this field (such as the goodFeaturesToTrack and calcOpticalFlowPyrLK function interfaces in the OpenCV library).

[0045] 3.1.2 Subtract the absolute displacement sequence from the synchronously extracted preset background displacement sequence to obtain the inter-frame displacement sequences corresponding to the visible light data and infrared data. For each observation area, in both the visible light mode and the infrared mode, the corresponding absolute displacement sequence is compared with the preset background displacement sequence in the same mode acquired simultaneously, and the vector difference is performed frame by frame according to the timestamp. Since the preset background displacement sequence essentially represents the mechanical disturbance trajectory of the UAV platform, the above vector difference operation can directly extract the common-mode component introduced by the platform disturbance from the absolute displacement of the conductor, thereby extracting the intrinsic micro-motion sequence of the conductor, and thus obtaining the inter-frame displacement sequences corresponding to the visible light data and the infrared data respectively.

[0046] 3.2 Calculate the cross-power spectrum of the visible light inter-frame displacement sequence and the infrared inter-frame displacement sequence within the same observation area, and extract the offset parameter corresponding to the phase correlation peak as the displacement compensation parameter, including: For the visible light inter-frame displacement sequence and the infrared inter-frame displacement sequence of the same observation area, the temporal displacement variance of each pixel over the entire time series is calculated, thereby transforming the motion features in the time dimension to the spatial dimension and constructing a two-dimensional visible light motion feature spatial distribution matrix and a two-dimensional infrared motion feature spatial distribution matrix characterizing the intensity of regional micro-motion; the formula for calculating the temporal displacement variance is as follows: (5) in, Coordinates in the spatial distribution matrix of motion features The value of the element at that position, and These represent the horizontal and vertical pixel index variables within the observation area, respectively; The total number of frames contained in the inter-frame displacement sequence; This refers to the time frame sequence number; coordinates The pixel at the th The magnitude of the displacement vector at the frame; For this pixel point in The average value of the displacement vector magnitude within a frame.

[0047] Frequency domain mapping is performed on the extracted two-dimensional visible light motion feature spatial distribution matrix and the two-dimensional infrared motion feature spatial distribution matrix to calculate the cross-modal cross-power spectrum and invert its phase correlation. The specific formulas for calculating the cross-power spectrum and phase correlation are as follows: (6) in, The spatial domain pulse function matrix characterizing phase correlation; This is the normalized cross-power spectrum calculated from the spatial distribution matrix of two-dimensional motion features; It is a two-dimensional inverse discrete Fourier transform (IDFT) operator; The frequency domain signal matrix is ​​obtained by two-dimensional discrete Fourier transform of the spatial distribution matrix of two-dimensional visible light motion characteristics; It is the complex conjugate of the frequency domain signal matrix after the two-dimensional infrared motion feature spatial distribution matrix is ​​transformed by the two-dimensional discrete Fourier transform; Modulo operation for obtaining the magnitude of a complex matrix.

[0048] The amplitudes of all elements within the spatial domain impulse function matrix are compared pixel-by-pixel to extract the Dirac peak with the largest amplitude, which is the phase correlation peak. Figure 2 As shown in the figure, the three-dimensional surface mapping result of the spatial domain impulse function matrix obtained based on the cross-power spectrum is displayed. It can be seen that the spatial domain impulse function matrix presents a single and sharp Dirac peak point in the two-dimensional optimization space. The two-dimensional coordinate position of this peak point on the bottom coordinate system directly indicates the pixel disparity hysteresis in the horizontal and vertical directions of the two sets of inter-frame displacement sequences caused by the static physical installation parallax of the heterogeneous sensor. Based on this, the system directly extracts the horizontal and vertical pixel disparity hysteresis (i.e., offset parameters) of this phase correlation peak as the displacement compensation parameters.

[0049] 3.3 Based on displacement compensation parameters, the coordinates of the abnormal region are corrected, including: The two-dimensional center pixel coordinates of the abnormal region selected under the infrared mode are extracted, and the displacement compensation parameter is superimposed on the two-dimensional center pixel coordinates as a spatial translation vector, so that the abnormal heating position under the infrared mode is accurately aligned to the same physical surface region in the visible light image, thereby completing the coordinate correction of the abnormal region.

[0050] This step extracts the inter-frame displacement sequence of multispectral inspection data relative to a preset background and calculates the cross power spectrum of the visible light inter-frame displacement sequence and the infrared inter-frame displacement sequence to extract displacement compensation parameters. This effectively eliminates heterogeneous parallax and UAV mechanical jitter, realizes coordinate correction of abnormal areas, and provides a unified spatial alignment benchmark for subsequent accurate extraction of appearance defect features of abnormal areas.

[0051] Step 4: Extract the appearance defect features of the corrected abnormal area and combine them with the internal heating power to generate quality inspection results, including: 4.1 Based on a pre-constructed nonlinear mapping relationship, the local relative rate of change in the abnormal region is converted into an anomaly severity score, as follows: Extract the local relative change rate of the abnormal area obtained from step 2.2 based on the internal heating power. Considering that the electrical degradation risk caused by internal heating of the conductor increases non-linearly with temperature deviation, a pre-constructed non-linear mapping relationship is used to amplify the local relative change rate into an anomaly severity score characterizing the severity of the risk. The calculation formula for the non-linear mapping relationship is as follows: (7) in, Score the degree of abnormality. This represents the local relative rate of change in the abnormal region. This is the preset non-linear scaling factor; a value of 10 is recommended. The preset exponential gain coefficient, recommended to be 0.05, is used to control the boundary of the sharpness of the conversion of heating deviation into quantifiable risk. The recommended value for the basic risk score intercept is 50.

[0052] It should be noted that the values ​​of the nonlinear scaling factor, exponential gain factor, and basic risk score intercept can all be obtained by statistical fitting based on historical thermoelectric fatigue failure physical test data of specific types of transmission conductors.

[0053] 4.2 The corrected coordinates of the abnormal areas are mapped to the visible light data of the multispectral inspection data, and the appearance defect features of the corresponding areas are extracted, as follows: Since the infrared anomaly coordinates have been aligned to the visible light global coordinate reference in step 3, the corrected two-dimensional center pixel coordinates of the anomaly region are directly used as the center point. Based on the actual physical step size equivalent to the fixed pixel step size when performing equidistant segmentation of the wire connected domain in step 1, the visible light image block corresponding to the range of the anomaly region is extracted in situ in the visible light data as the target feature conjugate block.

[0054] The visible light grayscale image of the target feature conjugate block is subjected to grayscale linear quantization. Specifically, for any pixel in the grayscale image, its original grayscale value between 0 and 255 is extracted. Then, according to the ratio of the target quantization level (i.e., 16) to the original total grayscale level (i.e., 256), the original grayscale value is scaled and rounded down proportionally, thereby converting the grayscale space of the visible light grayscale image into a quantized image composed of 16 discrete levels from 0 to 15. Based on the quantized image, the Gray-Level Co-occurrence Matrix (GLCM) algorithm is used. By setting the matrix pixel generation step size parameter to 1, the GLCM is calculated in four directions: 0°, 45°, 90°, and 135°. Statistical features of texture characteristics such as contrast, energy, and correlation of local pixels are obtained as texture parameters characterizing the corrosion and oxidation state of the conductor surface.

[0055] Simultaneously, the same Canny edge detection algorithm as used in step 1.2 to identify candidate edge point sets is employed to extract the local contour of the target feature conjugate block. This involves calculating the global grayscale gradient magnitude and direction of pixels within the block and performing non-maximum suppression to eliminate false edges. Subsequently, using high and low thresholds specific to visible light data (e.g., high and low thresholds set to 150 and 50 respectively), strong edge points are retained and weak edge points that meet the conditions are connected, thereby outputting a binarized contour map characterizing the micromorphology of the block. Since the target feature conjugate block has been unified to the global coordinate reference of the visible light image, the local tangent of the first axial trajectory extracted in step 1 within the target feature conjugate block is directly called as the reference center line. Subsequently, the spatial variance of the normal offset distance from each edge pixel in the binarized contour map to the reference center line is calculated as a parameter characterizing the morphological distortion caused by wire strand breakage or detachment.

[0056] After the texture parameters and morphological distortion parameters are spliced ​​together, they together constitute the appearance defect features.

[0057] 4.3 Input the anomaly score and appearance defect features into the trained defect classification model to obtain quality inspection results including defect type and risk level, as follows: In this embodiment, the defect classification model adopts a multi-class support vector machine (SVM) topology based on radial basis function (RBF). To address the need for solving two-dimensional labels of multiple types and levels, the model internally employs a one-to-one strategy to construct multiple binary classifier nodes. During the model training phase, multi-dimensional feature vectors containing labeled ground truth values ​​are extracted from historical inspection samples. The anomaly score, various texture parameters, and morphological distortion parameters of the samples are concatenated into an input vector. Manually labeled defect types and risk levels are combined and encoded to form a single-class label (e.g., "scattered stock - severe") as the output label, which is then input into the SVM model. Subsequently, the Sequential Minimum Optimization (SMO) algorithm is used to solve the dual problem of the support vector machine. Specifically, in each iteration, two Lagrange multipliers that violate the KKT conditions are selected, while other multipliers are fixed. The quadratic programming subproblem of these two multipliers is solved analytically, thereby continuously iterating and updating the hyperplane parameters and kernel function weights of each binary classifier node, enabling it to decouple multi-dimensional features.

[0058] In the actual inference process, the defect classification model takes the multi-dimensional joint feature vector formed by concatenating the appearance defect features and the anomaly degree score as input, and uses the radial basis kernel function to implicitly map the joint feature vector to a high-dimensional feature space. By calculating the inner product distance between the input feature vector and the model support vector and substituting it into the boundary decision function of the hyperplane, the combined category label corresponding to the target region is calculated through the voting mechanism of each binary classifier node. Then, the corresponding defect type (such as broken strand, scattered strand, or corrosion) and risk level (such as general, severe, or critical) are decoded and separated, thereby generating the final quality inspection result.

[0059] To further verify the classification effectiveness and generalization ability of the defect classification model described in this embodiment when dealing with heterogeneous multidimensional features, Figure 3 The confusion matrix evaluation results of the model on the independent test sample set are given; for example... Figure 3 As shown, the horizontal axis of the matrix represents the predicted combination results output by the model inference, and the vertical axis represents the manually labeled true combination category labels. The percentage values ​​marked in the figure represent the normalized prediction ratio of the model for each combination category. Blank grids without marked values ​​represent a normalized prediction ratio of 0.0% for that combination category, meaning that the model did not make any form of misjudgment or omission among related categories. From the distribution pattern of the data in the figure, it can be seen that, thanks to the feature-level fusion of visible light morphological distortion, GLCM texture parameters and infrared anomaly score, as well as the high-dimensional nonlinear mapping of the radial basis function (RBF), the model has significant advantages in categories such as "normal-no risk" and "corrosion-moderate". The model maintained a high classification confidence level in the combination categories with significant feature differences. Meanwhile, the test results objectively reflect that, for fault conditions with relatively continuous physical evolution (such as the inter-class boundaries between "scattered strands - severe" and "scattered strands - critical"), the model exhibits a small amount of reasonable prediction bias. However, the vast majority of these prediction biases are limited to adjacent risk levels within the same defect type, with very few serious misjudgments occurring across defect types. These test results objectively demonstrate that the pre-trained defect classification model in this embodiment, within an acceptable engineering error range, achieves comprehensive decoupling and accurate assessment of complex transmission line degradation modes and their derived electrical risks, possessing significant practical engineering application value.

[0060] This step extracts appearance defect features by mapping the corrected coordinates of the abnormal region to visible light data, and inputs them together with the anomaly degree score obtained by converting the local relative change rate into the defect classification model. This achieves multimodal feature coupling between the internal thermodynamic state and the external spatial physical form, establishes a diagnostic link with logical verification of internal and external features, and improves the output accuracy and environmental resistance of quality inspection results.

[0061] Example 2, Figure 4A device for testing the quality of transmission line conductors is presented, comprising the following modules: The contour recognition module is used to identify the conductor contour and discretize the conductor into several observation areas based on the multispectral inspection data of the transmission line. The heat generation analysis module is used to calculate the surface heat loss and axial heat conduction of each observation area based on the surface temperature, ambient temperature and thermal properties of the conductor, invert the internal heat generation power of each observation area, and filter out abnormal areas in combination with preset judgment thresholds. The coordinate correction module is used to extract the inter-frame displacement sequence of the multispectral inspection data relative to the preset background, and after calculating the cross-modal phase correlation of the inter-frame displacement sequence, to perform coordinate correction on the abnormal region. The quality inspection module is used to extract the appearance defect features of the corrected abnormal area and combine them with the internal heating power to generate quality inspection results.

[0062] Example 3, Figure 5 An electronic device is provided, comprising: a memory for storing a computer program; and a processor for implementing the steps of any of the methods when executing the computer program.

[0063] The above formulas are all dimensionless calculations. The formulas are derived from software simulations using a large amount of collected data, and are the closest to the real situation. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0064] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0065] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0066] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0067] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0068] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. 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 quality inspection of transmission line conductors, characterized in that, Includes the following steps: Based on multispectral inspection data of transmission lines, the outline of the conductor is identified and the conductor is discretized into several observation areas; Based on the surface temperature, ambient temperature, and thermal properties of the conductor in each observation area, the surface heat loss and axial heat conduction in each observation area are calculated, the internal heating power of each observation area is obtained by inversion, and abnormal areas are screened in combination with the preset judgment threshold. After extracting the inter-frame displacement sequence of the multispectral inspection data relative to a preset background and calculating the cross-modal phase correlation of the inter-frame displacement sequence, coordinate correction is performed on the abnormal region. Extract the appearance defect features of the abnormal area after correction, and combine them with the internal heating power to generate quality inspection results.

2. The method according to claim 1, characterized in that, The multispectral inspection data includes visible light data and infrared data; the identification of the conductor contour includes: Based on visible light data and infrared data, the first axial trajectory and the second axial trajectory representing the direction of the conductor are extracted respectively; Using the first axial trajectory and the second axial trajectory as a reference, the connected domain boundaries of the corresponding conductor regions are extracted to obtain the visible light conductor profile and the infrared conductor profile.

3. The method according to claim 2, characterized in that, The steps for generating the first axial trajectory and the second axial trajectory include: Candidate edge point sets are extracted from the visible light data and infrared data in the initial frame, respectively; Based on the spatial geometric features of overhead power transmission lines, curve fitting is performed on the candidate edge point set to obtain the first axial trajectory and the second axial trajectory.

4. The method according to claim 1, characterized in that, The calculation formulas for surface heat loss and axial heat conduction are as follows: ， ， in, For heat loss on the surface, The heat transfer coefficient of the conductor surface. The outer surface area of ​​the observation area. For the first infrared mode in multispectral inspection data Surface temperature of each observation area The ambient temperature; For axial heat conduction, Thermal conductivity, Let be the cross-sectional area of ​​the conductor. The discretization step size for the observation area. , They are respectively the first infrared modes. The, the Surface temperature of the observation area.

5. The method according to claim 1, characterized in that, The filtered abnormal regions include: Based on the difference between the heat generation power inside the target observation area and the heat generation power inside the adjacent observation area, the local relative rate of change is calculated. By comparing the local relative change rate with a preset judgment threshold, abnormal areas are filtered out.

6. The method according to claim 1, characterized in that, The coordinate correction of the abnormal region includes: Extract the inter-frame displacement sequence corresponding to the visible light data and infrared data in the multispectral inspection data; Calculate the cross power spectrum of the visible light inter-frame displacement sequence and the infrared inter-frame displacement sequence within the same observation area, and extract the offset parameter corresponding to the phase correlation peak as the displacement compensation parameter. Based on the displacement compensation parameters, the coordinates of the abnormal region are corrected.

7. The method according to claim 6, characterized in that, The extraction of the inter-frame displacement sequence corresponding to the visible light data and infrared data in the multispectral inspection data includes: The absolute displacement sequences of each observation area were extracted from the visible light data and infrared data, respectively. The inter-frame displacement sequences corresponding to the visible light data and infrared data are obtained by subtracting the absolute displacement sequence from the synchronously extracted preset background displacement sequence.

8. The method according to claim 5, characterized in that, The generated quality inspection results include: Based on a pre-constructed nonlinear mapping relationship, the local relative rate of change in the abnormal region is converted into an anomaly score. The corrected coordinates of the abnormal area are mapped to the visible light data of the multispectral inspection data, and the appearance defect features of the corresponding area are extracted. The anomaly score and appearance defect features are input into a trained defect classification model to obtain quality inspection results that include defect type and risk level.

9. A device for detecting the quality of transmission line conductors, characterized in that, Includes the following modules: The contour recognition module is used to identify the conductor contour and discretize the conductor into several observation areas based on the multispectral inspection data of the transmission line. The heat generation analysis module is used to calculate the surface heat loss and axial heat conduction of each observation area based on the surface temperature, ambient temperature and thermal properties of the conductor, invert the internal heat generation power of each observation area, and filter out abnormal areas in combination with preset judgment thresholds. The coordinate correction module is used to extract the inter-frame displacement sequence of the multispectral inspection data relative to the preset background, and after calculating the cross-modal phase correlation of the inter-frame displacement sequence, to perform coordinate correction on the abnormal region. The quality inspection module is used to extract the appearance defect features of the corrected abnormal area and combine them with the internal heating power to generate quality inspection results.

10. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the steps of the method as claimed in any one of claims 1 to 8.

Images