Insulator contamination detection method, electronic device, storage medium, and program product

Abstract

Embodiments of the present application provide an insulator contamination detection method, an electronic device, a storage medium and a program product. The method comprises: an electronic device acquiring a plurality of original images of a target insulator collected by hyperspectral imaging technology. The electronic device generates hyperspectral data of the target insulator using the plurality of original images. The electronic device inputs the hyperspectral data of the target insulator into a trained prediction model to predict a contamination degree index of the surface of the target insulator. The method achieves the effect of improving the prediction accuracy of the surface contamination degree.

Description

Technical Field

[0001] This application relates to the field of insulator pollution detection, and more particularly to an insulator pollution detection method, electronic equipment, storage medium, and program product. Background Technology

[0002] Insulators are critical equipment in power systems, used to support conductors and ensure insulation performance. However, during operation, contaminants such as salt, alkali, dust, or industrial pollutants can easily accumulate on the surface of insulators, leading to a decline in insulation performance and even causing flashover faults, threatening the safety of the power grid.

[0003] Traditional pollution detection technologies are divided into two categories: contact and non-contact. Contact methods, such as the equivalent salt density method and the leakage current method, require direct contact with the insulator, which is cumbersome to operate and easily affected by environmental interference. Non-contact methods, such as infrared imaging and ultraviolet imaging, are easy to implement, but are not sensitive to light pollution or dry conditions and have limited accuracy.

[0004] Therefore, there is an urgent need for an insulator pollution status assessment method that can balance detection accuracy and real-time performance and adapt to complex environmental conditions. Summary of the Invention

[0005] This application provides an insulator contamination detection method, electronic device, storage medium, and program product to improve the accuracy of insulator surface contaminant monitoring.

[0006] In a first aspect, embodiments of this application provide an insulator pollution detection method, including:

[0007] Multiple raw images of the target insulator were acquired using hyperspectral imaging technology; and the raw images were then used to process the target insulator to obtain hyperspectral data of the target insulator.

[0008] The hyperspectral data of the target insulator is input into a trained prediction model to predict the pollution level index of the surface of the target insulator; wherein the prediction model is a random forest model including multiple decision trees.

[0009] In one example, the step of inputting the hyperspectral data of the target insulator into a trained prediction model to predict the surface contamination level index of the target insulator includes:

[0010] The hyperspectral data of the target insulator is input into the prediction model corresponding to the pollution to predict the area ratio of the pollution on the surface of the target insulator.

[0011] The degree of contamination on the surface of the target insulator is determined based on the area percentage of each type of contaminant.

[0012] In one example, inputting the hyperspectral data of the target insulator into a pollution prediction model to predict the area ratio of pollution on the surface of the target insulator includes:

[0013] The hyperspectral data of the target insulator is divided into multiple sub-data, each sub-data including the spectral features corresponding to multiple pixels in the hyperspectral data;

[0014] Each piece of the sub-data is input into a decision tree of the prediction model to predict the proportion of dirt in the sub-data; wherein, the proportion of dirt represents the proportion of pixels covered by dirt in the sub-data out of all pixels.

[0015] Based on the pollution percentage predicted by each decision tree of the prediction model, the area percentage of pollution on the surface of the target insulator is determined.

[0016] In one example, determining the area percentage of contamination on the surface of the target insulator based on the contamination percentage predicted by each decision tree of the prediction model includes:

[0017] The area percentage of the contamination on the surface of the target insulator is determined based on the average of the contamination percentages predicted by each decision tree.

[0018] In one example, determining the pollution level index of the target insulator surface based on the area proportion of each type of contaminant includes:

[0019] Based on the preset weighting coefficient for each type of contamination, a weighted sum of the area proportions of each type of contamination is calculated to obtain a contamination level index of the surface of the target insulator.

[0020] In one example, processing the target insulator using the original image to obtain hyperspectral data of the target insulator includes:

[0021] The target insulator is modeled using the original image to obtain a modeled image of the target insulator;

[0022] The region of interest (ROI) is extracted from the modeling image to obtain the ROI image of the target insulator.

[0023] Hyperspectral data of the target insulator is constructed based on the spectral features of each pixel in the region of interest image.

[0024] In one example, constructing the hyperspectral data of the target insulator based on the spectral features of each pixel in the region of interest image includes:

[0025] The spectral features of each pixel in the region of interest image are input into the trained feature extraction model to obtain the feature data of each pixel of the target insulator.

[0026] The hyperspectral data of the target insulator is constructed based on the feature data of each pixel of the target insulator.

[0027] In one example, before modeling the target insulator using the original image, the method further includes:

[0028] The original image is subjected to black and white reference correction using preset standard black correction data and preset standard white correction data.

[0029] In one example, the acquisition of multiple raw images of the target insulator using hyperspectral imaging technology includes:

[0030] Obtain the height information of the target insulator;

[0031] Based on the height of the target insulator, multiple data acquisition planes are determined; each data acquisition plane is perpendicular to the target insulator.

[0032] Using a drone, the target insulator is photographed from all angles on each of the data acquisition planes to obtain multiple original images.

[0033] Secondly, embodiments of this application provide an insulator pollution detection device, comprising:

[0034] The acquisition module is used to acquire multiple original images of the target insulator using hyperspectral imaging technology; and to process the target insulator using the original images to obtain hyperspectral data of the target insulator.

[0035] The processing module is used to input the hyperspectral data of the target insulator into a trained prediction model to predict the pollution level index of the surface of the target insulator; wherein the prediction model is a random forest model including multiple decision trees.

[0036] In one example, the processing module is used to include:

[0037] The hyperspectral data of the target insulator is input into the prediction model corresponding to the pollution to predict the area ratio of the pollution on the surface of the target insulator.

[0038] The degree of contamination on the surface of the target insulator is determined based on the area percentage of each type of contaminant.

[0039] In one example, the processing module is used to include:

[0040] The hyperspectral data of the target insulator is divided into multiple sub-data, each sub-data including the spectral features corresponding to multiple pixels in the hyperspectral data;

[0041] Each piece of the sub-data is input into a decision tree of the prediction model to predict the proportion of dirt in the sub-data; wherein, the proportion of dirt represents the proportion of pixels covered by dirt in the sub-data out of all pixels.

[0042] Based on the pollution percentage predicted by each decision tree of the prediction model, the area percentage of pollution on the surface of the target insulator is determined.

[0043] In one example, the processing module is used to include:

[0044] The area percentage of the contamination on the surface of the target insulator is determined based on the average of the contamination percentages predicted by each decision tree.

[0045] In one example, the processing module is used to include:

[0046] Based on the preset weighting coefficient for each type of contamination, a weighted sum of the area proportions of each type of contamination is calculated to obtain a contamination level index of the surface of the target insulator.

[0047] In one example, the module is used for:

[0048] The target insulator is modeled using the original image to obtain a modeled image of the target insulator;

[0049] The region of interest (ROI) is extracted from the modeling image to obtain the ROI image of the target insulator.

[0050] Hyperspectral data of the target insulator is constructed based on the spectral features of each pixel in the region of interest image.

[0051] In one example, the module is used for:

[0052] The spectral features of each pixel in the region of interest image are input into the trained feature extraction model to obtain the feature data of each pixel of the target insulator.

[0053] The hyperspectral data of the target insulator is constructed based on the feature data of each pixel of the target insulator.

[0054] In one example, the module is used for:

[0055] The original image is subjected to black and white reference correction using preset standard black correction data and preset standard white correction data.

[0056] In one example, the module is used for:

[0057] Obtain the height information of the target insulator;

[0058] Based on the height of the target insulator, multiple data acquisition planes are determined; each data acquisition plane is perpendicular to the target insulator.

[0059] Using a drone, the target insulator is photographed from all angles on each of the data acquisition planes to obtain multiple original images.

[0060] Thirdly, embodiments of this application provide an electronic device, including: a memory and a processor;

[0061] The memory stores computer-executed instructions;

[0062] The processor executes computer execution instructions stored in the memory, causing the processor to perform the first aspect and / or various possible implementations of the first aspect as described above.

[0063] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the first aspect and / or various possible implementations of the first aspect.

[0064] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the first aspect and / or various possible implementations of the first aspect.

[0065] The insulator pollution detection method, electronic device, storage medium, and program product provided in this application acquire multiple original images of the target insulator using hyperspectral imaging technology, generate hyperspectral data of the target insulator using these multiple original images, and then predict the surface pollution level index of the target insulator by inputting the hyperspectral data of the target insulator into a trained prediction model. This improves the prediction accuracy of the surface pollution level and enhances the environmental adaptability and detection stability of the pollution detection. Attached Figure Description

[0066] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0067] Figure 1 A schematic diagram of the data structure for hyperspectral imaging provided in this application;

[0068] Figure 2 A schematic diagram of the data structure for the spectral curves provided in this application;

[0069] Figure 3 A flowchart illustrating the insulator pollution detection method provided in this application;

[0070] Figure 4 A flowchart illustrating the insulator pollution detection method provided in this application;

[0071] Figure 5 A flowchart illustrating the insulator pollution detection method provided in this application;

[0072] Figure 6 A flowchart illustrating the insulator pollution detection method provided in this application;

[0073] Figure 7 This is a schematic diagram of the insulator pollution detection device provided in this application;

[0074] Figure 8 A schematic diagram of the structure of the electronic device provided in this application.

[0075] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0076] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0077] Insulators are key components in power systems, playing a vital role in supporting conductors and ensuring electrical insulation. During actual operation, various contaminants inevitably accumulate on the surface of insulators.

[0078] Insulator contamination is generally caused by natural pollution, icing and snow accumulation, and industrial pollution. Natural pollution includes dust, saline-alkali soil, seawater or petrel pollutants, and bird droppings. Industrial pollution includes exhaust fumes from chemical plants, smelters, and thermal power plants; dust from cement plants, coal mines, and other mining operations; and acidic water mist from circulating water cooling towers or spray pools.

[0079] The continuous accumulation of these contaminants can significantly reduce the electrical performance of insulators. When a critical value is reached, it may cause creepage or even flashover accidents, seriously affecting the reliability of power grid operation. With the continuous development of power systems, insulator pollution has gradually become one of the key factors affecting power grid reliability.

[0080] To quantify the degree of insulator contamination, the industry generally uses Equivalent Salt Deposit Density (ESDD) and Non-soluble Deposit Density (NSDD) as core indicators.

[0081] ESDD characterizes the content of soluble conductive substances and is defined as the amount of sodium chloride (NaCl) dissolved in a given deionized water that has the same volume conductivity as natural deposits washed from a given surface of an insulator, divided by the area of ​​that surface, and is generally expressed in mg / cm2.

[0082] NSDD reflects the density of insoluble deposits and is defined as the amount of insoluble residue washed off a given surface of an insulator divided by the area of ​​that surface, typically expressed in mg / cm².

[0083] Studies have shown that insulator flashover voltage decreases with increasing salt density, dust density, and fog conductivity. That is, insulator flashover voltage exhibits a significant negative correlation with factors such as ESDD, NSDD, and ambient humidity. However, the effects of salt density and dust density on flashover voltage are independent of each other. Therefore, effective monitoring and assessment of insulator pollution levels are of great significance.

[0084] However, existing pollution detection technologies have significant limitations in practical applications: traditional contact measurement methods are cumbersome and destructive, while non-contact detection technologies are easily affected by environmental conditions. Therefore, developing an efficient, accurate, and adaptable insulator pollution condition monitoring technology is of great practical significance for ensuring the safe and stable operation of the power grid.

[0085] Therefore, this application proposes a method for detecting insulator pollution. To adapt to the development trend of intelligent and automated power grids and improve the operation and maintenance level of power transmission and transformation, this application utilizes spectroscopic technology to measure the degree of insulator pollution, taking into account the currently widespread use of drone technology in power grids. This measurement method is quick, flexible, low-cost, has no special requirements for the measurement environment, and yields high accuracy.

[0086] The field of spectral imaging can be divided into multispectral imaging, hyperspectral imaging, and hyperspectral imaging.

[0087] Hyperspectral Imaging Systems (HIS) can acquire hundreds of consecutive narrow-band spectral information, capturing subtle spectral changes, and are therefore widely used in scenarios requiring target classification and analysis. Furthermore, spectral imaging images can reflect the size, shape, and color of samples. Spectral information can also reflect internal properties such as physical structure, chemical composition, and their content. Due to differences in the composition and structure of substances, there is a selective absorption and reflection of photons of different wavelengths within the substance; therefore, the reflectance spectrum of a substance exhibits a "fingerprint" effect.

[0088] This hyperspectral imaging can generate three-dimensional (3D) data. This 3D data can be used as follows: Figure 1 As shown, this includes two spatial dimensions, x and y. The spectral dimensions are shown. Taking the pixels corresponding to x1 and y1 coordinates as an example, in... In terms of dimensions, its values ​​can form as follows Figure 2 The spectral curve shown is illustrated. The horizontal axis of the spectral curve represents the spectral band. The vertical axis of the spectral curve represents the numerical values ​​measured within that spectral band.

[0089] Based on the three-dimensional data generated by this hyperspectral imaging, it can be used as follows: Figure 3 The steps shown enable the training and use of the prediction model, thereby achieving accurate detection of insulator contamination.

[0090] Specifically, after acquiring the raw 3D data through HIS, the electronic device can optimize the raw data by using black and white reference correction. Furthermore, the electronic device can model the raw data and extract the Region of Interest (ROI) to obtain the hyperspectral data of the insulator.

[0091] The hyperspectral data can include two-dimensional image features. Each pixel in these image features represents the insulator. The hyperspectral data can also include one-dimensional spectral features. That is, each pixel can correspond to the entire spectrum, forming a spectral curve.

[0092] The electronic device can divide the collected and labeled hyperspectral data of insulators into training and test sets. After preprocessing the hyperspectral data, the electronic device can use the hyperspectral data of the insulators in the training set to train the constructed prediction model, resulting in a trained prediction model. The electronic device can then use the hyperspectral data of the insulators in the test set to validate the trained prediction model.

[0093] Furthermore, the electronic device can input the hyperspectral data of the target insulator actually collected into the verified prediction model, thereby enabling the application of the prediction model and the acquisition of the pollution level index of the target insulator.

[0094] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0095] Figure 4 A flowchart illustrating the insulator pollution detection method provided in this application is shown below. Figure 4 As shown, the method includes:

[0096] S101. Multiple original images of the target insulator are acquired using hyperspectral imaging technology; and the original images are used to process the target insulator to obtain hyperspectral data of the target insulator.

[0097] For example, an electronic device uses a drone equipped with a hyperspectral imaging system to capture and acquire multiple raw images of the target insulator. Based on these raw images, the electronic device can construct hyperspectral data of the target insulator using a spectral reconstruction algorithm.

[0098] In one example, the hyperspectral imaging system may be a pushbroom imaging spectrometer, a staring imaging spectrometer, or a snapshot imaging spectrometer.

[0099] In one example, the hyperspectral imaging system can be mounted on a drone to collect data on insulators on a power tower in the field.

[0100] In one example, during the acquisition process, the electronic device controls the hyperspectral camera to acquire spectral data within a preset wavelength range with a preset scanning step size and exposure time, while simultaneously recording the spatial coordinate information of each pixel.

[0101] In one example, the hyperspectral data contains two-dimensional spatial information and one-dimensional spectral information. The three-dimensional data structure formed by combining the two-dimensional spatial information and the one-dimensional spectral information contains reflectance information for each pixel in tens to hundreds of consecutive bands.

[0102] In one example, an electronic device can stitch and integrate the hyperspectral data by preprocessing multiple original images of the target insulator, such as radiometric correction, geometric correction, and noise suppression, to obtain the final hyperspectral data of the target insulator.

[0103] S102. Input the hyperspectral data of the target insulator into the trained prediction model to predict the surface contamination level of the target insulator. The prediction model is a random forest model that includes multiple decision trees.

[0104] For example, the electronic device inputs the pre-processed hyperspectral data into a pre-trained random forest prediction model to predict the degree of contamination and obtain a contamination index of the surface of the target insulator.

[0105] In one example, the prediction model is a random forest (RF) model. The bootstrap resampling used in the random forest algorithm involves random sampling with replacement, repeated n times to form n sample sets. Each sample set generates n decision trees, and these n decision trees are used to build the "forest" in parallel. Finally, each decision tree votes independently on the input test sample, and the final voting result is the identification result of the contaminant component.

[0106] In one example, each random forest model consists of multiple decision trees. Optionally, the number of decision trees can be n. For example, n can be 5, 8, 10, etc.

[0107] In one example, the electronic device can employ a parallel computing architecture to accelerate the prediction process by simultaneously executing prediction tasks for multiple decision trees using a multi-core CPU or GPU.

[0108] In one example, the random forest model can obtain a final prediction by constructing multiple decision trees and taking the average or majority vote of their predictions. For instance, this process can be as follows: Figure 5 As shown.

[0109] In one example, by constructing multiple decision trees, it is possible to process large-scale data of thousands of dimensions without feature selection or deletion. It can maintain a balance of classification errors even when the distribution of class sizes is unbalanced, making it more robust than a single decision tree and exhibiting good generalization performance.

[0110] The insulator pollution detection method provided in this application acquires multiple original images of the target insulator using hyperspectral imaging technology, and then uses these multiple original images to generate hyperspectral data of the target insulator. Furthermore, by inputting the hyperspectral data of the target insulator into a trained prediction model, the method predicts the pollution level index of the target insulator's surface. This achieves accurate analysis of high-latitude, nonlinear hyperspectral data, improves the prediction accuracy of surface pollution level, and enhances the environmental adaptability and detection stability of the pollution detection method.

[0111] In one example, the process of acquiring the original image in step S101 may include:

[0112] S1011. Obtain the height information of the target insulator.

[0113] For example, the electronic device can first obtain the height information of the target insulator.

[0114] In one example, electronic devices can obtain the height information of the target insulator by real-time ranging using a lidar (LiDAR) mounted on a drone.

[0115] In another example, the electronic device can acquire an image of the target insulator using a visual sensor such as a binocular camera, and determine the height information of the target insulator based on the image.

[0116] For example, electronic devices can determine the height information of a target insulator using a monocular camera and the shooting position, through an ensemble algorithm.

[0117] For example, electronic devices can use binocular cameras to perform 3D modeling to determine the height information of the target insulator.

[0118] In another example, the electronic device can read the specification parameters of the insulator from a pre-stored database of power grid equipment to determine the height information of the target insulator.

[0119] In one example, the height information of the target insulator indicates the vertical distance from the bottom to the top of the target insulator.

[0120] S1012. Based on the height of the target insulator, determine multiple data acquisition planes. The data acquisition planes are planes perpendicular to the target insulator.

[0121] For example, the electronic device automatically plans the optimal data acquisition planar distribution scheme based on the precisely measured height information of the target insulator and the fixed parameters collected from the hyperspectral information.

[0122] In one example, the data acquisition plane is a virtual imaging plane perpendicular to the target insulator. The drone can then use this data acquisition plane to perform panoramic imaging of the insulator.

[0123] In one example, the electronic device may use an equal-spacing partitioning method or an adaptive density partitioning method to determine the number and location of data acquisition planes.

[0124] In one example, the electronic device can determine the number of sampling planes based on the divisor of the height of the target insulator and the preset spacing.

[0125] In another example, the electronic device can determine the number of sampling planes based on the height of the target insulator and a preset mapping table.

[0126] For example, when the height of the target insulator is less than a preset threshold, two acquisition planes can be set, one above and one below. Alternatively, when the height of the target insulator is greater than or equal to the preset threshold, three acquisition planes can be set, one above, one in the middle, and one below.

[0127] In one example, the arrangement of multiple acquisition planes ensures that the surface information of the target insulator can be accurately acquired, thereby improving the processing effect of subsequent data.

[0128] S1013. Using a drone, take panoramic photos of the target insulator on each data acquisition plane to obtain multiple raw images.

[0129] For example, after determining multiple data acquisition planes, the electronic device can control a drone equipped with a hyperspectral imaging system to perform an automated panoramic imaging task according to the preset acquisition planes, and obtain multiple raw images corresponding to the acquisition planes.

[0130] For example, the electronic device can acquire two raw images from each acquisition plane. These two raw images can be located on opposite sides of the target insulator. Alternatively, the electronic device can acquire three raw images from each acquisition plane. In these three raw images, any two images must be captured at a 120-degree angle. Another example is that the electronic device can acquire four raw images from each acquisition plane. These four raw images can be located in the front, back, left, and right directions of the target insulator. And so on.

[0131] In this embodiment, by setting multiple acquisition planes based on the height information of the target insulator and acquiring multiple original images based on each acquisition plane, the original images of the target insulator are acquired, providing more accurate and comprehensive original data for subsequent data processing.

[0132] In one example, before modeling the target insulator using the original image in step S101, the electronic device also needs to correct the original image to improve its usability. The specific process includes:

[0133] S1014. Perform black and white reference correction on the original image using preset standard black correction data and preset standard white correction data.

[0134] For example, the electronic device performs radiometric calibration on the acquired raw image using preset standard black correction data and standard white correction data to obtain the corrected raw image.

[0135] In one example, the electronic device can acquire an image taken of a whiteboard with a reflectivity close to 1, and use this image as standard white correction data. Optionally, this standard white correction data can be denoted as... The standard white correction data can be considered the upper limit of this spectral data.

[0136] In one example, the electronic device can acquire an image captured by a closed lens and use this image as standard black correction data. Optionally, this standard black correction data can be denoted as... This standard black correction data can serve as the lower limit for spectral data.

[0137] In one example, the correction formula for the original image can be written as:

[0138]

[0139] in, This is the original image. This is the original image after correction.

[0140] In this embodiment, black-and-white reference correction is used to eliminate sensor dark current noise and illumination inhomogeneity in hyperspectral imaging, thereby improving the reliability of the original image.

[0141] In one example, the process of generating hyperspectral data of the target insulator from the original image in step S101 includes:

[0142] S1015. Model the target insulator using the original image to obtain the modeled image of the target insulator.

[0143] For example, the electronic device processes the original image sequence using multi-view 3D reconstruction technology to generate a 3D model of the target insulator.

[0144] In one example, the electronic device may employ a sparse reconstruction based on Structure from Motion (SfM) combined with a dense reconstruction process using Multi-View Stereo (MVS). Alternatively, the electronic device may be modeled using a deep learning-based 3D reconstruction network.

[0145] In one example, during the modeling process, electronic devices can simultaneously align hyperspectral data and three-dimensional geometric information in the original image to establish a multidimensional data association that includes spatial coordinates and spectral features.

[0146] In one example, the electronic device can unfold a three-dimensional image of the target insulator onto a two-dimensional plane after modeling is completed.

[0147] In one example, this modeling process can be achieved by stitching together multiple original images.

[0148] S1015. Extract the region of interest from the modeling image to obtain the region of interest image where the target insulator is located.

[0149] For example, electronic devices automatically identify and extract the Region of Interest (ROI) of a target insulator based on prior knowledge base and image feature analysis.

[0150] In one example, the region of interest image may include pixels of the target insulator, and the region of interest image may exclude pixels other than those of the target insulator.

[0151] In one example, the region of interest extraction can be achieved based on shape template matching. Alternatively, the region of interest extraction can be achieved based on deep learning semantic segmentation. Or, the region of interest extraction can be determined based on spectral feature clustering.

[0152] In one example, the electronic device performs edge refinement and region verification to ensure the accuracy of the image boundaries of the region of interest while preserving the complete spectral features.

[0153] In one example, for insulators with multi-skirt structures, the electronic equipment can also achieve independent segmentation of individual skirts, creating conditions for local pollution analysis.

[0154] S1016. Construct hyperspectral data of the target insulator based on the spectral characteristics of each pixel in the region of interest image.

[0155] For example, an electronic device can acquire the spectral features of each pixel in the region of interest image. The region of interest image constitutes the y-axis and y-axis dimensions of the hyperspectral data. Furthermore, the spectral features constitute the... The dimension of the axis.

[0156] In this embodiment, by stitching and integrating multiple original images and extracting regions of interest, the final hyperspectral data of the target insulator is obtained. This achieves the deduplication and integration of effective data in the original images, thereby improving the efficiency of using the hyperspectral data of the target insulator in subsequent applications.

[0157] In one example, constructing the hyperspectral data of the target insulator in step S1016 may further include feature extraction from spectral features directly acquired from the original image. This process includes:

[0158] S10161. Input the spectral features of each pixel in the region of interest image into the trained feature extraction model to obtain the feature data of each pixel of the target insulator.

[0159] In one example, each pixel of the region of interest image may include spectral features. These spectral features contain feature data from the full spectrum.

[0160] In one example, the full-spectrum feature data contains a large amount of redundant information due to the high correlation between adjacent data.

[0161] For example, electronic devices can extract effective features from full-spectrum feature data and eliminate irrelevant information through feature extraction, thereby reducing the amount of data.

[0162] Electronic devices can input the spectral features of each pixel in the region of interest image into a trained feature extraction model to obtain effective feature data with reduced data volume.

[0163] In one example, the feature extraction process can be implemented using a competitive adaptive reweighted sampling (CARS) algorithm, which selects the best combination of wavelengths from multi-component spectral data based on the principle of "survival of the fittest".

[0164] In one example, the feature extraction process can also be implemented using different architectures such as Convolutional Neural Networks (CNN), Spectral-Attention Network, Graph Neural Networks (GNN), and Principal Components Analysis (PCA).

[0165] S10162. Construct hyperspectral data of the target insulator based on the feature data of each pixel of the target insulator.

[0166] For example, an electronic device can use the feature data of each pixel to replace the spectral features of that pixel to generate the final hyperspectral data.

[0167] In this embodiment of the application, by filtering the feature data of the full spectrum of the target insulator, a method is used to obtain effective features with less data, thereby improving the efficiency of subsequent calculations.

[0168] In one example, step S102 involves inputting the hyperspectral data of the target insulator into a trained prediction model to predict an index of the surface contamination level of the target insulator, including:

[0169] S1021. Input the hyperspectral data of the target insulator into the prediction model corresponding to the pollution to predict the area ratio of pollution on the surface of the target insulator.

[0170] For example, the electronic device can input the feature data of each pixel in the hyperspectral data of the target insulator into the prediction model corresponding to the pollution, and predict whether the pixel is the pollution. Based on the prediction result of each pixel, the electronic device can determine the area ratio of pollution on the surface of the target insulator.

[0171] In one example, the model can be implemented using deep learning-based semantic segmentation networks such as U-Net and DeepLabv3+, or traditional machine learning classifiers such as support vector machines and random forests.

[0172] In one example, the electronic device uses a parallel computing architecture to perform pixel-by-pixel predictions on the hyperspectral data of the target insulator, generating a pollution distribution probability map.

[0173] In one example, the prediction result for each pixel indicates the probability that the contaminant is present at that pixel's location. Based on this probability and a preset threshold, a judgment is made regarding the presence or absence of the contaminant.

[0174] In one example, the natural contaminants on the insulator surface are diverse and complex in composition. For instance, coal smoke pollution in air pollution contains high levels of SO2, NOx, SiO2, Al2O3, and Fe2O3, while industrial and saline-alkali areas have high NaCl content. Therefore, this application sets up six types of contaminants: ferric sulfate, calcium carbonate, alumina, sodium chloride, calcium sulfate, and kaolin. A corresponding model is set up for each type of contaminant to make predictions.

[0175] In one example, the electronic device can input the hyperspectral data of the target insulator into six prediction models corresponding to six types of pollution, and predict the area proportion and distribution information of the six types of pollution.

[0176] S1022. Determine the pollution level index of the target insulator surface based on the area proportion of each type of contaminant.

[0177] For example, the electronic device calculates the comprehensive index of the target insulator based on the area ratio of various types of contaminants and in conjunction with a pre-established contamination degree calculation model, thereby determining the contamination degree index of the target insulator's surface.

[0178] For example, the comprehensive calculation algorithm can be a weighted summation method that can take into account the influence coefficients of different types of pollution.

[0179] For example, the comprehensive calculation algorithm can be a fuzzy logic-based method, which can handle the uncertainty of the calculation standard.

[0180] For example, the comprehensive calculation algorithm can be a deep learning-based method that can directly map the area percentage to the final pollution level.

[0181] In one example, during the calculation process, the electronic device will refer to historical data and environmental factors such as humidity and distance from pollution sources to correct the results, thereby improving the reliability of the calculation.

[0182] For example, electronic devices can generate correction weights based on information such as humidity and distance. The electronic device can then use the product of these correction weights and the pollution level indicator to determine the corrected pollution level indicator.

[0183] In one example, electronic devices can determine the severity of various types of pollution in a target insulator based on the pollution level index and the index ranges corresponding to the four predefined levels of cleanliness, light pollution, moderate pollution, and heavy pollution.

[0184] In one example, the electronic device also generates a visual map of pollution distribution, with annotations on the pollution distribution map.

[0185] This application's embodiments achieve refined quantitative calculation of insulator surface contamination through high-precision pixel-level prediction. Furthermore, an intelligent comprehensive calculation model fully considers the synergistic effects of various types of contamination, making the assessment of contamination levels more scientific and reliable. Additionally, this application significantly improves detection efficiency through automated calculation processes, meeting the needs of large-scale power grid inspections and enhancing the reliability of power grid operation.

[0186] In one example, when the prediction model is a random forest model, the random forest model may include multiple decision trees. In step S1021, the hyperspectral data of the target insulator is input into each decision tree of the prediction model corresponding to the pollution, and the area ratio of pollution on the surface of the target insulator is predicted. The specific process may include:

[0187] S10211. Divide the hyperspectral data of the target insulator into multiple sub-data, each sub-data including the spectral features corresponding to multiple pixels in the hyperspectral data.

[0188] For example, an electronic device can divide the hyperspectral data of a target insulator into multiple sub-data sets, where a single pixel is treated as a data point. Each sub-data set includes spectral features corresponding to multiple pixels in the hyperspectral data.

[0189] In one example, the electronic device may divide the pixels in the hyperspectral data of the target insulator into multiple parts using an average distribution method.

[0190] In one example, an electronic device can group adjacent pixels into the same subdata based on their positional relationship.

[0191] In another example, the electronic device can randomly assign pixels to various sub-data.

[0192] S10212. Input each subset of data into a decision tree of the prediction model to predict the percentage of contamination in the subset. The percentage of contamination represents the proportion of pixels covered by contamination in the subset.

[0193] For example, an electronic device can input each data set into a decision tree and use the decision tree to make a decision about each pixel in that data set, determining whether the pixel is dirty. Based on this decision, the electronic device can statistically determine the percentage of dirty pixels in a data set.

[0194] In one example, the number of sub-data points can be the same as the number of decision trees. The electronic device can input each sub-data point into a separate decision tree to predict the percentage of contamination in each sub-data point. For example, when including sub-data points 1, 2, and 3, and decision trees 1, 2, and 3, the electronic device can input sub-data point 1 into decision tree 1, sub-data point 2 into decision tree 2, and sub-data point 3 into decision tree 3.

[0195] In another example, the number of decision trees can exceed the number of subdata points. An electronic device can feed a subset of the subdata points into different decision trees, each predicting a percentage of contamination. For instance, when including subdata points 1 and 2, and decision trees 1, 2, 3, and 4, subdata point 1 can be input into decision trees 1 and 2, and subdata point 2 into decision trees 3 and 4.

[0196] S10213. Based on the pollution percentage predicted by each decision tree of the prediction model, determine the area percentage of pollution on the surface of the target insulator.

[0197] For example, the electronic device can summarize the prediction results of each decision tree to obtain the area percentage of surface contamination of the target insulator as a whole.

[0198] In one example, the electronic device can mark dirty pixels in the region of interest map based on the judgment result of whether each pixel is dirty, and complete the drawing of the dirt distribution map.

[0199] In one example, when the number of sub-data points can match the number of decision trees, the electronic device can determine the number of dirty pixels in each sub-data point based on the proportion of dirt in that sub-data point. Furthermore, the electronic device can count the total number of dirty pixels on the surface of the target insulator. The electronic device can then determine the area proportion of dirt on the surface of the target insulator based on the ratio of the total number of pixels on the surface of the target insulator to the total number of dirty pixels.

[0200] In another example, when the number of decision trees exceeds the number of sub-data points, the electronic device can first determine the actual contamination percentage of a sub-data point based on the average contamination percentage of the same sub-data point. This actual contamination percentage is a more accurate value after verification by multiple decision trees. Subsequently, the electronic device can calculate the area percentage of contamination on the surface of the target insulator by counting the total number of contaminated pixels on the surface.

[0201] In this embodiment, decision-making and summarization are performed using decision trees, which avoids the imbalance problem of individual decision trees and improves the accuracy of decision results.

[0202] In one example, the step of summarizing the area percentage of surface contamination in step S10213 includes:

[0203] S102131. Determine the area percentage of contamination on the surface of the target insulator based on the average of the contamination percentages predicted by each decision tree.

[0204] For example, since each decision tree makes the same number of decisions, the electronic device can use the mean calculation method to calculate the area ratio of dirt on the overall surface of the target insulator.

[0205] In this embodiment, by calculating the mean, the bias that may be generated by a single decision tree is effectively suppressed, which significantly enhances the robustness of the system in complex environments and improves the accuracy of the prediction results.

[0206] In one example, the process of summarizing and calculating the pollution level index of electronic devices in step S1022 may include:

[0207] S10221. Based on the preset weighting coefficients for each type of contamination, calculate the weighted sum of the area proportions of various contaminations to obtain the contamination level index of the target insulator surface.

[0208] For example, electronic devices can be pre-set with weighting coefficients for each type of contaminant. The electronic device can obtain a final contamination level index by weighted summation of the area proportions of various contaminants.

[0209] In one example, the electronic device can perform a weighted summation based on the predicted results of each type of stain for each pixel to obtain a comprehensive pollution index for each pixel, and then draw a schematic diagram of the comprehensive pollution distribution of the target insulator.

[0210] In one example, target insulators with a greater impact can have a higher weighting coefficient, while target insulators with a smaller impact can have a lower weighting coefficient, thus better compensating for the impact of different types of pollution on the insulator's lifespan.

[0211] In this embodiment of the application, by using weighted summation, the impact of different types of contamination can be reduced, thereby improving the accuracy of the final judgment result.

[0212] In one example, Figure 6 The entire process of measuring insulator contamination using a hyperspectral drone is shown.

[0213] For example, an electronic device can first train a prediction model.

[0214] During the training phase, the electronic device can acquire spectral data from multiple insulators and treat the spectral data of each insulator as a sample. The electronic device can then divide this sample into two parts at a ratio of 70% and 30%. The 70% part serves as training data, and the 30% part serves as test data.

[0215] Electronic devices can perform CARS feature extraction on training data. Subsequently, the extracted data is used to train the RF model, resulting in the final random forest model.

[0216] Electronic devices can use test data to validate the random forest model and apply the validated random forest model to the model usage phase.

[0217] During the usage phase, after drawing the drone's inspection route, the electronic equipment acquires the raw images collected by the drone during the inspection process. These raw images are the hyperspectral imaging data.

[0218] Electronic devices can process the original image to obtain spectral data and then extract CARS features. Subsequently, a random forest model can be used to classify the extracted data, determining whether each pixel is contaminated. Based on the judgment results for all pixels, the proportion of contaminated pixels can be statistically determined, and the final contamination level index can be established.

[0219] Figure 7 This is a schematic diagram of the insulator pollution detection device provided in this application, as shown below. Figure 7 As shown, the insulator pollution detection device 200 provided in this embodiment includes:

[0220] The acquisition module 201 is used to acquire multiple original images of the target insulator using hyperspectral imaging technology; and to process the target insulator using the original images to obtain hyperspectral data of the target insulator.

[0221] The processing module 202 is used to input the hyperspectral data of the target insulator into the trained prediction model to predict the pollution level index of the surface of the target insulator; wherein, the prediction model is a random forest model including multiple decision trees.

[0222] In one example, processing module 202 is used to include:

[0223] The hyperspectral data of the target insulator is input into the prediction model corresponding to the pollution to predict the area ratio of pollution on the surface of the target insulator.

[0224] The degree of contamination on the surface of the target insulator is determined based on the area proportion of each type of contaminant.

[0225] In one example, processing module 202 is used to include:

[0226] The hyperspectral data of the target insulator is divided into multiple sub-data, each of which includes the spectral features corresponding to multiple pixels in the hyperspectral data.

[0227] Each subset of data is input into a decision tree of the prediction model to predict the percentage of contamination in the subset; where the percentage of contamination represents the proportion of pixels covered by contamination in the subset.

[0228] Based on the pollution percentage predicted by each decision tree in the prediction model, the area percentage of pollution on the surface of the target insulator is determined.

[0229] In one example, processing module 202 is used to include:

[0230] The area percentage of contamination on the surface of the target insulator is determined based on the average of the contamination percentages predicted by each decision tree.

[0231] In one example, processing module 202 is used to include:

[0232] Based on the preset weighting coefficients for each type of contamination, the weighted sum of the area proportions of each type of contamination is calculated to obtain the contamination level index of the target insulator surface.

[0233] In one example, module 201 is used for:

[0234] The target insulator is modeled using the original image to obtain a modeled image of the target insulator;

[0235] Region of interest (ROI) extraction is performed on the modeling image to obtain the ROI image of the target insulator;

[0236] Hyperspectral data of the target insulator is constructed based on the spectral features of each pixel in the region of interest image.

[0237] In one example, module 201 is used for:

[0238] The spectral features of each pixel in the region of interest image are input into the trained feature extraction model to obtain the feature data of each pixel of the target insulator.

[0239] Hyperspectral data of the target insulator is constructed based on the feature data of each pixel of the target insulator.

[0240] In one example, module 201 is used for:

[0241] The original image is subjected to black and white reference correction using preset standard black correction data and preset standard white correction data.

[0242] In one example, module 201 is used for:

[0243] Obtain the height information of the target insulator;

[0244] Based on the height of the target insulator, multiple data acquisition planes are determined; each data acquisition plane is perpendicular to the target insulator.

[0245] Using drones, the target insulators are photographed from all angles on each data acquisition plane to obtain multiple raw images.

[0246] The insulator pollution detection device provided in this embodiment can perform the method provided in the above method embodiment. Its implementation principle and technical effect are similar, and will not be described in detail here.

[0247] Figure 8 A schematic diagram of the structure of the electronic device provided in this application. Figure 8 As shown, the electronic device 300 provided in this embodiment includes at least one processor 301 and a memory 302. Optionally, the electronic device 300 further includes a communication component 303. The processor 301, memory 302, and communication component 303 are connected via a bus 304.

[0248] In a specific implementation, at least one processor 301 executes computer execution instructions stored in memory 302, causing at least one processor 301 to perform the above-described method.

[0249] The specific implementation process of processor 301 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0250] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0251] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0252] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0253] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0254] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the above-described method.

[0255] The aforementioned readable storage medium can be implemented by 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. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0256] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an application-specific integrated circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0257] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0258] 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 units can be selected to achieve the purpose of this embodiment according to actual needs.

[0259] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0260] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory, random access memory, magnetic disks, or optical disks.

[0261] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0262] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A method for detecting pollution in insulators, characterized in that, include: Multiple raw images of the target insulator were acquired using hyperspectral imaging technology; and the raw images were then used to process the target insulator to obtain hyperspectral data of the target insulator. The hyperspectral data of the target insulator is input into a trained prediction model to predict the pollution level index of the surface of the target insulator; wherein the prediction model is a random forest model including multiple decision trees.

2. The method according to claim 1, characterized in that, The step of inputting the hyperspectral data of the target insulator into the trained prediction model to predict the surface contamination level index of the target insulator includes: The hyperspectral data of the target insulator is input into the prediction model corresponding to each type of contaminant to predict the area ratio of the contaminant on the surface of the target insulator. The degree of contamination on the surface of the target insulator is determined based on the area percentage of each type of contaminant.

3. The method according to claim 2, characterized in that, The step of inputting the hyperspectral data of the target insulator into the pollution prediction model to predict the area ratio of pollution on the surface of the target insulator includes: The hyperspectral data of the target insulator is divided into multiple sub-data, each sub-data including the spectral features corresponding to multiple pixels in the hyperspectral data; Each piece of the sub-data is input into a decision tree of the prediction model to predict the proportion of dirt in the sub-data; wherein, the proportion of dirt represents the proportion of pixels covered by dirt in the sub-data out of all pixels. Based on the pollution percentage predicted by each decision tree of the prediction model, the area percentage of pollution on the surface of the target insulator is determined.

4. The method according to claim 3, characterized in that, Determining the area percentage of contamination on the surface of the target insulator based on the contamination percentage predicted by each decision tree of the prediction model includes: The area percentage of the contamination on the surface of the target insulator is determined based on the average of the contamination percentages predicted by each decision tree.

5. The method according to claim 2, characterized in that, The determination of the surface contamination level index of the target insulator based on the area proportion of each type of contaminant includes: Based on the preset weighting coefficient for each type of contamination, a weighted sum of the area proportions of each type of contamination is calculated to obtain a contamination level index of the surface of the target insulator.

6. The method according to any one of claims 1-5, characterized in that, The process of processing the target insulator using the original image to obtain the hyperspectral data of the target insulator includes: The target insulator is modeled using the original image to obtain a modeled image of the target insulator; The region of interest (ROI) is extracted from the modeling image to obtain the ROI image of the target insulator. Hyperspectral data of the target insulator is constructed based on the spectral features of each pixel in the region of interest image.

7. The method according to claim 6, characterized in that, The step of constructing hyperspectral data of the target insulator based on the spectral features of each pixel in the region of interest image includes: The spectral features of each pixel in the region of interest image are input into the trained feature extraction model to obtain the feature data of each pixel of the target insulator. The hyperspectral data of the target insulator is constructed based on the feature data of each pixel of the target insulator.

8. The method according to claim 6, characterized in that, Before modeling the target insulator using the original image, the method further includes: The original image is subjected to black and white reference correction using preset standard black correction data and preset standard white correction data.

9. The method according to any one of claims 1-5, characterized in that, The multiple original images of the target insulator acquired through hyperspectral imaging technology include: Obtain the height information of the target insulator; Based on the height of the target insulator, multiple data acquisition planes are determined; each data acquisition plane is perpendicular to the target insulator. Using a drone, the target insulator is photographed from all angles on each of the data acquisition planes to obtain multiple original images.

10. An insulator pollution detection device, characterized in that, include: The acquisition module is used to acquire multiple raw images of the target insulator using hyperspectral imaging technology; The original image is then used to process the target insulator to obtain hyperspectral data of the target insulator; The processing module is used to input the hyperspectral data of the target insulator into a trained prediction model to predict the pollution level index of the surface of the target insulator; wherein the prediction model is a random forest model including multiple decision trees.

11. An electronic device, characterized in that, include: Memory, processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the processor to perform the method as described in any one of claims 1-9.

12. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1-9.

13. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method described in any one of claims 1-9.

Images