CN115901796A - Mining area tower insulator contamination state identification method based on multi-image feature fusion - Google Patents
Mining area tower insulator contamination state identification method based on multi-image feature fusion Download PDFInfo
- Publication number
- CN115901796A CN115901796A CN202211505793.3A CN202211505793A CN115901796A CN 115901796 A CN115901796 A CN 115901796A CN 202211505793 A CN202211505793 A CN 202211505793A CN 115901796 A CN115901796 A CN 115901796A
- Authority
- CN
- China
- Prior art keywords
- image
- value
- mining area
- characteristic
- images
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Image Analysis (AREA)
Abstract
A mining area pole tower insulator contamination state identification method based on multi-image feature fusion comprises the following steps of 1: acquiring visible light images, infrared images and ultraviolet images of the insulator of the tower in the mining area from multiple angles; and 2, step: respectively carrying out image segmentation on the infrared image and the visible light image by adopting an improved seed region growing method; and step 3: selecting original characteristic parameters of the visible light image, the infrared image and the ultraviolet image respectively by utilizing a Fisher criterion; and 4, step 4: trimming and correlating the characteristic parameters by using an improved ID3 algorithm, and obtaining optimized information gain by using an equalization coefficient; and 5: and setting a threshold value based on the optimization information gain, and classifying the pollution grades of the insulators of the towers in the mining area. The advantages are that: the method combines three images, namely visible light images, infrared images and ultraviolet images, obtains relatively less quantity of each image, has high information identification certainty, provides guarantee for the identification accuracy of the insulator contamination state of the tower in the mining area, and has high accuracy.
Description
Technical Field
The invention belongs to the field of electrical equipment fault detection and diagnosis, and particularly relates to a method for identifying a filthy state of a pole tower insulator in a mining area based on multi-image feature fusion.
Background
The insulator is one of important insulating parts of a power grid, and the insulation defect of the insulator can cause great loss and is directly related to energy safety. In the traditional transmission line insulator operation and detection process, the problems of false detection, missed detection, low defect detection rate, high consumption of manpower and material resources for defect screening and the like are easy to occur in the detection and evaluation process depending on manual experience. In addition, the decrease in the insulation performance of the insulator is accompanied by the distortion of the surface local electric field and the increase in current leakage, which may cause surface discharge, heat generation, and the like.
Particularly, mine pole towers are mainly concentrated in an area within 100 meters near an operation platform, intensive factory and mining industries are mainly used in the periphery, and when the mine pole towers meet strong wind and sand storm weather, the content of sand particles and coal dust in the air is greatly increased and the mine pole towers move at high speed, so that on one hand, the accumulated dirt of the insulator is increased, and on the other hand, sand particles can impact the surface of the insulator from all directions to damage the external insulation performance of the insulator. If in rainy days or snowy days, the occurrence of mine area tower insulator discharge is easily caused, the stable operation of a mine area electric power system is seriously influenced, and when lightning current flows through the dirt-accumulating insulator, inferior value insulators are easily formed, and once the insulators exist on a line, the weak point of insulation is formed on the line, so that the lightning stroke flashover rate of the line is increased, and particularly when lightning stroke on the tower or a lightning conductor occurs, lightning stroke 'counterattack' is easily caused, so that the lightning stroke resistance level of the whole line is reduced.
The infrared imaging sensing technology is widely used for detecting various power transmission and transformation equipment of a power grid. Through the long-distance infrared thermal imaging detecting instrument that uses, can visual detection power transmission and transformation equipment surface unusual situation that generates heat fast, provide technical guarantee for solving the nature trouble that generates heat fast. However, the number of infrared images shot by the on-site infrared imager is huge, and the on-site infrared imager needs manual grading diagnosis; meanwhile, in the process of live-line infrared detection of insulators of towers in a mining area, due to the fact that the insulators are polluted, various interference factors such as various pseudo-color types, complex background interference, different shooting angles, infrared images, superposed graphic characters and the like exist, and certain challenges and difficulties are caused to the accuracy of intelligent identification of abnormal heating fault points of the infrared images.
Disclosure of Invention
In order to solve the problems, the invention provides a mining area pole and tower insulator contamination state identification method based on multi-image feature fusion.
The technical scheme adopted by the invention is as follows:
a mining area pole tower insulator contamination state identification method based on multi-image feature fusion is characterized by comprising the following steps:
step 1: respectively utilizing a visible light camera, an infrared thermal imager and an ultraviolet imager to obtain visible light images, infrared images and ultraviolet images of the insulator of the tower in the mining area from multiple angles;
step 2: respectively carrying out image segmentation on the infrared image and the visible light image by adopting an improved seed region growing method;
and step 3: selecting original characteristic parameters of the visible light image, the infrared image and the ultraviolet image respectively by utilizing a Fisher criterion;
and 4, step 4: trimming and correlating the characteristic parameters by using an improved ID3 algorithm, and obtaining optimized information gain by using an equalization coefficient;
and 5: and (4) based on the optimized information gain obtained by the improved ID3 algorithm in the step (4), setting a threshold value, and classifying the insulator pollution levels of the towers in the mining area.
Preferably, in step 2, when the infrared image and the visible light image are respectively subjected to image segmentation by using an improved seed region growing method, the following steps are specifically performed:
1) The YCbCr color space can be obtained by carrying out image graying processing and converting the RGB color space, and the conversion formula is as follows:
wherein, R, G, B respectively refer to the colors red, green, blue of RGB color space, Y, cb, cr respectively refer to the brightness component, blue chroma component, red chroma component of color space;
2) Extracting the color characteristic value and the texture characteristic value of the pixel point, and utilizing the obtained characteristic valuesSeed selection is carried out, and the seed point is marked as (x) 0 ,y 0 ) (ii) a Selecting pixel points with the same gray value and points with the same adjacent gray value in the target image as initial areas, and setting the edges of the initial areas to contain n points: (x) 0 ,y 0 )1 、(x 0 ,y 0 )2 、 …(x 0 ,y 0 ) n As new seeds;
3) With seed points (x) 0 ,y 0 ) As a starting point, judging whether the gray difference Ds between the pixel point and the pixel point in the n multiplied by n neighborhood meets the following growth rule or not: ds is a group of<δ s, wherein δ s is a set threshold;
4) If so, carrying out region combination, regarding the point (x, y) as a new seed point, repeating the step 3) on the new seed point (x, y) until the judgment condition is not satisfied, stopping growth, and obtaining a filthy growth region Z;
5) For seed point (x) 0 ,y 0 )1 、(x 0 ,y 0 )2 、…(x 0 ,y 0 ) n, repeating the steps (3) and (4) to obtain a growth area Z 1 、Z 2 、 …Zn；
6) Merging all the dirty growth areas to obtain a target Mn:
Mn= Z∪Z 1 ∪Z 2 ∪ … ∪Zn；
7) Traversing the seed pixel points, and finishing region combination to obtain a region Mn year which is a segmented image;
8) According to the step 2) to the step 7), dividing the insulator of the tower in the mining area into M areas M 1 、M 2 、…Mm；
9) Converting the YCbCr color space of each region into the RGB color space respectively, wherein the formula is as follows:
more preferably, the visible light threshold is δ s =15, and the infrared image threshold is δ s =3.
As a further preferred, the step 3 of selecting the original characteristic parameters of the visible light image, the infrared image and the ultraviolet image comprises the following steps:
step 3.1: for a visible light image, examining luminance components, blue chrominance components and C red chrominance components of red, green, blue and Y color spaces of an RGB color space, and calculating the mean value, the maximum value, the minimum value, the range and the variance of each feature, wherein the total 6 color features are obtained;
for the infrared image, calculating the mean value, the median value, the maximum value, the minimum value, the mode value, the range, the variance, the skewness, the kurtosis, the energy and the entropy of the relative temperature of the surface region of the insulator disc, taking total 11 statistics as temperature characteristics, simultaneously setting a temperature range, and inspecting the ratio of each temperature on the surface of the disc and the coverage rate of a temperature rise region;
for ultraviolet images, 20s are taken as a group, the mean value, the median value, the maximum value, the minimum value, the mode value, the range, the variance, the skewness, the kurtosis, the energy and the entropy of the spot areas of 400 images in each group are calculated, and 11 statistics are used as ultraviolet discharge characteristics in total.
Step 3.2: respectively selecting original characteristics of the visible light image, the infrared image and the ultraviolet image by using a Fisher criterion;
assume the existence of in the data setnEach sample is divided into M categories, and M is recorded 1 ,m 2 ,…,m M Each class respectively containing nkA sample is obtained;
and &>
Is as followskThe intra-class variance and the inter-class variance of the dimensional features on the training set respectively have the following expressions:
in the formula, x k RepresentCharacteristic value of k dimension of single individual, c k Expressed as the mean of the k-dimension features, c is expressed as the mean of all features;
by inter-class variance
And in-class variance ≥>
Ratio J F (k) To judge the degree of distinction of the features, the feature function is
J F (k) The method is characterized in that the method is a Fisher criterion of characteristics, the larger the function value is, the higher the discrimination degree of the characteristics is, and three characteristics which can better represent pollution identification to a visible light image, an infrared image and an ultraviolet image are respectively extracted;
normalizing each group of characteristic quantities to realize data comparability, wherein the normalization formula is
Wherein v represents a feature, d represents a number of feature groups, and j represents a sample number;
represents the jth feature v, <' > in the feature group d>
Represents the minimum value of a feature v in the feature group d>
Represents the minimum value of the feature v in the feature group d;
and taking the evolution of the square sum of the three characteristic normalization values as an evaluation parameter of the image level, namely information Gain (Ai).
As a further preference, in step 4, the optimized features are trimmed and associated by using an improved ID3 algorithm, that is, a data set is constructed, the features of the three images are mutually referenced, the identification characteristics and the efficiency of the three images are considered to be different, and the ID3 algorithm is adopted to realize decision-level fusion of the images according to the feature parameter data; then when the equalization coefficient is used for obtaining the optimized information gain, the method comprises the following steps:
1) Identification efficiency of each feature in various images obtained based on step 3Gain(Ai)；
2) Establishing a characteristic data sample set S of the dirty image, and comparing the characteristic A i Information gain ofGain(A i )Performing correction to obtain corrected information gainGain′(Ai), the expression of which is as follows:
wherein, the first and the second end of the pipe are connected with each other,n 1 the number of classifications of the contamination level is represented,f(n 1 ) In order to modify the parameters of the device,
number of classifications of each contamination classn 1 The larger the correction parameter isf(n 1 ) The smaller; when the classification of the pollution grades is more, the correction parameter is smaller, and the corrected information gain is smaller, so that the influence of multi-value deviation on the information gain is corrected;
3) In order to avoid excessive gain rate correction compensation caused by a certain characteristic, a characteristic deviation threshold T is introduced, the degree of multi-value deviation is measured and controlled, and for a set S, the method comprises the following stepsn 2 The attribute bias threshold T is generally taken as all conditional entropiesE(Ai) Is expressed as follows
Wherein n is 2 Number of representation features, conditional entropyE(Ai) The degree of distinction of each characteristic;
4) The method comprises the steps of balancing the influence of multi-value deflection on information gain and correcting the information gain by using an equalization coefficient to realize optimization of the information gain;
as a further preferred option, the set threshold value is used for classifying the pollution levels of insulators of towers in the mining area as follows:
taking the evolution of the square sum of the numerical values of the equalization coefficient R (Ai) multiplied by 100 percent as the judgment parameter of the image grade; and comparing and classifying the numerical values of the three judging parameters with the classified data of the contamination grade of the insulator of the tower in the mining area, and taking the highest contamination grade in the three judging parameters as the final grade.
The invention has the beneficial effects that:
the optimized multi-image characteristic parameters obtained by the method can avoid multi-value deviation, and compare the size relationship between the optimized characteristic parameters and the defined pollution level; screening and judging the pollution grades of the insulator disc images of the towers in various mining areas. The algorithm can overcome the problem of multi-value deviation and improve the effectiveness of judgment.
In the prior art, the temperature rise of the surface of the insulator is only seen by infrared rays, and the invention performs combined judgment on the area of ultraviolet spots and the pollution chroma of visible light images on the basis of infrared rays, so that the classification of the pollution grades of the insulators in a mining area is more reasonable.
Drawings
FIG. 1 is a schematic flow chart of a method for identifying the contamination state of a pole and tower insulator in an ore district based on multi-image feature fusion, according to the invention;
FIG. 2 is a feature fusion diagram of the mining area pole tower insulator contamination state identification method based on multi-image feature fusion;
FIG. 3 is a schematic diagram of grading of the mining area pole tower insulator contamination state identification method based on multi-image feature fusion.
Detailed Description
Examples
As shown in fig. 1, a schematic flow chart of the method for identifying the contamination state of the insulator of the tower in the mining area based on the fusion of the multiple image features is shown. The invention discloses a mining area pole tower insulator contamination state identification method based on multi-image feature fusion, which comprises the following steps of:
step 1: and respectively acquiring a visible light image, an infrared image and an ultraviolet image of the insulator of the tower in the mining area from multiple angles by using a visible light camera, an infrared thermal imager and an ultraviolet imager.
Step 2: respectively carrying out image segmentation on the infrared image and the visible light image by adopting an improved seed region growing method; the method comprises the following specific steps:
1) The YCbCr color space can be obtained by carrying out image graying processing and converting the RGB color space, and the YCbCr color space can be obtained
The equation is as follows:
wherein, R, G, B respectively refer to the colors red, green, blue of RGB color space, Y, cb, cr respectively refer to the brightness component, blue chroma component, red chroma component of color space;
2) Extracting the color characteristic value and the texture characteristic value of the pixel point, selecting seeds by using the obtained characteristic values, and recording the seed points as (x) 0 ,y 0 ) (ii) a Selecting pixel points with the same gray value and points with the same adjacent gray value in the target image as initial areas, and setting the edges of the initial areas to contain n points: (x) 0 ,y 0 )1 、(x 0 ,y 0 )2 、 …(x 0 ,y 0 ) n As new seeds;
3) With seed point (x) 0 ,y 0 ) As a starting point, judging whether the gray difference Ds between the gray difference Ds and the pixel points in the n multiplied by n neighborhood meets the following growth rule or not: ds is a group of<δ s, where δ s is a set threshold valueThe visible light threshold is set to 15, and the infrared image threshold is set to 3;
4) If so, carrying out region combination, regarding the point (x, y) as a new seed point, repeating the step (3) on the new seed point (x, y) until the judgment condition is not satisfied, stopping growth, and obtaining a filthy growth region Z;
5) For seed point (x) 0 ,y 0 ) 1 、(x 0 ,y 0 ) 2 、…(x 0 ,y 0 ) n Repeating the steps (3) and (4) to obtain a growth area Z 1 、Z 2 、 …Zn；
6) Merging all the dirty growth areas to obtain a target Mn:
Mn= Z∪Z 1 ∪Z 2 ∪ … ∪Zn；
7) Traversing the seed pixel points, and finishing region combination to obtain a region Mn year which is the segmented image;
8) According to the step 2) to the step 7), dividing the insulator of the tower in the mining area into M areas M 1 、M 2 、…Mm；
9) Converting the YCbCr color space of each region into RGB color space respectively, wherein the formula is as follows:
and step 3: the method for selecting the original characteristic parameters of the visible light image, the infrared image and the ultraviolet image respectively by utilizing the Fisher criterion comprises the following steps:
step 3.1: for a visible light image, examining luminance components, blue chrominance components and C red chrominance components of red, green, blue and YCbCr color spaces of an RGB color space, and calculating the mean value, the maximum value, the minimum value, the range and the variance of each feature, wherein the total 6 color features are obtained;
for the infrared image, calculating the mean value, the median value, the maximum value, the minimum value, the mode value, the range, the variance, the skewness, the kurtosis, the energy and the entropy of the relative temperature of the insulator disc surface area, taking total 11 statistics as temperature characteristics, simultaneously setting a temperature range, and investigating the ratio of each temperature on the disc surface and the coverage rate of a temperature rise area;
for ultraviolet images, 20s are taken as a group, the mean value, the median value, the maximum value, the minimum value, the mode value, the range, the variance, the skewness, the kurtosis, the energy and the entropy of the spot areas of 400 images in each group are calculated, and 11 statistics are used as ultraviolet discharge characteristics in total.
Step 3.2: assuming that there are n samples in the data set, dividing the samples into M categories, and recording M 1 ,m 2 ,⋯,m M Each class respectively containing n k A sample;
and &>
For the intra-class variance and the inter-class variance of the kth dimension feature on the training set, the expressions are respectively:
in the formula, x k Expressed as the k-dimension characteristic value of a single individual, c k Expressed as the mean of the k-dimensional features, c is expressed as the mean of all features;
by inter-class variance
And intra-class variance>
Ratio J F (k) To judge the degree of distinction of the features, the feature function is
In the formula, J F (k) The method is characterized in that the method is a Fisher criterion of the characteristics, the larger the function value is, the higher the distinguishing degree of the characteristics is, and three characteristics which can better reflect pollution identification to visible light images, infrared images and ultraviolet images are respectively extracted;
after the insulator of the tower in the mining area selected by the embodiment is processed, the visible light image finally selects the maximum values of the brightness component and the C red chrominance component of the red and YCbCr color spaces of each group of RGB color space, the infrared temperature rise image selects the mean value, the energy and the maximum value of each group of temperature rise, and the ultraviolet image selects the mean value, the range and the skewness of the area of each group of light spots;
normalizing each group of characteristic quantities to realize data comparability, wherein the normalization formula is
Wherein v represents a feature, d represents a number of feature groups, and j represents a sample number;
represents the jth feature v, <' > in the feature group d>
Represents the minimum value of a feature v in the feature group d>
Represents the minimum value of the feature v in the feature group d;
and taking the evolution of the square sum of the three characteristic normalization values as an evaluation parameter of the image level, namely information Gain (Ai).
In this embodiment, the results of calculating and identifying efficiency Gain (Ai) of three characteristics selected from the visible light image, the infrared temperature rise image and the ultraviolet image are as follows:
visible light image: gain (maximum value of red) =0.32
Gain (maximum value of luminance component) =0.43
Gain (maximum value of red chroma component) =0.38
Infrared temperature rise image: gain (mean value of temperature rise) =0.42
Gain (energy of temperature rise) =0.54
Gain (energy maximum) =0.25
Ultraviolet image: gain (mean value of spot area) =0.35
Gain (range of spot area) =0.24
Gain (deviation of spot area) =0.21
And 4, step 4: trimming and associating the optimized features by using an improved ID3 algorithm, namely constructing a data set, wherein three image features are mutually referenced, and the ID3 algorithm is adopted to realize decision-level fusion of images according to feature parameter data in consideration of different recognition characteristics and efficiencies of the three images; then when the equalization coefficient is used to obtain the optimized information gain, the steps include:
1) Based on the visible light image, the infrared temperature rise image and the ultraviolet image obtained in the step 3, the recognition efficiencies Gain (Ai) of the 9 features selected are respectively Gain (maximum red value), gain (maximum brightness component), gain (maximum red chromaticity component), gain (mean temperature rise), gain (energy of temperature rise), gain (maximum energy value), gain (mean light spot area), gain (range of light spot area) and Gain (bias of light spot area), a filthy image feature data sample set S is established, the information gains Gain (Ai) of the maximum value of the feature Ai red, the maximum brightness component, the maximum red chromaticity component, the mean temperature rise, the energy of temperature rise, the maximum energy, the mean light spot area, the range of light spot area and the bias of light spot area are corrected to obtain corrected information Gain (Ai), and the expression is as follows:
wherein, the first and the second end of the pipe are connected with each other,n 1 the number of classifications of the contamination level is indicated,f(n 1 ) In order to modify the parameters of the device,
number of classifications of each contamination classn 1 The larger the correction parameter is, the larger the correction parameter isf(n 1 ) The smaller; when the classification of the pollution grades is more, the correction parameter is smaller, the corrected information gain is smaller, and therefore the influence of multi-value deviation on the information gain is corrected, n 1 A value of 4;
3) In order to avoid excessive gain rate correction compensation caused by a certain characteristic, a characteristic deviation threshold T is introduced, the degree of multi-value deviation is measured and controlled, and for a set S, n is provided 2 The attribute bias threshold T is generally an average value of all conditional entropies E (Ai), and is expressed as follows
Wherein n is 2 Expressing the number of the features, wherein the conditional entropy E (Ai) is the discrimination of each feature; the attribute deviation thresholds T of the insulators of the towers in the mining area selected by the embodiment are respectively T Visible light =0.38，T Infrared ray =0.40，T Ultraviolet light =0.27。
4) The method comprises the following steps of balancing the influence of multi-value deviation on information gain by using an equalization coefficient and correcting the information gain to realize the optimization of the information gain:
the effect of the equalization coefficient R (Ai) is equivalent to the optimal value of the effect of the combined action of the information Gain' (Ai) after the correction and the attribute deviation threshold T.
Visible light: r (maximum of red) =0.043;
r (maximum value of luminance component) =0.05;
r (maximum of red chroma component) =0.048;
infrared temperature rise image: r (mean value of temperature rise) =0.051;
r (energy of temperature rise) =0.057;
r (energy maximum) =0.043;
ultraviolet image: r (mean of spot areas) =0.038;
r (range of spot area) =0.032;
r (spot area skewness) =0.030;
multiplying the evolution of the square sum of the numerical values of the equalization coefficients R (Ai) by 100 percent to obtain an evaluation parameter of the image grade;
R visible light =（0.043 2 +0.05 2 +0.048 2 ） ½ ×100%=8.2%；
R Infrared ray =（0.051 2 +0.057 2 +0.043 2 ） ½ ×100%=8.8%；
R Ultraviolet light =（0.038 2 +0.032 2 +0.030 2 ） ½ ×100%=5.8%。
And 5: and (5) based on the optimized information gain obtained by the improved ID3 algorithm in the step (4), setting a threshold value, and classifying the pollution grades of the insulators of the towers in the mining area.
In step 5, based on the optimized information gain obtained by the improved ID3 algorithm in step 4, a characteristic threshold is set, and the pollution levels of the insulators on the towers in the mining area are classified, as shown in fig. 3.
Table 1 mining area pole tower insulator filthy grade classification table based on multi-image feature fusion
And comparing and classifying the numerical values of the three judging parameters with data in the grade classification table, and taking the highest pollution grade in the three judging parameters as the final grade. The evaluation parameter R obtained according to the embodiment of the invention Visible light 、R Infrared ray 、R Ultraviolet light And comparing and classifying the numerical values of the three judging parameters with the data in the grade classification table, and classifying the highest pollution grade-the pollution grade in the three judging parameters into a second grade as a final grade.
The existing single image characteristic pollution grade determination method is adopted to classify based on a single image characteristic mining area pole and tower insulator pollution grade classification table in the table 2, the visible light image R (red) component averaging method pollution grade classification result of the mining area pole and tower insulator in the embodiment is two-grade, the infrared temperature rise image temperature rise maximum value method pollution grade classification result is two-grade, and the ultraviolet image spot area ratio method pollution grade classification result is two-grade, which is consistent with the result of the embodiment of the invention.
Table 2 classification table for insulator contamination grade of mining tower based on single image characteristics
By continuously measuring the pollution levels of 100 insulators, the error rate is within 2 percent, the method can avoid the influence of single image characteristic detection error on the judgment precision, and can realize more precise grade division so as to adopt proper decontamination measures.
The present invention is not limited to the above embodiments, but various modifications and changes can be made by those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (6)
1. A mining area pole tower insulator contamination state identification method based on multi-image feature fusion is characterized by comprising the following steps:
step 1: respectively utilizing a visible light camera, an infrared thermal imager and an ultraviolet imager to obtain visible light images, infrared images and ultraviolet images of the insulator of the tower in the mining area from multiple angles;
step 2: respectively carrying out image segmentation on the infrared image and the visible light image by adopting an improved seed region growing method;
and step 3: selecting original characteristic parameters of the visible light image, the infrared image and the ultraviolet image respectively by utilizing a Fisher criterion;
and 4, step 4: trimming and correlating the characteristic parameters by using an improved ID3 algorithm, and obtaining optimized information gain by using an equalization coefficient;
and 5: and (4) based on the optimized information gain obtained by the improved ID3 algorithm in the step (4), setting a threshold value, and classifying the insulator pollution levels of the towers in the mining area.
2. The mining area tower insulator contamination state identification method based on multi-image feature fusion according to claim 1, wherein in the step 2, when an improved seed region growing method is adopted to respectively carry out image segmentation on the infrared image and the visible light image, the method specifically comprises the following steps:
1) The YCbCr color space can be obtained by carrying out image graying processing and converting the RGB color space, and the conversion formula is as follows:
wherein, R, G, B respectively refer to red, green, blue, Y, cb, cr respectively refer to luminance component, blue chrominance component, red chrominance component of the color space;
2) Extracting the color characteristic value and the texture characteristic value of the pixel point, selecting seeds by using the obtained characteristic values, and recording the seed points as (x) 0 ,y 0 ) (ii) a Selecting pixel points with the same gray value and points with the same adjacent gray value in the target image as initial areas, and setting the edges of the initial areas to contain n points: (x) 0 ,y 0 )1 、(x 0 ,y 0 )2 、 …(x 0 ,y 0 ) n As new seeds;
3) With seed points (x) 0 ,y 0 ) As a starting point, judging whether the gray difference Ds between the pixel point and the pixel point in the n multiplied by n neighborhood meets the following growth rule or not: ds is a group of<δ s, wherein δ s is a set threshold;
4) If so, carrying out region combination, regarding the point (x, y) as a new seed point, repeating the step 3) on the new seed point (x, y) until the judgment condition is not satisfied, stopping growth, and obtaining a filthy growth region Z;
5) For seed point (x) 0 ,y 0 )1 、(x 0 ,y 0 )2 、…(x 0 ,y 0 ) n, repeating the steps (3), (b), (c) and (d)4) Obtaining a growth region Z 1 、Z 2 、 …Zn；
6) Merging all the dirty growth areas to obtain a target Mn:
Mn= Z∪Z 1 ∪Z 2 ∪ … ∪Zn；
7) Traversing the seed pixel points, and finishing region combination to obtain a region Mn year which is the segmented image;
8) According to the step 2) to the step 7), dividing the insulator of the tower in the mining area into M areas M 1 、M 2 、…Mm；
9) Converting the YCbCr color space of each region into the RGB color space respectively, wherein the formula is as follows:
3. the mining area tower insulator contamination state identification method based on multi-image feature fusion as claimed in claim 2, wherein a visible light threshold is set to be δ s =15, and an infrared image threshold is set to be δ s =3.
4. The mining area tower insulator contamination state identification method based on multi-image feature fusion as claimed in claim 1, wherein the step 3 of selecting original feature parameters of the visible light image, the infrared image and the ultraviolet image comprises the following steps:
step 3.1: for a visible light image, examining a luminance component, a blue chrominance component and a C red chrominance component of a red, green, blue and YCbCr color space of an RGB color space, and calculating the mean value, the maximum value, the minimum value, the range and the variance of each feature, wherein the total number of 6 color features;
for the infrared image, calculating the mean value, the median value, the maximum value, the minimum value, the mode value, the range, the variance, the skewness, the kurtosis, the energy and the entropy of the relative temperature of the surface region of the insulator disc, taking total 11 statistics as temperature characteristics, simultaneously setting a temperature range, and inspecting the ratio of each temperature on the surface of the disc and the coverage rate of a temperature rise region;
for ultraviolet images, 20s are taken as a group, the mean value, the median value, the maximum value, the minimum value, the mode value, the range, the variance, the skewness, the kurtosis, the energy and the entropy of the light spot areas of 400 images in each group are calculated, and 11 statistics are taken as ultraviolet discharge characteristics in total;
step 3.2: assuming that there are n samples in the data set, the samples are classified into M categories, and M is recorded 1 ,m 2 ,⋯,m M Each class respectively containing n k A sample is obtained;
and &>
For the intra-class variance and inter-class variance of the kth dimension features on the training set,
the expressions are respectively:
in the formula, x k Expressed as the k-dimension characteristic value of a single individual, c k Expressed as the mean of the k-dimension features, c is expressed as the mean of all features;
by inter-class variance
And in-class variance ≥>
Ratio J F (k) To judge the degree of distinction of the features, the feature function is
J F (k) The method is characterized in that the method is a Fisher criterion of the characteristics, the larger the function value is, the higher the distinguishing degree of the characteristics is, and three characteristics which can better reflect pollution identification to visible light images, infrared images and ultraviolet images are respectively extracted;
normalizing each group of characteristic quantities to realize data comparability, wherein the normalization formula is
Wherein v represents a feature, d represents a number of feature groups, and j represents a sample number;
representing the jth feature v in the set of features d,
represents the minimum value of the characteristic v in the characteristic group d, is/are>
Represents the minimum value of the feature v in the feature group d;
and taking the evolution of the square sum of the three characteristic normalization values as an evaluation parameter of the image level, namely information Gain (Ai).
5. The mining area tower insulator contamination state identification method based on multi-image feature fusion as claimed in claim 1, wherein in step 4, the optimized features are trimmed and associated by using an improved ID3 algorithm, and when the optimized information gain is obtained by using the equalization coefficient, the steps include:
1) Obtaining the recognition efficiency Gain (Ai) of each feature in each type of image based on the step 3;
2) Establishing a filth image characteristic data sample set S, and correcting the information Gain (Ai) of the characteristic Ai to obtain a corrected information Gain' (Ai), wherein the expression is as follows:
wherein the content of the first and second substances,n 1 the number of classifications of the contamination level is represented,f(n 1 ) In order to modify the parameters of the device,
number of classifications of each pollution classn 1 The larger the correction parameter isf(n 1 ) The smaller; when the classification of the pollution grades is more, the correction parameter is smaller, and the corrected information gain is smaller, so that the influence of multi-value deviation on the information gain is corrected;
3) In order to avoid excessive gain rate correction compensation caused by a certain characteristic, a characteristic deviation threshold T is introduced, the degree of multi-value deviation is measured and controlled, and for a set S, the method comprises the following stepsn2Characteristic, attribute bias threshold T is usually taken to be all conditional entropiesE(Ai) Is expressed as follows
Wherein n is 2 Number of representation features, conditional entropyE(Ai) The degree of distinction of each characteristic;
4) And balancing the influence of multi-value deviation on the information gain and correcting the information gain by using the equalization coefficient to realize the optimization of the information gain.
6. The mining area pole and tower insulator pollution state identification method based on multi-image feature fusion as claimed in claim 1, wherein the threshold is set to classify the mining area pole and tower insulator pollution grades as follows:
taking the evolution of the square sum of the numerical values of the equalization coefficient R (Ai) multiplied by 100 percent as the judgment parameter of the image grade; and comparing and classifying the numerical values of the three judging parameters with classified data of the pollution grades of the insulators of the towers in the mining area, and taking the highest pollution grade in the three judging parameters as a final grade.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211505793.3A CN115901796A (en) | 2022-11-29 | 2022-11-29 | Mining area tower insulator contamination state identification method based on multi-image feature fusion |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211505793.3A CN115901796A (en) | 2022-11-29 | 2022-11-29 | Mining area tower insulator contamination state identification method based on multi-image feature fusion |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115901796A true CN115901796A (en) | 2023-04-04 |
Family
ID=86475878
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211505793.3A Pending CN115901796A (en) | 2022-11-29 | 2022-11-29 | Mining area tower insulator contamination state identification method based on multi-image feature fusion |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115901796A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117611784A (en) * | 2023-11-30 | 2024-02-27 | 华北电力大学 | Cooperative control method and device for multispectral high-timeliness discharge detection |
-
2022
- 2022-11-29 CN CN202211505793.3A patent/CN115901796A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117611784A (en) * | 2023-11-30 | 2024-02-27 | 华北电力大学 | Cooperative control method and device for multispectral high-timeliness discharge detection |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108734689B (en) | Method for detecting scattered strands of conducting wires based on region growth | |
| CN111798412B (en) | Intelligent diagnosis method and system for defects of power transformation equipment based on infrared image | |
| CN108765373B (en) | Insulator abnormity automatic detection method based on integrated classifier online learning | |
| CN114359416B (en) | Building outer wall hollowing leakage abnormity detection and positioning method | |
| CN105445607B (en) | A kind of electrical equipment fault detection method drawn based on thermoisopleth | |
| CN106680285B (en) | Method for recognizing insulator contamination state based on infrared image assisted visible light image | |
| CN103196380B (en) | Transmission line of electricity image collecting device and electric power line ice-covering thickness detection system | |
| CN107727662B (en) | Battery piece EL black spot defect detection method based on region growing algorithm | |
| CN106950472B (en) | insulator detection method based on infrared and ultraviolet imaging | |
| CN111402247B (en) | Machine vision-based method for detecting defects of suspension clamp on power transmission line | |
| CN115409833A (en) | Hot spot defect detection method of photovoltaic panel based on unsharp mask algorithm | |
| CN103234465A (en) | Method and system for detecting icing thickness of electric transmission line | |
| CN112435208B (en) | Insulator region detection method for significant edge guidance | |
| CN115901796A (en) | Mining area tower insulator contamination state identification method based on multi-image feature fusion | |
| CN106408025A (en) | Classification and recognition method of aerial image insulators based on image processing | |
| CN111008967B (en) | Insulator RTV coating defect identification method | |
| CN109523543A (en) | A kind of wire strand breakage detection method based on Edge Distance | |
| CN116758081B (en) | Unmanned aerial vehicle road and bridge inspection image processing method | |
| CN110490261B (en) | Positioning method for power transmission line inspection image insulator | |
| CN111951247B (en) | Power equipment ultraviolet image diagnosis method and system | |
| CN203190947U (en) | Transmission line image acquisition device and transmission line icing thickness detection system | |
| CN114581407B (en) | Self-adaptive defect detection method for photovoltaic module | |
| CN110363773B (en) | Cable type detection system and method based on image processing | |
| CN112037137A (en) | Method and device for eliminating fuzzy region of insulator disc surface edge in infrared image | |
| CN112113667A (en) | Current transformer fault diagnosis method based on infrared thermal image information |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |