A method and system for detecting the pollution level of insulators based on multi-source spectral sensing

Abstract

This application relates to the field of multi-source spectral sensing technology, providing a method and system for detecting the degree of insulator contamination based on multi-source spectral sensing. The method includes: acquiring spectral image data of the insulator at multiple focal lengths and scattering spectral data of the insulator surface; extracting contamination layer thickness information from the spectral image data; fusing the contamination layer thickness information and the scattering spectral data, inputting the fusion result into a graph neural network, and determining the spatial correlation between location units through the graph neural network, where each location unit corresponds to a spatial partition region of the insulator; and generating three-dimensional contamination distribution data of the insulator based on the spatial correlation, thereby solving the problems of low spatial resolution and low accuracy in insulator contamination detection in existing technologies, and improving the spatial resolution and accuracy of insulator contamination detection.

Description

Technical Field

[0001] This application relates to the field of multi-source spectral sensing technology, and in particular to a method and system for detecting the degree of pollution of insulators based on multi-source spectral sensing. Background Technology

[0002] High-voltage transmission line insulators operate in complex environments for extended periods, and their surfaces are prone to accumulating contaminants, leading to decreased insulation performance and even flashover accidents. Traditional testing methods are insufficient to accurately measure the uneven contamination on the surfaces of complex insulator configurations (such as anti-flashover skirts). There is an urgent need for a high-precision testing technology that can combine contaminant thickness distribution and compositional characteristics to achieve a three-dimensional visual assessment of contamination levels.

[0003] Currently, a method for detecting insulator pollution based on single hyperspectral imaging is widely used. This method acquires surface reflectance spectral data of insulators using a hyperspectral camera with a fixed focal length, uses spectral features to invert the pollution components, and combines image segmentation technology to estimate the pollution coverage area.

[0004] This method relies on spectral data from a single perspective, making it difficult to accurately reflect the three-dimensional thickness distribution of the contamination layer. Furthermore, the lack of integration of contamination scattering characteristics data results in insufficient ability to distinguish between surface and subsurface contamination. In addition, the lack of modeling for the complex physical structure of insulators leads to discrepancies between the analysis of contamination diffusion paths and actual operating conditions. Summary of the Invention

[0005] This application provides a method and system for detecting the degree of insulator contamination based on multi-source spectral sensing, in order to solve the problems of low spatial resolution and low accuracy in insulator contamination detection in the prior art.

[0006] Firstly, this application provides a method for detecting the pollution level of insulators based on multi-source spectral sensing, including:

[0007] Acquire spectral image data of the insulator at multiple focal lengths and scattering spectral data of the insulator surface;

[0008] Extract the dirt layer thickness information from the spectral image data;

[0009] The contamination layer thickness information and the scattering spectrum data are fused together, and the fusion result is input into a graph neural network. The graph neural network is used to determine the spatial relationship between position units, wherein the position units correspond to the spatial partition regions of the insulator.

[0010] Based on the spatial correlation, three-dimensional pollution distribution data of the insulator is generated.

[0011] Optionally, the step of fusing the dirt layer thickness information and the scattering spectral data, inputting the fusion result into a graph neural network, and determining the spatial correlation between location units through the graph neural network includes:

[0012] The dirt layer thickness information and the scattering spectrum data are aligned according to position units to form a joint feature vector for each position unit;

[0013] The joint feature vector is input into a graph neural network, and the target space connection weights between adjacent position units are calculated by the graph neural network.

[0014] Based on the target spatial connection weights, the spatial association between location units is determined.

[0015] Optionally, calculating the target spatial connection weights between adjacent location units using the graph neural network includes:

[0016] Identify the physical connection topology between position units from the physical spatial structure of the insulator;

[0017] The similarity of dirt thickness and spectral features between adjacent units are extracted from the joint feature vector to form a two-dimensional similarity index;

[0018] The physical connection topology and the two-dimensional similarity index are fused into a spatial constraint factor;

[0019] The spatial constraint factor is coupled with the joint feature vector to generate target spatial connection weights between adjacent position units.

[0020] Optionally, the coupling of the spatial constraint factor with the joint feature vector to generate target spatial connection weights between adjacent location units includes:

[0021] The spatial constraint factors are converted into edge constraints of a graph neural network;

[0022] The physical configuration features of the spatial constraint factor are injected into the joint feature vector to generate an enhanced feature vector;

[0023] The graph neural network performs constraint-guided processing on the enhanced feature vector and the edge constraint conditions, and outputs the initial spatial connection weight values.

[0024] Based on the pollution diffusion characteristics of the insulator, the initial spatial connection weight value is optimized and adjusted to generate the target spatial connection weight.

[0025] Optionally, extracting the dirt layer thickness information from the spectral image data includes:

[0026] The optical depth parameters of the dirt layer are calculated based on the differences in spectral focus variation in the spectral image data.

[0027] Based on the optical depth parameters and combined with the preset image sharpness variation pattern, dirt layer thickness information is generated.

[0028] Optionally, generating dirt layer thickness information based on the optical depth parameters and a preset image sharpness variation pattern includes:

[0029] The optical depth parameters are matched and mapped with the image sharpness variation pattern;

[0030] Based on the matching mapping results, the quantified value of the dirt layer thickness is calculated using a linear interpolation method;

[0031] The quantized values ​​of the contamination layer thickness of all location units are aggregated to form contamination layer thickness information.

[0032] Optionally, generating three-dimensional pollution distribution data of the insulator based on the spatial correlation includes:

[0033] Based on the spatial relationships, construct the contamination transfer paths between location units;

[0034] The thickness of the contamination layer is calculated according to the contamination transmission path;

[0035] The calculation results are mapped to a preset three-dimensional insulator model to generate three-dimensional pollution distribution data.

[0036] Secondly, this application provides an insulator pollution level detection system based on multi-source spectral sensing, comprising:

[0037] The acquisition module is used to acquire spectral image data of the insulator at multiple focal lengths and scattering spectral data of the insulator surface;

[0038] The extraction module is used to extract dirt layer thickness information from the spectral image data;

[0039] A fusion module is used to fuse the contamination layer thickness information and the scattering spectrum data, input the fusion result into a graph neural network, and determine the spatial relationship between position units through the graph neural network, wherein the position units correspond to the spatial partition regions of the insulator.

[0040] The generation module is used to generate three-dimensional pollution distribution data of the insulator based on the spatial correlation.

[0041] Thirdly, this application provides a computing device, including a processor and a memory, wherein the memory stores a computer program, and the processor is configured to run the computer program to perform a method for detecting the degree of insulator pollution based on multi-source spectral sensing as described in any of the first aspects.

[0042] Fourthly, this application provides a computer storage medium storing computer program instructions thereon, which, when executed by a processor, implement the insulator pollution level detection method based on multi-source spectral sensing as described in any one of the first aspects.

[0043] This application provides a method for detecting the degree of insulator contamination based on multi-source spectral sensing. The method includes: acquiring spectral image data of the insulator at multiple focal lengths and scattering spectral data of the insulator surface; extracting contamination layer thickness information from the spectral image data; fusing the contamination layer thickness information and the scattering spectral data, inputting the fusion result into a graph neural network, determining the spatial correlation between position units through the graph neural network, wherein the position units correspond to the spatial partition regions of the insulator; and generating three-dimensional contamination distribution data of the insulator based on the spatial correlation.

[0044] The technical solution provided in this application has the following beneficial effects:

[0045] This application achieves complementary acquisition of contamination layer depth information and surface / subsurface composition information through multifocal spectral imaging and simultaneous acquisition of scattering spectra, providing a multi-source data foundation for three-dimensional contamination analysis. Based on the focal point variation differences of multifocal spectra, the optical thickness of the contamination layer is quantified, solving the challenge of thickness measurement for uneven contamination accumulation on complex insulator surfaces. Through synergistic analysis of thickness and spectral characteristics, the comprehensive characterization of the physical properties and chemical composition of contamination is enhanced, providing multi-dimensional data support for spatial correlation modeling. Graph neural networks are used to capture the diffusion patterns of contamination between spatial partitioning units of the insulator, establishing a contamination distribution correlation model that conforms to actual physical processes. Combining contamination thickness and spatial diffusion paths, the three-dimensional contamination distribution morphology is reconstructed, enabling a visual assessment of contamination levels.

[0046] Furthermore, this application aligns the contamination layer thickness information and scattering spectral data according to positional units to form a joint feature vector, inputs it into a graph neural network to calculate the target spatial connection weights between adjacent units, and determines the spatial correlation of contamination diffusion based on the weights.

[0047] Furthermore, by fusing multi-source data and using spatial modeling with graph neural networks, the diffusion path of contaminants on the complex surface of insulators can be accurately depicted, improving the physical rationality and accuracy of the three-dimensional distribution analysis of contaminants.

[0048] These or other aspects of this application will become more apparent in the following description of the embodiments. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 A flowchart illustrating a method for detecting the degree of insulator contamination based on multi-source spectral sensing, provided in an embodiment of this application;

[0051] Figure 2 A schematic diagram of the structure of an insulator pollution level detection system based on multi-source spectral sensing provided in an embodiment of this application;

[0052] Figure 3 This is a schematic diagram of the structure of a computing device provided in an embodiment of this application. Detailed Implementation

[0053] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0054] In some of the processes described in the specification, claims, and accompanying drawings of this application, multiple operations appearing in a specific order are included. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or may be executed in parallel. The operation numbers, such as 101, 102, etc., are merely used to distinguish different operations and do not themselves represent any execution order. Furthermore, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first," "second," etc., in this document are used to distinguish different messages, devices, modules, etc., and do not represent a chronological order, nor do they limit "first" and "second" to different types.

[0055] Existing methods for insulator pollution detection based on single hyperspectral imaging have limitations: they rely on spectral data with a fixed focal length, resulting in insufficient analytical capability for the three-dimensional thickness distribution of the pollution layer; simultaneously, they lack coordinated analysis of surface and subsurface scattering spectra, making it difficult to distinguish the depth characteristics of pollution components; furthermore, they fail to consider the influence of the complex insulator configuration on the pollution diffusion path, leading to systematic deviations between the detection results and actual operating conditions. These shortcomings stem from the dual constraints of a single data dimension and a lack of physical modeling, necessitating a novel detection method that integrates multi-source optical features and spatial topological correlations.

[0056] To address the aforementioned issues, this application proposes a method for detecting insulator contamination levels based on multi-source spectral sensing. Its innovation lies in achieving three-dimensional reconstruction of contamination distribution through simultaneous acquisition of multi-focal-length spectral imaging and scattering spectra, combined with spatial correlation modeling using graph neural networks. Specifically, the contamination layer thickness is obtained by extracting depth information from spectral images at different focal lengths, surface / subsurface scattering spectral data are fused to form a joint feature vector, and graph neural networks are used to analyze the contamination diffusion correlation between spatial partitioning units of the insulator, ultimately generating three-dimensional contamination distribution data reflecting the actual physical process. This method overcomes the dimensional limitations of single-spectral detection. Through collaborative calculation of multi-source optical data and topological structure, it solves the problem of accurate measurement of contamination thickness and composition, and corrects diffusion path deviations caused by complex configurations, providing a high-precision three-dimensional visualization solution for insulator contamination status assessment.

[0057] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0058] Figure 1 A flowchart of an insulator pollution level detection method based on multi-source spectral sensing provided in this application embodiment is shown below. Figure 1 As shown, the method includes:

[0059] Step 101: Acquire spectral image data of the insulator at multiple focal lengths and scattering spectral data of the insulator surface.

[0060] In step 101, the spectral image data represents the surface reflection spectrum information of the insulator acquired by a hyperspectral camera at different focal lengths, including the correspondence between wavelength and reflection intensity, used to analyze the optical properties of the contamination layer. The scattering spectrum data represents the scattered light signals of the insulator surface and subsurface obtained by a laser scatterer, reflecting the size, composition, and distribution characteristics of the contamination particles, including surface scattering spectrum data and subsurface scattering spectrum data.

[0061] In this embodiment, a hyperspectral camera is moved along the optical axis at preset focal length intervals to acquire multiple sets of spectral images of the insulator surface under different focusing states. Simultaneously, a laser scatterer is used to illuminate the insulator surface at a fixed angle, receiving the scattered light signals and converting them into spectral data. The multi-focal-length spectral images are used for subsequent depth information extraction, and the scattered spectral data is used for pollution component analysis; together, they form the basis of multi-source optical detection.

[0062] For example, in the inspection of FXBW4-110 insulators at Substation A, a hyperspectral camera with a focal length step of 0.2mm was used to acquire 20 sets of spectral images, while a 532nm laser scattering instrument was used to obtain the scattering spectrum. The focal length interval was calculated based on the depth-of-field formula Δz=2λ / (NA)^2, where λ=500nm is the center wavelength of visible light and NA=0.25 is the numerical aperture of the lens, ensuring that adjacent images cover a continuous depth range.

[0063] Step 102: Extract the dirt layer thickness information from the spectral image data.

[0064] In step 102, the contamination layer thickness information characterizes the physical thickness of the contamination deposit on the insulator surface and is obtained by inverting the focal difference of the multifocal spectral image.

[0065] In this embodiment, a sharpness analysis is performed on the multifocal spectral image sequence, the image gradient value of each position unit at different focal lengths is calculated, and a sharpness-focal length variation curve is constructed. The optical depth parameter is calculated based on the focal length difference corresponding to the peak of the curve, and combined with the preset sharpness-thickness mapping relationship, it is converted into actual dirt layer thickness data.

[0066] For example, for the insulator of line B, the spectral image gradient value Σ|G(x,y)| / n of the edge unit of its shed is extracted, where G(x,y) is the pixel gradient and n is the total number of pixels; according to the sharpness threshold calibrated in the laboratory, when the gradient value drops to 50% of the peak value, the corresponding focal length difference is determined to be 0.3mm, and the thickness T=0.2mm is calculated by interpolation formula T=k·Δf (k=0.67 is the calibration coefficient).

[0067] Step 103: Fuse the contamination layer thickness information and the scattering spectrum data, input the fusion result into a graph neural network, and determine the spatial relationship between position units through the graph neural network, wherein the position units correspond to the spatial partition regions of the insulator.

[0068] In step 103, a graph neural network (GNN) is a deep learning model used to process data with topological structures. Its nodes represent spatial partitioning units of the insulator, and edges represent the pollution diffusion correlation between units. Spatial connection weights are calculated by aggregating the feature information of adjacent nodes, thereby modeling the diffusion law of pollution on the insulator surface. Spatial correlation refers to the degree of mutual influence of pollution diffusion between units at different locations on the insulator surface, which is quantitatively represented by the connection weight values ​​calculated by the GNN. A larger weight value indicates a higher probability of pollution diffusion along that path. Spatial partitioning regions refer to several detection units divided into the surface of the insulator based on its physical structural characteristics (such as skirt spacing, curvature, etc.). These units are obtained through 3D scanning or discretization of CAD models. Each unit corresponds to a node in the GNN, used for the extraction and correlation analysis of local pollution features.

[0069] In this embodiment, the thickness information and the scattering spectrum are aligned by coordinates and combined into a feature vector containing thickness and dual-wavelength scattering rate. After being input into the graph neural network, the network calculates the connection weights based on the similarity of node features and physical proximity, where nodes represent insulator partitioning units and edge weights characterize the correlation strength of pollution diffusion. Finally, the spatial correlation matrix of the entire graph is output.

[0070] For example, for the skirt partitioning of the insulator of power station C, the thickness of 0.18mm and the scattering rate of 0.35 / 0.28 (532nm / 1064nm) are combined into a feature vector; the graph neural network adopts a three-layer structure, and the connection weight is calculated by the weight formula W=exp(-d^2 / 2σ^2)*S, where d=8mm is the unit spacing, σ=5mm is the attenuation coefficient, and S=0.82 is the feature similarity.

[0071] Step 104: Generate three-dimensional pollution distribution data of the insulator based on the spatial correlation.

[0072] In step 104, the three-dimensional contamination distribution data represents a quantitative result reflecting the accumulation morphology and thickness variation of contamination in the three-dimensional space on the insulator surface.

[0073] In this embodiment, a pollution diffusion path network is constructed based on spatial connection weights, and thickness values ​​are superimposed along strongly correlated paths (weight > 0.7) to generate a three-dimensional contour map; spatial mapping is performed in conjunction with the insulator CAD model to achieve visualization of pollution distribution.

[0074] For example, for the D line insulator, 12 main diffusion paths are extracted based on a weight threshold of 0.7. The thickness of the downstream unit is calculated recursively along the path according to a thickness attenuation coefficient of 0.9, and finally a three-dimensional contour map with an interval of 0.1 mm is generated, in which the thickness of the red area is ≥0.3 mm.

[0075] This method accurately extracts the thickness and composition characteristics of contamination through the coordinated acquisition of multi-focal-length and scattering spectra; it uses graph neural networks to model the spatial correlation of contamination diffusion and combines physical configuration constraints to generate three-dimensional distribution data, thus realizing high-precision visualization detection of non-uniform contamination in complex-configuration insulators and providing a reliable basis for contamination degree assessment.

[0076] To address the challenges of multi-source data fusion and spatial correlation modeling in insulator pollution detection, in some embodiments, step 103 involves fusing the pollution layer thickness information and the scattering spectral data, inputting the fusion result into a graph neural network, and using the graph neural network to determine the spatial correlation between location units.

[0077] Step 201: Align the dirt layer thickness information with the scattering spectrum data according to the position units to form a joint feature vector for each position unit.

[0078] In step 201, the joint feature vector refers to multidimensional data formed by combining the dirt layer thickness value and the scattering spectrum feature according to the same position unit. The thickness value reflects the dirt accumulation height, and the scattering spectrum feature includes the optical properties of surface and subsurface dirt. The combination of the two can comprehensively characterize the physical and chemical properties of dirt.

[0079] In this embodiment, the thickness data extracted from the spectral image of each location unit is first matched and aligned with the scattering spectral data of the corresponding coordinates to ensure data spatial consistency; then the thickness value and the dual-wavelength scattering rate are combined in a fixed order to form a vector containing three feature dimensions; finally, the same operation is performed on all location units to construct a complete set of joint feature vectors, providing standardized input for subsequent graph neural network processing.

[0080] Step 202: Input the joint feature vector into the graph neural network, and calculate the target space connection weights between adjacent position units through the graph neural network.

[0081] In step 202, the spatial connection weight between adjacent location units specifically refers to the local connection strength between directly adjacent units; while the "spatial association relationship between location units" includes the global association relationship between adjacent and non-adjacent units, which is the overall topological relationship derived through the transitivity of the adjacent connection weight. The two represent a progressive relationship between local and global.

[0082] In this embodiment, the joint feature vector is input into a pre-trained graph neural network. The network first calculates the physical proximity based on the spatial coordinates of the location unit; then compares the similarity of the feature vectors of adjacent units; finally, the proximity and similarity are fused and calculated through the neural network layer, and the connection weight value between 0 and 1 is output. The larger the weight, the stronger the correlation of pollution spread.

[0083] Step 203: Determine the spatial association between location units based on the target spatial connection weights.

[0084] In this embodiment, firstly, a weight threshold is set to filter associated unit pairs; then, a directed connection relationship is constructed according to the weight value; finally, all valid connections are integrated to form a complete spatial association network, which can intuitively show the possible diffusion paths and core accumulation areas of pollution.

[0085] Here is a specific example:

[0086] In the testing of FXBW4-110 insulators at substation A, the obtained thickness data and scattering spectrum data were first aligned according to the coordinates of the location units. The thickness of a certain unit at the edge of the shed was 0.2 mm, and the scattering rates at wavelengths of 532 nm and 1064 nm corresponding to this unit were 0.35 and 0.28, respectively. These data were combined into a three-dimensional joint feature vector in the order of thickness + dual-wavelength scattering rate. Then, the joint feature vector of all location units was input into a graph neural network. This network adopts a three-layer computational structure. The first layer expands the feature vector dimension to 64 dimensions, the second layer performs a 128-dimensional feature transformation, and the third layer outputs scalar weight values. For two adjacent units with a spacing of 8 mm, the network calculates the connection weight of 0.75 based on their feature similarity S=0.82 and physical distance d=8 mm using the weight calculation formula W=exp(-d² / 2σ²)*S, where σ=5 mm is the distance attenuation coefficient, determined by experimental calibration. Finally, based on the preset weight threshold of 0.7, the adjacent unit pairs that constitute an effective spatial association are selected.

[0087] In this embodiment, by using multi-source data fusion and spatial correlation analysis of graph neural networks, the synergistic utilization of pollution thickness and composition characteristics is realized, accurately depicting the diffusion law of pollution on the surface of complex-configuration insulators, providing a reliable correlation basis for the reconstruction of three-dimensional pollution distribution, and improving the physical rationality and practicality of the detection results.

[0088] To further improve the accuracy of insulator pollution diffusion path analysis, in some embodiments, step 202: calculating the target spatial connection weights between adjacent location units using the graph neural network includes:

[0089] Step 301: Identify the physical connection topology between position units from the physical spatial structure of the insulator.

[0090] In step 301, the physical spatial structure of the insulator is obtained from the actual three-dimensional geometric configuration of the insulator object, including physical parameters such as shed spacing, tilt angle, and curvature, which are obtained directly from insulator design drawings or three-dimensional scanning data. The physical connection topology refers to the connection relationship between various positional units on the insulator surface based on the actual physical configuration, including spatial characteristics such as overlapping areas between sheds and adjacent spacing, reflecting the possible diffusion paths of pollution on the actual physical structure.

[0091] In this embodiment, the spatial coordinates and geometric parameters of each skirt unit are first extracted based on the three-dimensional structural model of the insulator; then the actual physical connection between units is analyzed, including the contact area of ​​adjacent skirts, air gap distance, etc.; finally, a topological relationship diagram representing the physical connection strength between units is constructed to provide a physical basis for subsequent spatial constraints.

[0092] Step 302: Extract the similarity of dirt thickness and spectral features between adjacent units from the joint feature vector to form a two-dimensional similarity index.

[0093] In step 302, thickness similarity reflects the consistency of the degree of dirt accumulation. Spectral feature similarity reflects the similarity of dirt composition. The two-dimensional similarity index is a comprehensive evaluation index that considers both dirt thickness similarity and spectral feature similarity simultaneously.

[0094] In this embodiment, the thickness difference and spectral feature difference of adjacent units are extracted from the joint feature vector; the thickness similarity is calculated using the relative difference method, and the spectral feature similarity is calculated using the vector angle method; the two similarities are combined into a comprehensive evaluation index according to a preset weight to fully reflect the degree of similarity of the fouling characteristics of adjacent units.

[0095] Step 303: Integrate the physical connection topology with the two-dimensional similarity index into a spatial constraint factor.

[0096] In step 303, the spatial constraint factor refers to a comprehensive parameter that integrates physical connection topology and two-dimensional similarity index, taking into account both the constraints of the actual physical structure and the degree of similarity of the contamination characteristics.

[0097] In this embodiment of the application, the physical connection strength and the two-dimensional similarity index are weighted and fused. Regions with strong physical connections are given higher fusion weights to ensure that the final spatial constraint factor not only conforms to the actual physical structure characteristics, but also reflects the possibility of pollution diffusion.

[0098] Step 304: Couple the spatial constraint factor with the joint feature vector to generate target spatial connection weights between adjacent position units.

[0099] In this embodiment, the spatial constraint factor and the joint feature vector are input into a special processing layer of the graph neural network. Through multi-level feature transformation and nonlinear calculation, a standardized connection weight value between 0 and 1 is output as the final basis for pollution diffusion analysis.

[0100] Here is a specific example:

[0101] In the inspection of FXBW4-110 insulators at substation A, the physical connection relationships between positional units were first identified based on the three-dimensional model of the insulator. The actual distance between skirt unit A and unit B was 5mm, with some overlap. Based on the overlap area ratio of 30%, the physical connection strength was determined to be 0.3. Next, the pollution characteristic data of these two units were extracted from the joint feature vector. Unit A has a thickness of 0.2mm and dual-wavelength scattering equivalences of 0.35 and 0.28, while unit B has a thickness of 0.18mm and dual-wavelength scattering equivalences of 0.33 and 0.26. The thickness similarity was calculated using the formula 1 - |T_A - T_B| / T_ma. The thickness similarity x is 0.9, where T_max = 0.2 mm is the maximum thickness difference. The similarity is 0.95 using the spectral feature similarity calculation formula (F_A·F_B) / (||F_A||×||F_B||). These two similarities are combined with a weight of 4:6 to form a two-dimensional similarity index of 0.93. Then, the physical connection strength of 0.3 and the similarity index of 0.93 are fused with a weight of 3:7 to obtain a spatial constraint factor of 0.75. Finally, this constraint factor and the joint feature vector are input into a graph neural network. After processing by the network through a three-layer computational structure, the target spatial connection weight between units A and B is 0.82.

[0102] In this embodiment, the calculated spatial connection weights are more consistent with actual working conditions by using the dual constraints of physical connection topology and similarity of pollution characteristics. This not only considers the limitation of the physical structure of the insulator on pollution diffusion, but also incorporates the influence of the pollution's own characteristics, thereby improving the accuracy and reliability of pollution diffusion path analysis and providing a more accurate basis for subsequent three-dimensional pollution distribution reconstruction.

[0103] To further improve the accuracy of pollution diffusion weight calculation, in some embodiments, step 304: coupling the spatial constraint factor with the joint feature vector to generate target spatial connection weights between adjacent location units includes:

[0104] Step 401: Convert the spatial constraint factor into edge constraints of the graph neural network.

[0105] In step 401, the edge constraint condition refers to converting the spatial constraint factor into a limiting parameter in the graph neural network that controls the edge connection strength, used to force the maintenance or weakening of the connection relationship between units at a specific location.

[0106] In this embodiment, the spatial constraint factors are first divided into different levels according to preset rules, then the corresponding constraint strength is set according to the level, and finally converted into edge connection control parameters that the graph neural network can recognize, so as to ensure that the network follows the actual physical connection constraints during the processing.

[0107] Step 402: Inject the physical configuration features of the spatial constraint factor into the joint feature vector to generate an enhanced feature vector.

[0108] In step 402, the physical configuration characteristics of the spatial constraint factor refer to quantifying key parameters in the physical spatial structure of the insulator (such as the overlapping area of ​​the skirts, the gap between suspension spaces, etc.) into computable constraints, which are formed by encoding the spatial relationship between the special geometric features of the skirts and the positional units. The enhanced feature vector refers to adding a new dimension reflecting the physical configuration characteristics on the basis of the original joint feature vector, so that the feature expression simultaneously includes pollution characteristics and spatial structure information.

[0109] In this embodiment, physical configuration-related parameters are extracted from spatial constraint factors, standardized, and used as new feature dimensions. These dimensions are then concatenated with the original joint feature vector to form an enhanced feature vector with expanded dimensions, providing more comprehensive input information for subsequent network processing.

[0110] Step 403: Perform constraint-guided processing on the enhanced feature vector and the edge constraint conditions through the graph neural network, and output the initial spatial connection weight values.

[0111] In step 403, constraint-guided processing refers to a special computational mode in which the graph neural network considers both feature data and edge constraints during the computation process. The initial spatial connection weight value refers to the preliminary association strength between adjacent units calculated by the graph neural network through forward propagation, considering only the augmented feature vectors and edge constraints. This value reflects the network's initial prediction of the probability of pollution diffusion based on the current input features and physical constraints, and has not yet been optimized based on the actual diffusion characteristics of the pollution.

[0112] In this embodiment, the enhanced feature vector is input into the first hidden layer of the graph neural network for feature transformation. When processing the second hidden layer, edge constraints are introduced to filter the connection relationships. Finally, the initial spatial connection weight values ​​are generated through the output layer. The whole process ensures that the network computation is guided by physical constraints.

[0113] Step 404: Based on the pollution diffusion characteristics of the insulator, optimize and adjust the initial spatial connection weight value to generate the target spatial connection weight.

[0114] In step 404, the pollution diffusion characteristics of the insulator refer to the physical laws of pollutant diffusion along the surface of the insulator in the actual environment, including phenomena such as salt spray deposition in the shed grooves and hydrophilic pollution accumulation along the edges, which are obtained through long-term field observation data or laboratory accelerated pollution accumulation tests.

[0115] In this embodiment, a pollution diffusion characteristic library is established based on historical detection data. The initial spatial connection weight value is compared with the typical patterns in the characteristic library. Weight values ​​that do not conform to the actual diffusion law are adjusted in a targeted manner, and finally, the target spatial connection weight that conforms to the physical reality is output.

[0116] Here is a specific example:

[0117] In the testing of FXBW4-110 insulators at substation A, the spatial constraint factor of 0.75 was first converted into edge constraints for a graph neural network. Specifically, this value was mapped to the constraint coefficients of the network edge computation layer. When the constraint coefficient was greater than 0.7, the connection was retained; when it was less than 0.5, the connection was disconnected; and when it was between the two, it was scaled linearly. Then, based on the original three-dimensional joint feature vector, the awning tilt angle feature and the overlapping area ratio feature were added to form a five-dimensional enhanced feature vector. Among them, the tilt angle of unit A was 15 degrees and the overlapping ratio was 30%, and the tilt angle of unit B was 12 degrees and the overlapping ratio was 25%. The enhanced feature vectors are then input into the graph neural network. The first layer of the network maps the five-dimensional features to a 64-dimensional space through the weight matrix W_1, and the second layer transforms them to 128 dimensions through W_2. W_1 and W_2 are parameter matrices obtained from network training. Edge constraints are introduced during the calculation of the second layer to mask connections that do not meet the constraints. The initial spatial connection weights of the output units A and B of the third layer are 0.78. Finally, based on the characteristic that pollution of this type of insulator is prone to deposit in the overlapping area of ​​the sheds, the initial weights are multiplied by a correction factor of 1.1 to obtain the final target spatial connection weights of 0.86.

[0118] In this embodiment, by introducing edge constraints and physical configuration features, the computation process of the graph neural network is made more in line with actual physical laws. Combined with the optimization and adjustment of the pollution diffusion characteristics, the final target space connection weights not only maintain the accuracy of data-driven approaches but also ensure consistency with actual working conditions, providing a reliable quantitative basis for the analysis of three-dimensional pollution distribution.

[0119] To further improve the accuracy of dirt thickness measurement, in some embodiments, step 102: extracting dirt layer thickness information from the spectral image data includes:

[0120] Step 501: Calculate the optical depth parameters of the dirt layer based on the spectral focus variation differences in the spectral image data.

[0121] In step 501, the difference in spectral focus variation in the spectral image data refers to the difference in focal position caused by the reflection / transmission characteristics of the contaminant layer in spectral images acquired at different focal lengths. This difference originates from the optical path difference caused by the change in the refractive index of the contaminant medium on light waves, and the raw data is obtained directly through multi-focal-length scanning. The optical depth parameter is a parameter obtained by analyzing the focus variation characteristics of multi-focal-length spectral images. It reflects the refraction and absorption characteristics of the contaminant layer on light waves and is used to characterize the optical penetration depth of the contaminant.

[0122] In this embodiment, firstly, a sharpness analysis is performed on the spectral image sequences at different focal lengths to calculate the sharpness index of each position unit in the image at each focal length; then, the optimal focal length corresponding to the peak sharpness of each unit is determined; finally, the optical depth parameter is calculated based on the gradient of sharpness change between adjacent focal lengths, and this parameter has a corresponding relationship with the actual thickness of the dirt layer.

[0123] Step 502: Based on the optical depth parameters and combined with the preset image sharpness variation law, generate dirt layer thickness information.

[0124] In step 502, the image sharpness variation law refers to the mapping relationship between the sharpness index and the dirt thickness established in advance through experiments, which is used to convert the optical parameters into actual thickness values.

[0125] In this embodiment, a pre-calibrated sharpness-thickness lookup table is invoked, and the corresponding thickness range is found based on the calculated optical depth parameters; a linear interpolation method is used to calculate the accurate thickness value; finally, the same operation is performed on all position units to generate complete dirt layer thickness distribution information.

[0126] Here is a specific example:

[0127] In the inspection of FXBW4-110 insulators at substation A, 20 sets of spectral image sequences were acquired using a hyperspectral camera with a focal length interval of 0.2 mm. Analysis of a unit at a certain position on the edge of the insulator skirt revealed that the unit's sharpness peaked in the 6th image set, with a sharpness index of 0.85. The sharpness in the adjacent 5th and 7th images were 0.82 and 0.81, respectively. The optical depth parameter was obtained by calculating the sharpness change rate ΔC / Δf = (0.85 - 0.81) / 0.4 mm = 0.1 mm^-1. The value is 0.25, where Δf is the change in focal length. According to the calibration curve established in the laboratory, the thickness of this type of insulator is 0.1mm when the optical depth parameter is 0.2 and 0.2mm when the parameter is 0.3. The pollution thickness of this unit is calculated to be 0.15mm using the linear interpolation formula T=0.1mm+(0.25-0.2)×(0.2mm-0.1mm) / (0.3-0.2). After processing the data of all position units in the same way, a complete pollution thickness distribution map is generated.

[0128] In this embodiment, non-contact and accurate measurement of the contamination layer thickness is achieved through multifocal spectral analysis and sharpness-thickness mapping conversion, solving the problem of thickness detection of uneven contamination on the surface of complex insulators and providing key parameter basis for contamination degree assessment.

[0129] To further improve the accuracy and reliability of dirt thickness measurement, in some embodiments, step 502: generating dirt layer thickness information based on the optical depth parameters and combined with a preset image sharpness variation law includes:

[0130] Step 601: Match and map the optical depth parameters with the image sharpness variation pattern.

[0131] In step 601, matching mapping refers to the process of comparing and searching the calculated optical depth parameters with the pre-calibrated sharpness-thickness correspondence to determine the thickness range of the optical parameters.

[0132] In this embodiment, a sharpness-thickness comparison table established in advance through standard sample experiments is first loaded. Then, the upper and lower bound parameter values ​​that are closest to the current optical depth parameters are found in the table to determine the corresponding thickness reference range, providing basic data for subsequent accurate calculations.

[0133] Step 602: Based on the matching mapping results, calculate the quantified value of the dirt layer thickness using the linear interpolation method.

[0134] In step 602, the linear interpolation method refers to a mathematical method that calculates the thickness value of an intermediate point according to a linear relationship when the parameter values ​​and thickness values ​​of two reference points are known. The quantified value of the contamination layer thickness refers to a specific numerical value that reflects the actual thickness of the contamination deposition layer at a specific location on the insulator surface, calculated by the linear interpolation method. This value is determined based on the mapping relationship between optical depth parameters and a preset sharpness-thickness law, and is the basic data unit for subsequently constructing the three-dimensional contamination distribution.

[0135] In this embodiment, a linear relationship between optical parameters and thickness is established based on the upper and lower bounds of the thickness reference interval determined by the matching mapping. The actual measured optical depth parameters are substituted into this relationship to calculate the accurate dirt thickness value, ensuring the continuity of the measurement results.

[0136] Step 603: Aggregate the quantized values ​​of the contamination layer thickness of all location units to form contamination layer thickness information.

[0137] In step 603, "all locations" refers to each location unit corresponding to the spatial partition region of the insulator, that is, all the local areas of the insulator surface that are divided and detected by multiple focal length spectral images.

[0138] In this embodiment of the application, the thickness values ​​calculated by each unit are systematically organized and stored according to the arrangement order of the insulator surface position units, and a complete dataset containing spatial position coordinates and thickness values ​​is constructed to provide a basis for subsequent analysis.

[0139] Here is a specific example:

[0140] In the inspection of FXBW4-110 insulators at Power Station B, the optical depth parameter of a certain unit in the middle of the shed was measured to be 0.28. Consulting the calibration data table for this type of insulator, it was found that parameter 0.25 corresponds to a thickness of 0.15mm, and parameter 0.3 corresponds to 0.2mm. Using the interpolation formula T=0.15mm+(0.28-0.25)×(0.2mm-0.15mm) / (0.3-0.25), the contamination thickness of this unit was calculated to be 0.18mm. In the interpolation formula, the numerator represents the difference between the current parameter and the lower limit parameter, and the denominator represents the parameter range. The same method was used to process all 128 units on the insulator surface. The thickness of the shed edge units was measured to be 0.2-0.25mm, and the thickness of the shed top units was 0.1-0.15mm. Finally, the thickness values ​​of all units were integrated according to the detection position coordinates to form a complete contamination thickness information table containing the position number and thickness value.

[0141] In this embodiment, a reliable conversion from optical parameters to thickness values ​​is achieved through precise matching mapping and interpolation calculation. Then, complete thickness distribution information is formed through system integration, providing an accurate and comprehensive thickness data foundation for assessing the contamination level of insulators and solving the problem of measuring the contamination thickness of complex surfaces.

[0142] To further improve the accuracy of the three-dimensional pollution distribution reconstruction, in some embodiments, step 104: generating the three-dimensional pollution distribution data of the insulator based on the spatial correlation includes:

[0143] Step 701: Based on the spatial relationship, construct the contamination transfer path between location units.

[0144] In step 701, the contamination transmission path refers to the possible diffusion route of contamination between various units on the insulator surface, determined based on spatial correlation, reflecting the actual propagation direction and intensity of the contamination.

[0145] In this embodiment, firstly, unit pairs with connection weights exceeding a set threshold are screened in the spatial association relationship. Then, the path direction is determined according to the weight value. Finally, the unit pairs that meet the conditions are connected to form a complete contamination transfer network, providing a path basis for subsequent thickness diffusion calculation.

[0146] Step 702: Calculate the thickness diffusion of the contamination layer according to the contamination transmission path.

[0147] In step 702, the thickness diffusion calculation refers to the calculation process of transferring and attenuating the dirt thickness value along the dirt transfer path based on the correlation strength.

[0148] In this embodiment, starting from the core area of ​​contamination, the thickness value of the upstream unit is multiplied by the corresponding attenuation coefficient and then transmitted to the downstream unit according to the direction of the transmission path. At the same time, the influence of the correlation strength on the attenuation degree on the path is considered, and finally the thickness distribution of all units under the influence of contamination diffusion is calculated.

[0149] Step 703: Map the calculation results to the preset three-dimensional model of the insulator to generate three-dimensional pollution distribution data.

[0150] In step 703, the three-dimensional model of the insulator refers to a digital three-dimensional spatial model established based on the actual physical structure of the insulator (including geometric parameters such as skirt size, spacing, and tilt angle). This model is used to map and visualize pollution distribution data.

[0151] In this embodiment of the application, the calculated thickness value is assigned to the corresponding area of ​​the three-dimensional model according to the spatial coordinates of the position unit, and different colors are used to represent different thickness ranges to generate a three-dimensional map that intuitively shows the distribution of dirt.

[0152] Here is a specific example:

[0153] In the inspection of FXBW4-110 insulators at substation B, 15 pairs of adjacent units with a weight value greater than 0.7 were first selected based on spatial correlation to construct the main pollution transmission path. The connection weight from unit A to unit B was 0.82, and the weight from unit B to unit C was 0.75. Then, starting from unit A with a maximum thickness of 0.28mm (the groove in the skirt), the thickness diffusion was calculated along the transmission path using the formula T_downstream=T_upstream×α, where the attenuation coefficient α=0.9-(0.9-0.7)×(1-W), and W is the connection weight value. The calculated thickness of unit B was 0.28×[0.9-0.2×(1-0.82)]. The thickness of unit C is 0.25 × [0.9 - 0.2 × (1 - 0.75)] = 0.22 mm. Finally, the thickness calculation results of all position units are mapped to the three-dimensional model of the insulator to generate a color distribution map with an interval of 0.1 mm. The red area is the umbrella skirt groove and adjacent area with a thickness ≥ 0.25 mm, the yellow area is the transition area of ​​0.15-0.25 mm, and the green area is the clean area with a thickness < 0.15 mm. The three-dimensional distribution characteristics of the dirt spreading outward from the umbrella skirt groove are fully presented. In the attenuation coefficient formula, 0.9 is the basic attenuation value, and 0.7 is the minimum attenuation limit. The larger the weight value, the smaller the attenuation, which is in line with the physical law of dirt diffusion.

[0154] In this embodiment of the application, by constructing the pollution transmission path and calculating the thickness diffusion, the spatial reconstruction of the pollution distribution is realized. Then, through three-dimensional visualization mapping, the accumulation of pollution on the insulator surface is presented intuitively, providing a reliable basis for pollution level assessment and maintenance decisions.

[0155] Figure 2 A schematic diagram of an insulator pollution level detection system based on multi-source spectral sensing provided in this application embodiment is shown below. Figure 2 As shown, the system includes:

[0156] The acquisition module 21 is used to acquire spectral image data of the insulator at multiple focal lengths and scattering spectral data of the insulator surface.

[0157] Extraction module 22 is used to extract dirt layer thickness information from the spectral image data.

[0158] The fusion module 23 is used to fuse the dirt layer thickness information and the scattering spectrum data, input the fusion result into the graph neural network, and determine the spatial relationship between the position units through the graph neural network, wherein the position units correspond to the spatial partition regions of the insulator.

[0159] The generation module 24 is used to generate three-dimensional pollution distribution data of the insulator based on the spatial correlation.

[0160] Figure 2 The aforementioned insulator pollution level detection system based on multi-source spectral sensing can perform... Figure 1 The implementation principle and technical effects of the insulator pollution level detection method based on multi-source spectral sensing described in the illustrated embodiment will not be repeated here. The specific operation methods of each module and unit in the multi-source spectral sensing-based insulator pollution level detection system described in the above embodiments have been described in detail in the embodiments related to this method, and will not be elaborated upon here.

[0161] In one possible design, Figure 2 The insulator pollution level detection system based on multi-source spectral sensing, as shown in the embodiment, can be implemented as a computing device, such as... Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32;

[0162] The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are invoked and executed by the processing component 32.

[0163] The processing component 32 is used to perform the above. Figure 1 The embodiment describes a method for detecting the degree of insulator contamination based on multi-source spectral sensing.

[0164] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above-described method. Alternatively, the processing component may be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above-described method.

[0165] Storage component 31 is configured to store various types of data to support operations at the terminal. The storage component can be implemented from any type of volatile or non-volatile storage device or a combination thereof, such as Static Random-Access Memory (SRAM), Electrically Erasable Programmable Read Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read Only Memory (PROM), Read Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0166] Of course, computing devices may also include other components, such as input / output interfaces, display components, communication components, etc.

[0167] Input / output interfaces provide interfaces between processing components and peripheral interface modules, which can be output devices, input devices, etc.

[0168] The communication components are configured to facilitate wired or wireless communication between computing devices and other devices.

[0169] The computing device can be a physical device or an elastic computing host provided by a cloud computing platform. In this case, the computing device can refer to a cloud server, and the aforementioned processing components, storage components, etc., can be basic server resources rented or purchased from the cloud computing platform.

[0170] This application also provides a computer storage medium storing a computer program, which, when executed by a computer, can perform the above-described functions. Figure 1 The embodiment shown is a method for detecting the degree of contamination of insulators based on multi-source spectral sensing.

[0171] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0172] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0173] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0174] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for detecting the degree of insulator pollution based on multi-source spectral sensing, characterized in that, include: Acquire spectral image data of the insulator at multiple focal lengths and scattering spectral data of the insulator surface; Extract the dirt layer thickness information from the spectral image data; The contamination layer thickness information and the scattering spectrum data are fused together, and the fusion result is input into a graph neural network. The graph neural network is used to determine the spatial relationship between position units, wherein the position units correspond to the spatial partition regions of the insulator. Based on the spatial correlation, three-dimensional pollution distribution data of the insulator is generated; The step of extracting dirt layer thickness information from the spectral image data includes: The optical depth parameters of the dirt layer are calculated based on the differences in spectral focus variation in the spectral image data. Based on the optical depth parameters and combined with the preset image sharpness variation pattern, dirt layer thickness information is generated; The step of generating dirt layer thickness information based on the optical depth parameters and a preset image sharpness variation pattern includes: The optical depth parameters are matched and mapped with the image sharpness variation pattern; Based on the matching mapping results, the quantified value of the dirt layer thickness is calculated using a linear interpolation method; The quantized values ​​of the contamination layer thickness of all location units are aggregated to form contamination layer thickness information.

2. The method according to claim 1, characterized in that, The process of fusing the dirt layer thickness information and the scattering spectral data, inputting the fusion result into a graph neural network, and determining the spatial relationships between location units through the graph neural network includes: The dirt layer thickness information and the scattering spectrum data are aligned according to position units to form a joint feature vector for each position unit; The joint feature vector is input into a graph neural network, and the target space connection weights between adjacent position units are calculated by the graph neural network. Based on the target spatial connection weights, the spatial association between location units is determined.

3. The method according to claim 2, characterized in that, The calculation of target spatial connection weights between adjacent position units using the graph neural network includes: Identify the physical connection topology between position units from the physical spatial structure of the insulator; The similarity of dirt thickness and spectral features between adjacent units are extracted from the joint feature vector to form a two-dimensional similarity index; The physical connection topology and the two-dimensional similarity index are fused into a spatial constraint factor; The spatial constraint factor is coupled with the joint feature vector to generate target spatial connection weights between adjacent position units.

4. The method according to claim 3, characterized in that, The coupling of the spatial constraint factor with the joint feature vector to generate target spatial connection weights between adjacent position units includes: The spatial constraint factors are converted into edge constraints of a graph neural network; The physical configuration features of the spatial constraint factor are injected into the joint feature vector to generate an enhanced feature vector; The graph neural network performs constraint-guided processing on the enhanced feature vector and the edge constraint conditions, and outputs the initial spatial connection weight values. Based on the pollution diffusion characteristics of the insulator, the initial spatial connection weight value is optimized and adjusted to generate the target spatial connection weight.

5. The method according to claim 1, characterized in that, The step of generating three-dimensional pollution distribution data for insulators based on the spatial correlation includes: Based on the spatial relationships, construct the contamination transfer paths between location units; The thickness of the contamination layer is calculated according to the contamination transmission path; The calculation results are mapped to a preset three-dimensional insulator model to generate three-dimensional pollution distribution data.

6. A system for detecting the degree of insulator pollution based on multi-source spectral sensing, characterized in that, include: The acquisition module is used to acquire spectral image data of the insulator at multiple focal lengths and scattering spectral data of the insulator surface; The extraction module is used to extract dirt layer thickness information from the spectral image data; A fusion module is used to fuse the contamination layer thickness information and the scattering spectrum data, input the fusion result into a graph neural network, and determine the spatial relationship between position units through the graph neural network, wherein the position units correspond to the spatial partition regions of the insulator. The generation module is used to generate three-dimensional pollution distribution data of the insulator based on the spatial correlation. The step of extracting dirt layer thickness information from the spectral image data includes: The optical depth parameters of the dirt layer are calculated based on the differences in spectral focus variation in the spectral image data. Based on the optical depth parameters and combined with the preset image sharpness variation pattern, dirt layer thickness information is generated; The step of generating dirt layer thickness information based on the optical depth parameters and a preset image sharpness variation pattern includes: The optical depth parameters are matched and mapped with the image sharpness variation pattern; Based on the matching mapping results, the quantified value of the dirt layer thickness is calculated using a linear interpolation method; The quantized values ​​of the contamination layer thickness of all location units are aggregated to form contamination layer thickness information.

7. A computing device, characterized in that, It includes a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are invoked and executed by the processing component to implement the insulator pollution level detection method based on multi-source spectral sensing as described in any one of claims 1 to 5.

8. A computer storage medium, characterized in that, The device contains a computer program that, when executed by a computer, implements a method for detecting the degree of insulator contamination based on multi-source spectral sensing as described in any one of claims 1 to 5.

Images