Insulator contamination component analysis method and analysis system based on deep learning

Abstract

The application relates to a kind of insulator pollution component analysis method and analysis system based on deep learning, method includes extracting insulator object image in image;The insulator object image is carried out gray processing, obtains insulator object gray image;Using sub-insulator object gray image cuts insulator spectral image, obtains sub-insulator spectral image;Analysis reference line is established on sub-insulator spectral image and using the pixel point on analysis reference line establishes analysis curve;Using autocorrelation calculation and fast fourier transform mode process analysis curve and obtain power spectrum and according to power spectrum select the pollution component identification model based on deep learning and determine the pollution component that sub-insulator object gray image includes.The insulator pollution component analysis method and analysis system based on deep learning disclosed in the application determine the pollution on insulator and its component by intelligent identification to the pollution area on insulator and model analysis mode to the pollution area component.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method and system for analyzing the pollution components of insulators based on deep learning. Background Technology

[0002] Insulator contamination refers to the degree of contamination on the surface of an insulator. Two commonly used evaluation indicators are equivalent salt density and equivalent ash density. Equivalent salt density refers to the amount of NaCl (mg / cm2) equivalent to the content of conductive substances in the contaminants attached to each square centimeter of the insulator surface. Equivalent salt density characterizes the conductivity of the insulator surface after the contaminants have been fully dissolved. An increase in equivalent salt density will significantly reduce the flashover voltage of the insulator.

[0003] Currently, the main methods for implementing these two evaluation indicators are to directly sample the insulators or collect insulator model samples. These methods are lacking in terms of ease of implementation, sample accuracy, and timeliness, and require technological updates.

[0004] A suitable approach is to use spectral imaging combined with machine processing to determine the contaminant composition on the insulator. This method enables long-distance, non-contact sampling and is superior in terms of ease of implementation, sample accuracy, and timeliness.

[0005] However, spectral images have problems such as large data volume and interference, and the contaminants on insulators also have the characteristic of not having a predetermined shape. Further research is needed on how to analyze them. Summary of the Invention

[0006] This invention provides a method and system for analyzing the pollution components of insulators based on deep learning. The method determines the pollution and its components on the insulator by intelligently identifying the polluted areas on the insulator and performing model analysis on the components of the polluted areas.

[0007] The above-mentioned objective of the present invention is achieved through the following technical solution:

[0008] In a first aspect, the present invention provides a method for analyzing the pollution components of insulators based on deep learning, comprising:

[0009] In response to a received image containing insulators, extract the image of the insulator object from the image;

[0010] Perform grayscale processing on the image of the insulator object to obtain a grayscale image of the insulator object;

[0011] The grayscale image of the insulator object is decomposed and grouped to obtain grayscale image groups of the insulator object. Each grayscale image group of the insulator object includes a type of grayscale image of the sub-insulator object. The area of ​​the grayscale image of the sub-insulator object is greater than or equal to the area threshold.

[0012] The sub-insulator spectral image is obtained by cropping the sub-insulator spectral image using the grayscale image of the sub-insulator object;

[0013] An analysis reference line is established on the spectral image of the sub-insulator, and an analysis curve is constructed using the pixels on the analysis reference line;

[0014] The power spectrum was obtained by processing the analysis curve using autocorrelation calculation and fast Fourier transform.

[0015] Based on the power spectrum, a pollution component identification model obtained from deep learning is selected to determine the pollution components included in the grayscale image of the sub-insulator object.

[0016] In one possible implementation of the first aspect, decomposing the grayscale image of the insulator object includes:

[0017] Content is extracted from the grayscale image of the insulator object using grayscale ranges, and a trend line is established on the grayscale image of the insulator object based on the content extraction results.

[0018] Analyze the convergence area of ​​statistical trend lines and determine the center of the convergence area;

[0019] Establish a direction line starting from the center position;

[0020] Calculate the rate of change of the numerical value along the direction line and determine the boundary points based on the rate of change of the numerical value;

[0021] By sequentially connecting the boundary points, a grayscale image of the sub-insulator object is obtained.

[0022] In one possible implementation of the first aspect, extracting content from the grayscale image of the insulator object using grayscale ranges and establishing a trend line on the grayscale image of the insulator object based on the content extraction results includes:

[0023] The content extraction results are classified into different levels, with each level of content extraction results corresponding to a grayscale range.

[0024] The content extraction results are divided into content extraction result regions, and each content extraction result region has the same area.

[0025] By connecting multiple content extraction result areas according to the trend, a trend line is obtained;

[0026] The aggregation area of ​​statistical trend lines is analyzed, and the content extraction result area that participates in the same trend line is adjusted to minimize the number of connected aggregation areas.

[0027] Among them, the content extraction result areas that participate in the same trend line come from different levels, and any two content extraction result areas in the same level do not participate in the same trend line.

[0028] In one possible implementation of the first aspect, calculating the rate of change of the numerical value along the direction line and determining the boundary points based on the rate of change of the numerical value includes:

[0029] Reference points are set at intervals along the direction line, with equal distances between adjacent reference points;

[0030] Calculate the average value of the reference point and the pixels surrounding the reference point, and use the average value as the value of the reference point;

[0031] Calculate the quadratic difference series of statistical values ​​and plot the quadratic difference series curve;

[0032] Identify similar regions on all obtained quadratic difference series curves;

[0033] Use the starting point of a similar region as the boundary point.

[0034] In one possible implementation of the first aspect, selecting a deep learning-based contaminant identification model based on the power spectrum includes:

[0035] The model features of the dirt component identification model based on deep learning are obtained. The model features include the number of frequencies, frequency intensity, frequency proportion and frequency interval.

[0036] The model features are used to match the power spectrum to determine the pollution component identification model;

[0037] Among them, the frequency intensity is less than or equal to the intensity of the corresponding frequency in the power spectrum.

[0038] In one possible implementation of the first aspect, after establishing the analysis curve using pixels on the analysis reference line, the method further includes removing interference from the analysis curve. Removing interference from the analysis curve includes:

[0039] The analysis curve is processed using wavelet transform to obtain multiple sub-analysis curves, each with a start point and an end point.

[0040] The transformation length is used to filter the sub-analysis curve once to obtain the grayscale image of the sub-insulator object after processing. The sharpness value of the grayscale image of the sub-insulator object after processing is calculated.

[0041] When the clarity value of the grayscale image of the sub-insulator object meets the requirements in one processing, stop using the transform length to perform a single screening of the sub-analysis curve.

[0042] In one possible implementation of the first aspect, there is at least one endpoint of a sub-analysis curve that has not been deleted between the start point and the end point of the deleted sub-analysis curve.

[0043] Delete other sub-analysis curves with the same frequency as the deleted sub-analysis curve;

[0044] When the content loss of the grayscale image of the sub-insulator object exceeds the allowable range, adjust the deleted sub-analysis curves, including:

[0045] Set a deletion length and retain the other sub-analysis curves whose length is greater than the deletion length;

[0046] Recalculate the content loss of the grayscale image of the sub-insulator object, and stop modifying the deletion length when the content loss is within the allowable range.

[0047] Secondly, the present invention provides a deep learning-based insulator pollution component analysis device, comprising:

[0048] The content extraction unit is used to extract the image of the insulator object from the received image containing the insulator.

[0049] The grayscale processing unit is used to perform grayscale processing on the image of the insulator object to obtain a grayscale image of the insulator object.

[0050] The first processing unit is used to decompose and group the grayscale image of the insulator object to obtain grayscale image groups of the insulator object. Each grayscale image group of the insulator object includes a type of grayscale image of the sub-insulator object, and the area of ​​the grayscale image of the sub-insulator object is greater than or equal to the area threshold.

[0051] The second processing unit is used to crop the insulator spectral image using the grayscale image of the sub-insulator object to obtain the sub-insulator spectral image;

[0052] The third processing unit is used to establish an analysis reference line on the spectral image of the sub-insulator and to establish an analysis curve using the pixels on the analysis reference line.

[0053] The spectrum processing unit is used to process and analyze the curves using autocorrelation calculation and fast Fourier transform to obtain the power spectrum;

[0054] The component determination unit is used to select a pollution component identification model based on deep learning according to the power spectrum and determine the pollution components included in the grayscale image of the sub-insulator object.

[0055] Thirdly, the present invention provides a deep learning-based insulator pollution component analysis system, the system comprising:

[0056] One or more memories for storing instructions; and

[0057] One or more processors are configured to call and execute the instructions from the memory to perform the methods described in the first aspect and any possible implementation thereof.

[0058] Fourthly, the present invention provides a computer-readable storage medium comprising:

[0059] The program, when run by a processor, is executed as described in the first aspect and any possible implementation thereof.

[0060] Fifthly, the present invention provides a computer program product, including program instructions that, when the program instructions are run by a computing device, execute the method described in the first aspect and any possible implementation thereof.

[0061] In a sixth aspect, the present invention provides a chip system including a processor for implementing the functions involved in the foregoing aspects, such as generating, receiving, transmitting, or processing data and / or information involved in the foregoing methods.

[0062] This chip system can consist of chips or include chips and other discrete components.

[0063] In one possible design, the chip system also includes a memory for storing necessary program instructions and data. The processor and the memory can be decoupled and located on different devices, connected via wired or wireless means, or the processor and the memory can be coupled to the same device. Attached Figure Description

[0064] Figure 1 This is a flowchart illustrating the steps of a deep learning-based insulator pollution component analysis method provided by the present invention.

[0065] Figure 2 This is a schematic diagram of decomposing and grouping a grayscale image of an insulator object, provided by the present invention.

[0066] Figure 3 This is a schematic diagram of a power spectrum provided by the present invention.

[0067] Figure 4 This is a schematic diagram of establishing a trend line on a grayscale image of an insulator object, provided by the present invention.

[0068] Figure 5 This is a schematic diagram of a trend line provided by the present invention.

[0069] Figure 6 This is a schematic diagram of a direction line provided by the present invention.

[0070] Figure 7 This is a schematic diagram of similar regions on a quadratic difference sequence curve provided by the present invention. Detailed Implementation

[0071] The technical solutions of the present invention will be further described in detail below with reference to the accompanying drawings.

[0072] This invention discloses a deep learning-based method for analyzing the pollution components of insulators. Please refer to [link / reference]. Figure 1 In some examples, the deep learning-based insulator pollution component analysis method disclosed in this invention includes the following steps:

[0073] S101, in response to the received image containing insulators, extract the image of the insulator object from the image;

[0074] S102, perform grayscale processing on the insulator object image to obtain a grayscale image of the insulator object;

[0075] S103, decompose and group the grayscale image of the insulator object to obtain grayscale image group of the insulator object. Each grayscale image group of the insulator object includes a type of grayscale image of the sub-insulator object. The area of ​​the grayscale image of the sub-insulator object is greater than or equal to the area threshold.

[0076] S104, use the grayscale image of the sub-insulator object to crop the spectral image of the insulator to obtain the spectral image of the sub-insulator;

[0077] S105, Establish an analysis reference line on the spectral image of the sub-insulator and use the pixels on the analysis reference line to establish an analysis curve;

[0078] S106 uses autocorrelation calculation and fast Fourier transform to process the analysis curve and obtain the power spectrum;

[0079] S107, Select the pollution component identification model based on deep learning according to the power spectrum and determine the pollution components included in the grayscale image of the sub-insulator object.

[0080] Overall, the deep learning-based insulator pollution composition analysis method provided by this invention mainly solves two problems: pollution identification and pollution composition analysis. Specifically, it identifies pollution using ordinary images and then analyzes the composition of pollution using spectral images.

[0081] In step S101, the insulator object image is first extracted from the received image containing the insulator. The insulator object image can be extracted using image feature extraction processing methods, since the type, shape and color of the insulator can be fixed, and the extraction can be performed with the help of a neural network that has been specifically trained.

[0082] In step S102, the insulator object image is processed to obtain a grayscale image of the insulator object. Then, in step S103, the grayscale image of the insulator object is decomposed and grouped. Please refer to [link to relevant documentation]. Figure 2 This process yields a grayscale image group of the insulator object. Decomposition involves obtaining the grayscale image portion of the insulator object containing contaminants, while grouping involves placing the same type of contaminants into a single group. Figure 2 The two solid rectangles at the middle arrow point form one group, and the two dashed rectangles form another group.

[0083] In other words, each grayscale image group of an insulator object includes a grayscale image of a sub-insulator object. Of course, the area of ​​the grayscale image of the sub-insulator object must be greater than or equal to the area threshold. The purpose is to avoid getting stuck in a local optimum or a loop with no solution during the processing.

[0084] In step S104, the grayscale image of the sub-insulator object is used to crop the spectral image of the insulator to obtain the spectral image of the sub-insulator. The grayscale image of the sub-insulator object corresponds to the location on the insulator where there is dirt. The spectral image of the sub-insulator obtained after cropping the spectral image of the insulator using the grayscale image of the sub-insulator object also corresponds to the location on the insulator where there is dirt.

[0085] Using this processing method, only the dirty parts of the insulator spectral image can be processed, thus avoiding the need to process the entire insulator spectral image.

[0086] In step S105, an analysis reference line is established on the spectral image of the sub-insulator, and an analysis curve is constructed using the pixels on the analysis reference line. Then, in step S106, the analysis curve is processed using autocorrelation calculation and fast Fourier transform to obtain the power spectrum. Figure 3 As shown in the figure, in step S107, the pollution component identification model based on deep learning is selected according to the power spectrum, and the pollution components included in the grayscale image of the sub-insulator object are determined.

[0087] Specifically, in the aforementioned steps, the areas on the insulator where contamination exists were first identified and the spectral images of the sub-insulators corresponding to the contamination areas were obtained. The spectral images of the sub-insulators included information on the composition of the contamination. The content of steps S105 to S107 is to analyze the composition information of the contamination.

[0088] When performing component analysis, an analysis reference line is first established on the spectral image of the sub-insulator, and the analysis curve is established using the pixels on the analysis reference line. The spectral image of the sub-insulator can be one or multiple images.

[0089] It should be understood that the spectral image is obtained in sequence according to the frequency windows, with each frequency window corresponding to a frequency range. Based on this, the spectral image of a sub-insulator can be regarded as being caused by multiple discrete spectral images, or it can be formed by merging these discrete spectral images.

[0090] After establishing an analysis reference line on the spectral image of the sub-insulator and using the pixels on the analysis reference line to establish the analysis curve, the analysis curve is processed using autocorrelation calculation and fast Fourier transform to obtain the power spectrum. After obtaining the power spectrum through autocorrelation calculation and fast Fourier transform, it is necessary to determine the pollution components by determining the power spectrum.

[0091] The specific method is to first determine the possible components based on the power spectrum. This method is a fuzzy judgment method. For example, suppose that the power spectrum corresponding to a certain pollution includes three components A, B, and C. If the power spectrum at this time also contains three components A, B, and C, then the presence of this type of pollution on the insulator will be suspected.

[0092] For the three components A, B, and C, it is generally the main component of a certain filth rather than all of the components. Based on this approach, we can first select a filth component identification model obtained from deep learning, resulting in multiple filth component identification models.

[0093] The power spectrum is then fed to the dirt component identification model, which determines whether the corresponding dirt component exists.

[0094] In the above method, the dirt component identification model is trained separately and only identifies one type of dirt component. Here, an identification model composed of multiple dirt component identification models that can identify multiple dirt components simultaneously is also a variant of the dirt component identification model mentioned in this invention and should be included within the scope of this invention.

[0095] In some possible implementations, the method of establishing an analysis reference line on the sub-insulator spectral image is to establish a straight line as long as possible on the sub-insulator spectral image. The number of analysis reference lines can be one or multiple. When the number of analysis reference lines is multiple, after obtaining the results, the result with the largest proportion is used as the final result.

[0096] In some cases, the specific method for decomposing the grayscale image of the insulator object is as follows:

[0097] S201, Use grayscale range to extract content from the grayscale image of the insulator object and establish a trend line on the grayscale image of the insulator object based on the content extraction results;

[0098] S202, statistically analyze the convergence area of ​​the trend lines and determine the center of the convergence area;

[0099] S203, establish a direction line starting from the center position;

[0100] S204, calculate the rate of change of the numerical value along the direction line and determine the boundary point based on the rate of change of the numerical value;

[0101] S205, sequentially connect the obtained boundary points to obtain the grayscale image of the sub-insulator object.

[0102] In steps S201 to S205, content is first extracted from the grayscale image of the insulator object using grayscale ranges, and a trend line is established on the grayscale image of the insulator object based on the content extraction results. Figure 4 As shown, the grayscale range is a numerical range. By using the grayscale range to process the grayscale image of the insulator object, a content extraction result can be obtained for each grayscale range.

[0103] The specific method for extracting content from the grayscale image of an insulator object using grayscale ranges and establishing a trend line on the grayscale image of the insulator object based on the content extraction results is as follows:

[0104] S301, classify the content extraction results into levels, with each level of content extraction results corresponding to a grayscale range;

[0105] S302, Divide the content extraction results into content extraction result regions, with each content extraction result region having the same area;

[0106] S303: Connect multiple content extraction result areas according to the trend to obtain a trend line;

[0107] S304, statistically analyze the convergence area of ​​trend lines and adjust the content extraction result area that participates in the same trend line to minimize the number of connected convergence areas.

[0108] Among them, the content extraction result areas that participate in the same trend line come from different levels, and any two content extraction result areas in the same level do not participate in the same trend line.

[0109] After obtaining the content extraction results, the results are first classified, with each level corresponding to a grayscale range. A suitable classification method is to first sort the grayscale ranges, with higher numbers indicating higher levels and thus higher classification grades.

[0110] Next, the content extraction results are divided into regions, each with the same area. This division involves creating independent regions from the extracted content. After region division, multiple content extraction result regions are connected according to a trend to obtain a trend line. Figure 5 (The dashed line in the middle).

[0111] This requires that the content extraction result areas participating in the same trend line come from different levels, and that any two content extraction result areas in the same level do not participate in the same trend line.

[0112] Following the trend refers to either the color becoming darker or lighter; either approach is acceptable, and there are no restrictions here.

[0113] Next, the convergence area of ​​the trend lines is statistically analyzed and the content extraction result area participating in the same trend line is adjusted to minimize the number of connected convergence areas. The convergence area of ​​the trend lines obtained here is generally the central or near-central area of ​​the filth.

[0114] Using this method, the influence of uncertain shapes can be ignored, and the center or near-center area of ​​the filth can be obtained more accurately.

[0115] In some examples, the specific methods for calculating the rate of change of values ​​along the direction line and determining the boundary points based on the rate of change of values ​​are as follows:

[0116] S401, reference points are set at intervals on the direction line, and the distance between adjacent reference points is equal;

[0117] S402, Calculate the average value of the reference point and the pixels around the reference point, and use the average value as the value of the reference point;

[0118] S403, Calculate the quadratic difference series of statistical values ​​and plot the quadratic difference series curve;

[0119] S404, determine the similar regions on all obtained quadratic difference series curves;

[0120] S405 uses the starting point of a similar region as the boundary point.

[0121] In steps S401 to S405, the direction line will be first... Figure 6Reference points are set at intervals on the dashed lines in the diagram. Then, the average value of the reference point and the pixels around the reference point is calculated and used as the value of the reference point. The purpose of calculating the average value of the reference point and the pixels around the reference point is to ensure the stability of the pixel average value feedback. This is because the value of a single pixel has a certain error, and when selecting a reference point, there may be a selection deviation due to an unsuitable position. The average value calculation method can avoid the above problems.

[0122] After obtaining the baseline values, the quadratic difference series of the values ​​is statistically analyzed, and the quadratic difference series curve is plotted using the quadratic difference series. Figure 7 (as shown in the figure), then determine the similar regions on all the obtained quadratic difference series curves, and finally take the starting point of the similar region as the boundary point.

[0123] Here, for similar regions on the curve of a quadratic difference series ( Figure 7 (As shown), this refers to a region where the slope suddenly changes because an edge appears. In this invention, the starting points of similar regions are used as boundary points, and these boundary points are then connected sequentially to obtain the boundary.

[0124] In some examples, after constructing the analysis curve using pixels on the analysis reference line, the process also includes removing interference from the analysis curve. The specific method for removing interference from the analysis curve is as follows:

[0125] S501 uses wavelet transform to process the analysis curve and obtains multiple sub-analysis curves, each with a start point and an end point.

[0126] S502, use the transformation length to perform a single screening of the sub-analysis curve to obtain a grayscale image of the sub-insulator object after processing, and calculate the sharpness value of the grayscale image of the sub-insulator object after processing.

[0127] S503: When the clarity value of the grayscale image of the sub-insulator object meets the requirements in one processing, stop using the transformation length to perform one screening of the sub-analysis curve.

[0128] In the above steps, the analysis curves are first processed using wavelet transform, resulting in multiple sub-analysis curves. Then, the sub-analysis curves are filtered using the transform length. The transform length refers to the deletion of sub-analysis curves that are less than a certain set length each time they are filtered. Each time a sub-analysis curve is deleted, the sharpness value of the grayscale image of the sub-insulator object is calculated.

[0129] When the sharpness value of the grayscale image of the insulator object meets the requirements in one processing, stop using the transformation length to filter the sub-analysis curve. If the sharpness value does not meet the requirements, adjust the setting length for the next filtering.

[0130] Methods for judging sharpness include Brenner gradient function, Tenengrad gradient function, Laplacian gradient function, SMD (gray-level variance) function, and SMD2 (gray-level variance product) function.

[0131] The following requirements also apply during the processing:

[0132] There must be at least one endpoint of an undeleted sub-analysis curve between the start and end points of the deleted sub-analysis curve. This approach avoids deleting isolated points and aims to preserve content within permissible limits.

[0133] This is because, according to the present invention, interference should be mixed with non-interference data. When a data exists alone, it may be interference data or non-interference data. If too much data is deleted, it will affect the judgment of the results in the later stage. Therefore, the present invention requires that there be at least one endpoint of the sub-analysis curve that has not been deleted between the start point and the end point of the deleted sub-analysis curve.

[0134] In addition, other sub-analysis curves with the same frequency as the deleted sub-analysis curve need to be deleted. This is a simplified processing method, which treats all sub-analysis curves with a certain frequency as interference data.

[0135] However, when the content loss of the grayscale image of the sub-insulator object exceeds the allowable range, the other sub-analysis curves to be deleted are adjusted, including:

[0136] Set a deletion length and retain the other sub-analysis curves whose length is greater than the deletion length;

[0137] Recalculate the content loss of the grayscale image of the sub-insulator object, and stop modifying the deletion length when the content loss is within the allowable range.

[0138] In some examples, the specific methods for selecting the deep learning-based fouling component identification model based on the power spectrum are as follows:

[0139] The model features of the dirt component identification model based on deep learning are obtained. The model features include the number of frequencies, frequency intensity, frequency proportion and frequency interval.

[0140] The model features are used to match the power spectrum to determine the pollution component identification model;

[0141] Among them, the frequency intensity is less than or equal to the intensity of the corresponding frequency in the power spectrum.

[0142] The above method first determines the model features, and then uses the model features to match the power spectrum to determine the pollution component identification model. The model features include four feature quantities: frequency quantity, frequency intensity, frequency proportion, and frequency interval.

[0143] When matching model features in the power spectrum, if one or more of the frequency quantity, frequency intensity, frequency proportion, and frequency interval match, the pollution component identification model will be retained. After this step is completed, multiple pollution component identification models can be selected from all pollution component identification models, and each pollution component identification model can be used to identify a pollution component.

[0144] This invention also provides a deep learning-based insulator pollution component analysis device, comprising:

[0145] The content extraction unit is used to extract the image of the insulator object from the received image containing the insulator.

[0146] The grayscale processing unit is used to perform grayscale processing on the image of the insulator object to obtain a grayscale image of the insulator object.

[0147] The first processing unit is used to decompose and group the grayscale image of the insulator object to obtain grayscale image groups of the insulator object. Each grayscale image group of the insulator object includes a type of grayscale image of the sub-insulator object, and the area of ​​the grayscale image of the sub-insulator object is greater than or equal to the area threshold.

[0148] The second processing unit is used to crop the insulator spectral image using the grayscale image of the sub-insulator object to obtain the sub-insulator spectral image;

[0149] The third processing unit is used to establish an analysis reference line on the spectral image of the sub-insulator and to establish an analysis curve using the pixels on the analysis reference line.

[0150] The spectrum processing unit is used to process and analyze the curves using autocorrelation calculation and fast Fourier transform to obtain the power spectrum;

[0151] The component determination unit is used to select a pollution component identification model based on deep learning according to the power spectrum and determine the pollution components included in the grayscale image of the sub-insulator object.

[0152] In one possible implementation of the first aspect, decomposing the grayscale image of the insulator object includes:

[0153] Content is extracted from the grayscale image of the insulator object using grayscale ranges, and a trend line is established on the grayscale image of the insulator object based on the content extraction results.

[0154] Analyze the convergence area of ​​statistical trend lines and determine the center of the convergence area;

[0155] Establish a direction line starting from the center position;

[0156] Calculate the rate of change of the numerical value along the direction line and determine the boundary points based on the rate of change of the numerical value;

[0157] By sequentially connecting the boundary points, a grayscale image of the sub-insulator object is obtained.

[0158] In one possible implementation of the first aspect, extracting content from the grayscale image of the insulator object using grayscale ranges and establishing a trend line on the grayscale image of the insulator object based on the content extraction results includes:

[0159] The content extraction results are classified into different levels, with each level of content extraction results corresponding to a grayscale range.

[0160] The content extraction results are divided into content extraction result regions, and each content extraction result region has the same area.

[0161] By connecting multiple content extraction result areas according to the trend, a trend line is obtained;

[0162] The aggregation area of ​​statistical trend lines is analyzed, and the content extraction result area that participates in the same trend line is adjusted to minimize the number of connected aggregation areas.

[0163] Among them, the content extraction result areas that participate in the same trend line come from different levels, and any two content extraction result areas in the same level do not participate in the same trend line.

[0164] In one possible implementation of the first aspect, calculating the rate of change of the numerical value along the direction line and determining the boundary points based on the rate of change of the numerical value includes:

[0165] Reference points are set at intervals along the direction line, with equal distances between adjacent reference points;

[0166] Calculate the average value of the reference point and the pixels surrounding the reference point, and use the average value as the value of the reference point;

[0167] Calculate the quadratic difference series of statistical values ​​and plot the quadratic difference series curve;

[0168] Identify similar regions on all obtained quadratic difference series curves;

[0169] Use the starting point of a similar region as the boundary point.

[0170] In one possible implementation of the first aspect, selecting a deep learning-based contaminant identification model based on the power spectrum includes:

[0171] The model features of the dirt component identification model based on deep learning are obtained. The model features include the number of frequencies, frequency intensity, frequency proportion and frequency interval.

[0172] The model features are used to match the power spectrum to determine the pollution component identification model;

[0173] Among them, the frequency intensity is less than or equal to the intensity of the corresponding frequency in the power spectrum.

[0174] In one possible implementation of the first aspect, after establishing the analysis curve using pixels on the analysis reference line, the method further includes removing interference from the analysis curve. Removing interference from the analysis curve includes:

[0175] The analysis curve is processed using wavelet transform to obtain multiple sub-analysis curves, each with a start point and an end point.

[0176] The transformation length is used to filter the sub-analysis curve once to obtain the grayscale image of the sub-insulator object after processing. The sharpness value of the grayscale image of the sub-insulator object after processing is calculated.

[0177] When the clarity value of the grayscale image of the sub-insulator object meets the requirements in one processing, stop using the transform length to perform a single screening of the sub-analysis curve.

[0178] In one possible implementation of the first aspect, there is at least one endpoint of a sub-analysis curve that has not been deleted between the start point and the end point of the deleted sub-analysis curve.

[0179] Delete other sub-analysis curves with the same frequency as the deleted sub-analysis curve;

[0180] When the content loss of the grayscale image of the sub-insulator object exceeds the allowable range, adjust the deleted sub-analysis curves, including:

[0181] Set a deletion length and retain the other sub-analysis curves whose length is greater than the deletion length;

[0182] Recalculate the content loss of the grayscale image of the sub-insulator object, and stop modifying the deletion length when the content loss is within the allowable range.

[0183] In one example, the unit in any of the above devices may be one or more integrated circuits configured to implement the above methods, such as one or more application-specific integrated circuits (ASICs), or one or more digital signal processors (DSPs), or one or more field-programmable gate arrays (FPGAs), or a combination of at least two of these integrated circuit forms.

[0184] For example, when the units in the device can be implemented through a processing element scheduler, the processing element can be a general-purpose processor, such as a central processing unit (CPU) or other processor capable of calling programs. Alternatively, these units can be integrated together to form a system-on-a-chip (SOC).

[0185] In this invention, various objects such as messages / information / devices / network elements / systems / devices / actions / operations / processes / concepts may be named. It is understood that these specific names do not constitute a limitation on the relevant objects. The names may be changed depending on the scenario, context, or usage habits. The understanding of the technical meaning of the technical terms in this invention should be mainly determined from their functions and technical effects embodied / performed in the technical solution.

[0186] Those skilled in the art will 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.

[0187] In the embodiments provided by this invention, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and 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 through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0188] 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.

[0189] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0190] It should also be understood that in the various embodiments of the present invention, "first," "second," etc., are merely used to indicate that multiple objects are different. For example, a first time window and a second time window are only used to indicate different time windows, and should not have any effect on the time window itself. The aforementioned "first," "second," etc., should not impose any limitations on the embodiments of the present invention.

[0191] It should also be understood that, in the various embodiments of the present invention, unless otherwise specified or in case of logical conflict, the terminology and / or descriptions between different embodiments are consistent and can be referenced by each other, and the technical features in different embodiments can be combined to form new embodiments according to their inherent logical relationships.

[0192] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a computer-readable 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 described in the various embodiments of the present invention. The aforementioned computer-readable storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0193] This invention also provides a deep learning-based insulator pollution component analysis system, the system comprising:

[0194] One or more memories for storing instructions; and

[0195] One or more processors are configured to retrieve and execute the instructions from the memory, performing the methods described above.

[0196] The present invention also provides a computer program product including instructions that, when executed, cause the terminal device and the network device to perform operations corresponding to the methods described above.

[0197] The present invention also provides a chip system including a processor for implementing the functions involved in the above description, such as generating, receiving, transmitting, or processing the data and / or information involved in the above methods.

[0198] This chip system can consist of chips or include chips and other discrete components.

[0199] The processor mentioned above can be a CPU, a microprocessor, an ASIC, or one or more integrated circuits that execute a program to control the method of transmitting the feedback information described above.

[0200] In one possible design, the chip system also includes a memory for storing necessary program instructions and data. The processor and the memory can be decoupled and located on different devices, connected via wired or wireless means to support the chip system in implementing the various functions described in the above embodiments. Alternatively, the processor and the memory can also be coupled to the same device.

[0201] Optionally, the computer instructions are stored in memory.

[0202] Optionally, the memory can be a storage unit within the chip, such as a register or cache. Alternatively, the memory can be a storage unit located outside the chip within the terminal, such as a ROM or other types of static storage devices that can store static information and instructions, such as RAM.

[0203] It is understood that the memory in this invention can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory.

[0204] Non-volatile memory can be ROM, programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory.

[0205] Volatile memory can be RAM, which is used as an external cache. There are many different types of RAM, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct memory bus random access memory.

[0206] The embodiments described herein are preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape, and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A deep learning-based insulator contamination component analysis method, characterized by, The method comprises the following steps: In response to the received image containing the insulator, the insulator object image in the image is extracted; The insulator object image is subjected to grayscale processing to obtain an insulator object grayscale image; The insulator object grayscale image is decomposed and grouped to obtain an insulator object grayscale image group, each insulator object grayscale image group comprising a type of sub-insulator object grayscale image, and the area of the sub-insulator object grayscale image being greater than or equal to an area threshold value; The sub-insulator object grayscale image is used to cut the insulator spectral image to obtain a sub-insulator spectral image; An analysis reference line is established on the sub-insulator spectral image, and an analysis curve is established using the pixel points on the analysis reference line; The analysis curve is processed using autocorrelation calculation and fast Fourier transform to obtain a power spectrum; Based on the power spectrum, a pollution component recognition model based on deep learning is selected, and the pollution component included in the sub-insulator object grayscale image is determined. 2.The deep learning-based insulator contamination component analysis method of claim 1, wherein, The decomposition of the insulator object grayscale image comprises: Content extraction is performed on the insulator object grayscale image using a grayscale interval, and a change trend line is established on the insulator object grayscale image according to the content extraction result; The collection area of the change trend line is counted, and the center position of the collection area is determined; A direction line is established with the center position as the starting point; The numerical change speed on the direction line is calculated, and the boundary points are determined according to the numerical change speed; The obtained boundary points are sequentially connected to obtain the sub-insulator object grayscale image. 3.The deep learning-based insulator contamination component analysis method of claim 2, wherein, The content extraction is performed on the insulator object grayscale image using a grayscale interval, and a change trend line is established on the insulator object grayscale image according to the content extraction result, which comprises: The content extraction result is classified, and each level of content extraction result corresponds to a grayscale interval; The content extraction result is divided to obtain content extraction result areas, and the areas of each content extraction result area are the same; The change trend line is obtained by connecting multiple content extraction result areas according to the trend; The collection area of the change trend line is counted, and the content extraction result areas participating in the same change trend line are adjusted to minimize the number of connected collection areas; The content extraction result areas participating in the same change trend line come from different levels, and any two content extraction result areas in the same level do not participate in the same change trend line. 4.The deep learning-based insulator contamination component analysis method of claim 2, wherein, The numerical change speed on the direction line is calculated, and the boundary points are determined according to the numerical change speed, which comprises: The reference points are set on the direction line at intervals, and the distance between adjacent reference points is equal; The mean value of the reference points and the pixel points around the reference points is calculated, and the mean value is taken as the numerical value of the reference point; The second difference value sequence is counted, and the second difference value sequence curve is drawn using the second difference value sequence; Similar regions on all obtained second difference value sequence curves are determined; The starting point of the similar region is used as the boundary point. 5.The deep learning-based insulator contamination component analysis method of claim 1, wherein The pollution component recognition model based on deep learning is selected according to the power spectrum, which comprises: The model characteristics of the pollution component recognition model based on deep learning are obtained, and the model characteristics include the frequency number, the frequency intensity, the frequency proportion, and the frequency interval; The model characteristics are matched in the power spectrum to determine the pollution component recognition model; The frequency intensity is less than or equal to the intensity of the corresponding frequency in the power spectrum. 6.The deep learning-based insulator contamination component analysis method according to claim 1 or 5, characterized in that, After the analysis curve is established using the pixel points on the analysis reference line, the method further includes removing interference on the analysis curve, and removing the interference on the analysis curve includes: The analysis curve is processed using a wavelet transform mode to obtain a plurality of sub-analysis curves, and each of the sub-analysis curves has a start position point and an end position point; The sub-analysis curves are screened using a transform length to obtain a first processed sub-insulator object gray image, and a definition value of the first processed sub-insulator object gray image is calculated; When the definition value of the first processed sub-insulator object gray image meets a requirement, the screening of the sub-analysis curves using the transform length is stopped. 7.The deep learning-based insulator contamination component analysis method of claim 6, wherein, The start position point and the end position point of the deleted sub-analysis curve are connected by at least one end point of a non-deleted sub-analysis curve; Other sub-analysis curves having the same frequency as the deleted sub-analysis curve are deleted; When the content loss of the sub-insulator object gray image exceeds an allowable range, the deleted other sub-analysis curves are adjusted, and the adjusting includes: A deletion length is set, and the other sub-analysis curves having a length greater than the deletion length are retained; The content loss of the sub-insulator object gray image is recalculated, and the deletion length is modified until the content loss is within the allowable range.

8. An insulator contamination component analysis device based on deep learning, characterized by, The system includes: a content extraction unit configured to extract an insulator object image from an image containing an insulator in response to the image being received; a gray processing unit configured to perform gray processing on the insulator object image to obtain an insulator object gray image; a first processing unit configured to decompose and group the insulator object gray image to obtain a plurality of insulator object gray image groups, each of the insulator object gray image groups including a plurality of sub-insulator object gray images, and each of the sub-insulator object gray images having an area greater than or equal to an area threshold; a second processing unit configured to cut an insulator spectral image using the sub-insulator object gray images to obtain a plurality of sub-insulator spectral images; a third processing unit configured to establish an analysis reference line on each of the sub-insulator spectral images and establish an analysis curve using pixel points on the analysis reference line; a spectrum processing unit configured to process the analysis curve using an autocorrelation calculation and a fast Fourier transform mode to obtain a power spectrum; a component determination unit configured to select a pollution component recognition model based on deep learning according to the power spectrum and determine a pollution component included in the sub-insulator object gray image.

9. An insulator contamination component analysis system based on deep learning, characterized by, The system includes: one or more memories configured to store instructions; and one or more processors configured to call and run the instructions from the memories to perform the method of any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes: a program configured to perform the method of any one of claims 1 to 7 when the program is run by a processor.

Images