CN108765373A - A kind of insulator exception automatic testing method based on integrated classifier on-line study - Google Patents
A kind of insulator exception automatic testing method based on integrated classifier on-line study Download PDFInfo
- Publication number
- CN108765373A CN108765373A CN201810386292.5A CN201810386292A CN108765373A CN 108765373 A CN108765373 A CN 108765373A CN 201810386292 A CN201810386292 A CN 201810386292A CN 108765373 A CN108765373 A CN 108765373A
- Authority
- CN
- China
- Prior art keywords
- insulator
- sample
- classifier
- image
- training
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/12—Edge-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/136—Segmentation; Edge detection involving thresholding
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/155—Segmentation; Edge detection involving morphological operators
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of insulator exception automatic testing method based on integrated classifier on-line study, the specific steps are, high-definition tripod video camera is carried by carrier of unmanned plane/shaft tower, shoot transmission line of electricity visible light insulation subgraph, the image got is pre-processed, image segmentation, image multi-feature extraction etc., obtain a variety of characteristic features of the abnormal subgraph that insulate and makees normalized；Then trade-off decision tree is as Weak Classifier, Weak Classifier is trained using normalization characteristic as its categorical attribute, repetition training obtains several decision trees and its classified weight, the high strong classifier of classification accuracy is generated finally by Nearest Neighbor with Weighted Voting, it timely updates strong classifier in combination with online learning art, the all types of abnormal subgraphs that insulate are separated, to do subsequent processing.The principle of the invention is simple, intuitive easy, abnormal in conjunction with image processing techniques and machine learning algorithm Weigh sensor diagnosing insulation, and new approaches and method are provided for insulator misoperation monitoring.
Description
Technical field
The invention belongs to electric transmission line isolator abnormality detection, machine learning and technical field of image processing, specifically relate to
And a kind of insulator exception automatic testing method based on integrated classifier on-line study.
Background technology
Insulator dosage in power grid is huge, type is various, is one of critical component of high pressure overhead power line, occupies
Very important status.On the one hand, it is responsible for the mechanical supporting function of conducting wire；On the other hand, play a part of insulation, prevent electric current
Channel ground connection is formed over the ground.But insulator is due to being chronically exposed in wild environment, it is more likely that various electrical and machines occurs
Tool failure, it is possible to cause power supply system to be interrupted, can even power grid be caused to paralyse under serious conditions, therefore insulator direct relation
To the safe and stable operation of transmission line of electricity.
In order to improve the intelligent level of line data-logging, digital image processing techniques are in insulator identification and state-detection
Application study is more and more.Insulator non-contact detection based on image procossing is low with danger, interference is small, simple equipments
Advantage has carried out some representative work both at home and abroad, but majority concentrates on the single defect recognition of insulator, cannot be certainly
Dynamic all kinds of abnormal insulators of detection so that there are certain difficulties for intelligent isolator detecting processing.
In addition, with the development of artificial intelligence, the machine learning of one of core research field is increasingly becoming modern intelligence
The key link and bottleneck of system.Machine learning is usually used in solving classification problem, by the study to training data, obtains wherein
Rule, then using study result (grader) to new data carry out class prediction.And integrated study (Ensemble
Learning it is) current one of the research hotspot in machine learning field, is all shown in the task of various different fields good
Stability, therefore be widely used.But currently with machine learning combination digital image processing techniques to insulator
The abnormal research for carrying out online classification detection is also seldom.
Invention content
The abnormal automatic detection side of insulator that the object of the present invention is to provide a kind of based on integrated classifier on-line study
Method burns out three classes exception insulator for the self-destruction urgently replaced, crackle, enamel and carries out fast and accurately analysis and detect, reduces
Dangerous odds.
The technical solution adopted by the present invention is a kind of abnormal automatic detection of the insulator based on integrated classifier on-line study
Method, which is characterized in that be specifically implemented according to the following steps：
Step 1, it acquires insulation subgraph and obtains insulation subsample D to be measured, specifically, unmanned plane or shaft tower is utilized to carry
High-definition tripod camera acquisition insulation subgraph determines that its best shooting is burnt wherein by constantly regulate video camera shooting angle
Away from so that it is guaranteed that the image collected is more easy to extract the effective characteristic parameter of variety classes insulator；
Step 2, training sample set is established, specifically, determining the insulator sample set of fault type as offline using oneself
Training sample set trains grader, the Exception Type of coded samples insulator, normal, self-destruction, crackle, enamel are burnt out and its
His five quasi-modes are separately encoded as Arabic numerals 1,2,3,4 and 5, as the category corresponding to attribute value in the sample set
Label；
Step 3, exhausted to the training of collected insulation subsample D to be measured in step 1 and the determination Exception Type in step 2
Edge subsample T is pre-processed, i.e., adaptively equal with contrast by collected insulation subgraph after gray processing is handled
Then weighing apparatusization method isolates target insulation subregion, by the new of acquisition to image enhancement using adaptive threshold fuzziness method
Bianry image be mapped on original sample, extract the insulation subgraph accurately divided, obtain sample to be tested D ' and training sample
T'；
Step 4, the image being partitioned into step 3 in the sample to be tested D ' and training sample T ' of insulation subregion is carried out special
Sign extraction, acquisition insulator essential characteristic fusion similarity measure values dHWD, insulator distance between commutator segments Δ l, insulator marginal information
IRDAnd its rate of gray levelEtc. features, it is made after normalized as sample attribute value construction training sample
This collection S and test sample collection P；
Step 5, the decision tree is recycled by the training sample set S as Weak Classifier using decision tree
Training, until obtaining Weak Classifier anticipation function htAnd its error rate,
Step 6, the Weak Classifier gone out according to off-line training in step 5 generate final strong classification using the ballot method of Weight
Device anticipation function h0 fin(z)；
That is, all Weak Classifiers is allowed to vote, then voting results are obtained most according to error rates of weak classifiers weighted sum
Whole strong classifier anticipation function, as shown in formula (7)
Step 7, test sample collection P described in step 4 is sequentially input in step 6 gained strong classifier, through described final
After strong classifier function judges identification, the class label result v of corresponding exception is obtainedfin；
Step 8, by step 7 output testing result vfinThe decision tree classifier classification results minimum with error rate in step 5
vminComparing calculation is carried out by formula (8),
That is, working as vfinAnd vminThe class label output insulator Exception Type V is matched when the two is consistentfin, complete diagnosis；
Otherwise, work as vfinAnd vminWhen the two is inconsistent, step 9 is executed；
Step 9, when judging in step 8, there is vfin≠vmin, that is, there are strong classifier classification results obtained by off-line training
When with decision tree result difference, grader is made to carry out on-line study process, timely update grader, and carries out detection next time
Judge.
The features of the present invention also characterized in that
In step 4, in the training sample set S, including 5 class insulator breakdowns, there are m group numbers per class insulator breakdown
According to then training sample set S is expressed as：
S=<(u11,v11),…,(u1m,v1m),…,(u5m,v5m)>,
Wherein, it is ω that i-th of training sample, which has equal initial weight,i=1/5m；The value range of i is [1, m].
Each feature normalization calculating process in the step 4 specifically,
Step 4.1, to essential characteristic fusion similarity measure values dHWDIt is calculated,
Step 4.2, to insulator distance between commutator segments Δ l, this feature is normalized,
Step 4.3, marginal information IRDThis feature is normalized,
Step 4.4, to rate of gray levelThis feature is normalized.
The step 4.1 specifically,
Step 4.1.1, the calculation formula (1) that similarity measure values are merged using normalized essential characteristic carry out spy substantially
Sign fusion similarity measure values dHWDCalculating,
Step 4.1.2 measures single features using histogram intersection method, obtains the measuring similarity of each feature
Value,
Histogram intersection method formula is：
Wherein, M and N is respectively the histogram of original normal insulation subgraph p and input sample q same dimensions, d (M, N)
The normalization of histogram intersection is indicated as a result, its value is closer to 1, then two Histogram Matchings is better.If image p and image
It is 1 that q normalization histograms, which exactly match then histogram intersection institute value, and complete mismatch value is 0.
The step 4.2 obtains insulation sub-connected domain specifically, the image concentrated to training sample is split processing, obtains
It obtains each connected domain center and indicates sub-pieces center, acquire the distance between adjacent sub-pieces of normalization：
Wherein, Δ l indicates adjacent insulator distance between commutator segments,L indicates insulator chain
Total length.(X1,Y1),(X2,Y2) be respectively two neighboring connected domain center point coordinate.
The step 4.3 specifically,
Step 4.3.1 carries out piecemeal to image first with adaptive multi-thresholding algorithm, and determines the best of segmented areas
Get over a little；Then use linear membership function by image pixel values transformation for fuzzy field, by getting over the mould for calculating proposition
The relevant parameter for pasting hyperbolic tangent function carries out nonlinear images increasing to each segmented areas application Retinex algorithm of image
By force；
Step 4.3.2 calls the binary segmentation that Canny operators carry out image to obtain the potential area edge of local anomaly, to it
Retention area, the maximum connected domain of length after progress dilation operation, calculate two features of its circularity R and length-width ratio D, work as circle
It spends R and is less than R0And length-width ratio D is more than D0When judgement artwork in there are crackle, circularity R calculates as follows:Wherein s
For connected domain area, C is connected domain perimeter.When region is round, R values are up to 1.
Length-width ratio D calculates as follows:Wherein, L is the length of connected domain, that is, surrounds the minimum rectangle of connected domain
Length.
Obtained normalization marginal information is denoted as：IRD=(R≤R0)&&(L≤L0)。
The step 4.4 specifically,
Step 4.4.1, to handling obtained marginal information I through step 4.3RDObtained image img after screening is added
Weight average gray processing, and average gray is weighted to image img by formula (3), obtain weighted average gray level image
Grays, specific computational methods are as follows：
The gray value after weighted average gray processing at coordinate points (x, y) is calculated using formula (3),
Grays (x, y)=(WrR(x,y)+WgG(x,y)+WbB(x,y))/3 (3)
Wherein, Wr=0.299, Wg=0.587, Wb=0.114, grays (x, y) indicate coordinate after weighted average gray processing
Gray value at point (x, y),
Step 4.4.2, using step 4.4.1 calculating as a result, formula (4) is pushed away to obtain by formula (3), according to formula (4) meter
Calculate longitudinal gray value of gray level image：
graysc_y(x)=grays (x, y=c) 1≤c≤Wcolumn (4)
Wherein, c indicates the row coordinate of longitudinal hatching line position, WcolumnIndicate the width of row；
It obtainsIndicate normalized longitudinal gray-value variation rate.
The specific step of the step 5 is：
Step 5.1, by normalized image featureIt is inputted as sample attribute
Weak Classifier, Weak Classifier algorithm by the weights based on training sample set before larger 80% carry out structure grader operation,
Generate corresponding decision tree ht:U → V,
Step 5.2, assess single classifier performance, in total training sample other than screening the sample for being used as training institute
There is sample all by the test sample as grader caused by assessment epicycle iteration.The error rate h of the base gradert:erriBy
Formula (5) can be calculated：
Wherein, ωt(i) weight of i-th of sample in t wheels is indicated；
Step 5.3, weight being adjusted according to error rate, promotes the weight of misjudgement sample, specific method is,
It enablesThen,
Wherein, αtThe Weak Classifier h generated for t wheelstEvaluation points, classification accuracy rate is higher, value it is bigger, it is on the contrary then
It is smaller；ZtIt is a normalization constant.ωt+1(i) it is the weight of i-th group of data of t+1 wheels；
The weight for distinguishing correct sample is reduced with this, will more consider that these are wrong in this way in training later
The sample sentenced；
On the basis of sample weights after the adjustment, repeats step 5.1 and 5.2 and be trained the new grader of generation again, i.e.,
The general Weak Classifier of several detectabilities.
The step 9 is specially：
Step 9.1, it will be added in training sample set S centered on the inconsistent sample of step 8 acquired results, utilize kNN
Neighbour's thought, the essential characteristic fusion similarity measure values that sample image is obtained according to the method for step 4.1 find out it away from minimum
K (k=m) a sample in the maximum a kind of reference category as the sample of quantity, matched, taken with above two result
The high result of probability as the sample classification and using the sample as on-line training sample replace original training sample collection correspond to
A minimum sample of weight in classification, to constitute new training sample set S '；
Step 9.2, return to step 5, the weight of training sample, is arranged according to off-line training empirical value and instructs after initialization update
Practice number, train Weak Classifier again, count its number correctly or incorrectly classified, find out simplified Weak Classifier error rate,
Sample weights are updated with this；
Step 9.3, according to step 6 strong classifier generating process, updated strong classifier h is obtainedn fin(x), wherein n
Newer number is represented, takes 0,1,2 ....
The invention has the advantages that
(1) compared with the single abnormality diagnostic method of existing insulator, the present invention proposes to classify using integrated Adaboost
Device presorts to target insulation subgraph, and the various exceptions of insulator are realized by image processing techniques and machine learning algorithm
Detection, can more intelligently realize on-line monitoring and fault diagnosis, reduce manual operation program, for modern transmission line of electricity
Insulator state maintenance provides a kind of effective means.
(2) the present invention is based on the insulator exception automatic testing method of integrated classifier on-line study, towards revealing, split
Line, enamel such as burn out at several abnormal insulators, on the basis of analyzing its characteristic feature with image processing techniques, use
Adaboost graders carry out the abnormal automatic detection of insulator, timely update grader in conjunction with on-line learning algorithm, improve classification
Device accuracy rate.This method principle is simple, intuitive easy, and a kind of new detection think of is provided for the safe and stable operation of transmission line of electricity
Road.
Description of the drawings
Fig. 1 is a kind of flow of the insulator exception automatic testing method based on integrated classifier on-line study of the present invention
Figure；
Fig. 2 is involved in a kind of insulator exception automatic testing method based on integrated classifier on-line study of the present invention
Integrated Adaboost grader structure charts；
Fig. 3 is involved in a kind of insulator exception automatic testing method based on integrated classifier on-line study of the present invention
Decision-tree model figure.
Specific implementation mode
The following describes the present invention in detail with reference to the accompanying drawings and specific embodiments.
The present invention is based on the insulator exception automatic testing methods of integrated classifier on-line study, as shown in Figure 1, specifically pressing
Implement according to following steps：
Step 1, it acquires insulation subgraph and obtains insulation subsample D to be measured, specifically, unmanned plane or shaft tower is utilized to carry
High-definition tripod camera acquisition insulation subgraph determines that its best shooting is burnt wherein by constantly regulate video camera shooting angle
Away from so that it is guaranteed that the image collected is more easy to extract the effective characteristic parameter of variety classes insulator.
Step 2, training sample set is established, specifically, determining the insulator sample set of fault type as offline using oneself
Training sample set trains grader, the Exception Type of coded samples insulator, normal, self-destruction, crackle, enamel are burnt out and its
His five quasi-modes are separately encoded as Arabic numerals 1,2,3,4 and 5, as the category corresponding to attribute value in the sample set
Label.
Step 3, exhausted to the training of collected insulation subsample D to be measured in step 1 and the determination Exception Type in step 2
Edge subsample T is pre-processed, i.e., adaptively equal with contrast by collected insulation subgraph after gray processing is handled
Then weighing apparatusization method isolates target insulation subregion, by the new of acquisition to image enhancement using adaptive threshold fuzziness method
Bianry image be mapped on original sample, extract the insulation subgraph accurately divided, obtain sample to be tested D ' and training sample
T’。
Step 4, the image being partitioned into step 3 in the sample to be tested D ' and training sample T ' of insulation subregion is carried out special
Sign extraction, acquisition insulator essential characteristic fusion similarity measure values dHWD, insulator distance between commutator segments Δ l, insulator marginal information
IRDAnd its rate of gray levelEtc. features, it is made after normalized as sample attribute value construction training sample
This collection S and test sample collection P.
In the wherein described training sample set S, including 5 class insulator breakdowns, there are m group data per class insulator breakdown, then
Training sample set S is expressed as：
S=<(u11,v11),…,(u1m,v1m),…,(u5m,v5m)>,
Wherein, it is ω that i-th of training sample, which has equal initial weight,i=1/5m；The value range of i is [1, m].
Each feature normalization calculating process is as follows：
Step 4.1, to essential characteristic fusion similarity measure values dHWDIt being calculated, specific computational methods are,
Step 4.1.1, the calculation formula (1) that similarity measure values are merged using normalized essential characteristic carry out spy substantially
Sign fusion similarity measure values dHWDCalculating,
Wherein, dH、dW、dDColor characteristic similarity, textural characteristics similarity, shape after image normalization are indicated respectively
Characteristic similarity.dHWDTo normalize final similarity.
Step 4.1.2 measures single features using histogram intersection method, obtains the measuring similarity of each feature
Value,
Histogram intersection method formula is：
Wherein, M and N is respectively the histogram of original normal insulation subgraph p and input sample q same dimensions, d (M, N)
The normalization of histogram intersection is indicated as a result, its value is closer to 1, then two Histogram Matchings is better.If image p and image
It is 1 that q normalization histograms, which exactly match then histogram intersection institute value, and complete mismatch value is 0.
Step 4.2, to insulator distance between commutator segments Δ l, this feature is normalized,
Processing is split to the image that training sample is concentrated and obtains insulation sub-connected domain, each connected domain center is obtained and indicates
The distance between adjacent sub-pieces of normalization are acquired at sub-pieces center：
Wherein, Δ l indicates adjacent insulator distance between commutator segments,L indicates insulator chain
Total length.(X1,Y1),(X2,Y2) be respectively two neighboring connected domain center point coordinate.
Step 4.3, marginal information IRDThis feature is normalized,
Step 4.3.1 carries out piecemeal to image first with adaptive multi-thresholding algorithm, and determines the best of segmented areas
Get over a little；Then use linear membership function by image pixel values transformation for fuzzy field, by getting over the mould for calculating proposition
The relevant parameter for pasting hyperbolic tangent function carries out nonlinear images increasing to each segmented areas application Retinex algorithm of image
By force；
Step 4.3.2 calls the binary segmentation that Canny operators carry out image to obtain the potential area edge of local anomaly, to it
Retention area, the maximum connected domain of length after progress dilation operation, calculate two features of its circularity R and length-width ratio D, work as circle
It spends R and is less than R0And length-width ratio D is more than D0When judgement artwork in there are crackle,
Circularity R calculates as follows:Wherein s is connected domain area, and C is connected domain perimeter.When region is circle
R values are up to 1 when shape.
Length-width ratio D calculates as follows:Wherein, L is the length of connected domain, that is, surrounds the minimum rectangle of connected domain
Length.
Obtained normalization marginal information is denoted as：IRD=(R≤R0)&&(L≤L0)。
Step 4.4, to rate of gray levelThis feature is normalized, the specific steps are,
Step 4.4.1, to handling obtained marginal information I through step 4.3RDObtained image img after screening is added
Weight average gray processing, and average gray is weighted to image img by formula (3), obtain weighted average gray level image
Grays, specific computational methods are as follows：
The gray value after weighted average gray processing at coordinate points (x, y) is calculated using formula (3),
Grays (x, y)=(WrR(x,y)+WgG(x,y)+WbB(x,y))/3 (3)
Wherein, Wr=0.299, Wg=0.587, Wb=0.114, grays (x, y) indicate coordinate after weighted average gray processing
Gray value at point (x, y),
Step 4.4.2, using step 4.4.1 calculating as a result, formula (4) is pushed away to obtain by formula (3), according to formula (4) meter
Calculate longitudinal gray value of gray level image：
graysc_y(x)=grays (x, y=c) 1≤c≤Wcolumn (4)
Wherein, c indicates the row coordinate of longitudinal hatching line position, WcolumnIndicate the width of row；
It obtainsIndicate normalized longitudinal gray-value variation rate.
Step 5, the decision tree is recycled by the training sample set S as Weak Classifier using decision tree
Training, until obtaining Weak Classifier anticipation function htAnd its error rate, specific step are：
Step 5.1, by normalized image featureIt is inputted as sample attribute
Weak Classifier, Weak Classifier algorithm by the weights based on training sample set before larger 80% carry out structure grader operation,
Generate corresponding decision tree ht:U → V, decision-tree model are as shown in Figure 3；
Step 5.2, assess single classifier performance, in total training sample other than screening the sample for being used as training institute
There is sample all by the test sample as grader caused by assessment epicycle iteration.The error rate h of the base gradert:erriBy
Formula (5) can be calculated：
Wherein, ωt(i) weight of i-th of sample in t wheels is indicated.
Step 5.3, weight being adjusted according to error rate, promotes the weight of misjudgement sample, specific method is,
It enablesThen,
Wherein, αtThe Weak Classifier h generated for t wheelstEvaluation points, classification accuracy rate is higher, value it is bigger, it is on the contrary then
It is smaller；ZtIt is a normalization constant.ωt+1(i) it is the weight of i-th group of data of t+1 wheels；
The weight for distinguishing correct sample is reduced with this, will more consider that these are wrong in this way in training later
The sample sentenced.
On the basis of sample weights after the adjustment, repeats step 5.1 and 5.2 and be trained the new grader of generation again, i.e.,
The general Weak Classifier of several detectabilities；
Step 6, the Weak Classifier gone out according to off-line training in step 5 generate final strong classification using the ballot method of Weight
Device anticipation function h0 fin(z).That is, all Weak Classifiers is allowed to vote, then voting results are weighted according to error rates of weak classifiers
Summation obtains final strong classifier anticipation function, such as shown in (7)
Step 7, test sample collection P described in step 4 is sequentially input in step 6 gained strong classifier, through described final
After strong classifier function judges identification, the class label result v of corresponding exception is obtainedfin。
Step 8, by step 7 output testing result vfinThe decision tree classifier classification results minimum with error rate in step 5
vminComparing calculation is carried out by formula (8),
That is, working as vfinAnd vminThe class label output insulator Exception Type V is matched when the two is consistentfin, complete diagnosis；
Otherwise, work as vfinAnd vminWhen the two is inconsistent, step 9 is executed.
Step 9, when judging in step 8, there is vfin≠vmin, that is, there are strong classifier classification results obtained by off-line training
When with decision tree result difference, grader is made to carry out on-line study process, timely update grader, and carries out detection next time
Judge, specific step is：
Step 9.1, it will be added in training sample set S centered on the inconsistent sample of step 8 acquired results, utilize kNN
Neighbour's thought, the essential characteristic fusion similarity measure values that sample image is obtained according to the method for step 4.1 find out it away from minimum
K (k=m) a sample in the maximum a kind of reference category as the sample of quantity, matched, taken with above two result
The high result of probability as the sample classification and using the sample as on-line training sample replace original training sample collection correspond to
A minimum sample of weight in classification, to constitute new training sample set S '；
Step 9.2, return to step 5, the weight of training sample, is arranged according to off-line training empirical value and instructs after initialization update
Practice number, train Weak Classifier again, count its number correctly or incorrectly classified, find out simplified Weak Classifier error rate,
Sample weights are updated with this；
Step 9.3, according to step 6 strong classifier generating process, updated strong classifier h is obtainedn fin(x), wherein n
Newer number is represented, takes positive integer 0,1,2 ....
The method of the present invention has following characteristics, and compared with the single abnormality diagnostic method of existing insulator, this method carries
Go out and presorted to target insulation subgraph using integrated Adaboost graders (as shown in Figure 2), passes through image procossing skill
Art and machine learning algorithm realize the various abnormal detections of insulator, can more intelligently realize that on-line monitoring and failure are examined
It is disconnected, reduce manual operation program, a kind of effective means is provided for modern times electric transmission line isolator repair based on condition of component.Side of the present invention
Normal plane several abnormal insulators such as burns out to self-destruction, crackle, enamel, in the base for analyzing its characteristic feature with image processing techniques
On plinth, the abnormal automatic detection of insulator is carried out using Adaboost graders, is timely updated grader in conjunction with on-line learning algorithm,
Improve grader accuracy rate.This method principle is simple, intuitive easy, and one kind is provided newly for the safe and stable operation of transmission line of electricity
Detection thinking.
In a kind of insulator exception automatic testing method based on integrated classifier on-line study of the present invention, in step 4
There are many selections, such as KNN, BP neural network, decision tree etc. when the Weak Classifier of training Adaboost algorithm, are examined in of the invention
Consider the type and abnormal image changing features of insulator exception, chooses decision tree as Weak Classifier (as shown in Figure 3).
The input attribute of Weak Classifier is the phase that insulator essential characteristic merges similarity measure values, extracts in step 4
Adjacent insulator distance between commutator segments, the marginal information of image and gray value of image change rate.Reason is that normal insulation and template are exhausted
Its similarity measurements highest of edge, and insulator reveal after between piece spacing can have mutation；The side of porcelain insulator crack image
There are the connected domains that circularity and length-width ratio exceed definite value in edge information；And enamel burns out and is characterized on the image in the presence of exception
It is closed chain, mainly black enclosed region, the mutation of color can cause insulator gray value to reduce suddenly, close to zero, break
The regularity of the sub- rate of gray level of normal insulation.
On-line training is carried out to gained strong classifier in step 6, so as to the strong classifier constantly updated, cardinal principle
It is, by comparing the Integrated Algorithm classification results of decision Tree algorithms and Adaboost+ decision trees, to be carried towards new input test image
Misclassification target is taken out, on-line training is carried out, achievees the purpose that the grader that timely updates.
The classifying identification method designed in the present invention is applied in the 150 width test insulation subgraph of inspection acquisition and is carried out
Verification, including 50 width normal insulation subgraphs, 25 width self-destruction insulation subgraph, 25 width have crackle insulation subgraph, 20 width
Enamel burn out insulation subgraph and 30 width other extremely insulation subgraphs, classification prediction result it is as shown in table 1.
1 insulator anomaly classification result of table
The experimental results showed that the present invention is based on the insulator exception automatic testing method of integrated classifier on-line study, it is right
The identification classification accuracies of abnormalities such as burn out up to 85% or more in insulator self-destruction, crackle, enamel, principle is simply easy
Row, can effectively identify insulator Exception Type, and one kind is provided certainly for electric transmission line isolator anomaly analysis in unmanned plane inspection
Dynamicization detects thinking and method.
Claims (9)
1. a kind of insulator exception automatic testing method based on integrated classifier on-line study, which is characterized in that specifically according to
Following steps are implemented：
Step 1, it acquires insulation subgraph and obtains insulation subsample D to be measured, specifically, unmanned plane or shaft tower is utilized to carry high definition
Monopod video camera acquisition insulation subgraph determines its best shooting focal length wherein by constantly regulate video camera shooting angle, from
And ensures the image collected and be more easy to extract the effective characteristic parameter of variety classes insulator；
Step 2, training sample set is established, specifically, determining the insulator sample set of fault type as off-line training using oneself
Sample set trains grader, the Exception Type of coded samples insulator, by normal, self-destruction, crackle, enamel burn out and other five
Quasi-mode is separately encoded as Arabic numerals 1,2,3,4 and 5, as the class label corresponding to attribute value in the sample set；
Step 3, to the training insulator of collected insulation subsample D to be measured in step 1 and the determination Exception Type in step 2
Sample T is pre-processed, i.e., by collected insulation subgraph after gray processing is handled, with contrast adaptive equalization
Then method isolates target insulation subregion to image enhancement using adaptive threshold fuzziness method, by new two of acquisition
Value image is mapped on original sample, is extracted the insulation subgraph accurately divided, is obtained sample to be tested D ' and training sample T '；
Step 4, feature is carried out to the image being partitioned into step 3 in the sample to be tested D ' and training sample T ' of insulation subregion to carry
It takes, acquisition insulator essential characteristic fusion similarity measure values dHWD, insulator distance between commutator segments Δ L, insulator marginal information IRDAnd
Its rate of gray level ▽ graysc_y(x) etc. features make it to construct training sample set as sample attribute value after normalized
S and test sample collection P；
Step 5, using decision tree as Weak Classifier, circuit training is carried out to the decision tree by the training sample set S,
Until obtaining Weak Classifier anticipation function htAnd its error rate,
Step 6, the Weak Classifier gone out according to off-line training in step 5 are pre- using the final strong classifier of ballot method generation of Weight
Survey function h0 fin(z)；
That is, all Weak Classifiers is allowed to vote, then voting results are obtained according to error rates of weak classifiers weighted sum final
Strong classifier anticipation function, as shown in formula (7)
Step 7, test sample collection P described in step 4 is sequentially input in step 6 gained strong classifier, through described final strong point
After class device function judges identification, the class label result v of corresponding exception is obtainedfin；
Step 8, by step 7 output testing result vfinThe decision tree classifier classification results v minimum with error rate in step 5min
Comparing calculation is carried out by formula (8),
That is, working as vfinAnd vminThe class label output insulator Exception Type V is matched when the two is consistentfin, complete diagnosis；
Otherwise, work as vfinAnd vminWhen the two is inconsistent, step 9 is executed；
Step 9, when judging in step 8, there is vfin≠vmin, that is, there are strong classifier classification results obtained by off-line training and determine
When plan tree result difference, grader is made to carry out on-line study process, timely update grader, and the detection carried out next time is sentenced
It is disconnected.
2. the insulator exception automatic testing method according to claim 1 based on integrated classifier on-line study, special
Sign is, in step 4, in the training sample set S, including 5 class insulator breakdowns, there are m group numbers per class insulator breakdown
According to then training sample set S is expressed as：
S=<(u11,v11),…,(u1m,v1m),…,(u5m,v5m)>,
Wherein, it is ω that i-th of training sample, which has equal initial weight,i=1/5m；The value range of i is [1, m].
3. the insulator exception automatic testing method according to claim 1 based on integrated classifier on-line study, special
Sign is, each feature normalization calculating process in the step 4 specifically,
Step 4.1, to essential characteristic fusion similarity measure values dHWDIt is calculated,
Step 4.2, to insulator distance between commutator segments Δ l, this feature is normalized,
Step 4.3, marginal information IRDThis feature is normalized,
Step 4.4, to rate of gray level ▽ graysc_y(x) this feature is normalized.
4. the insulator exception automatic testing method according to claim 3 based on integrated classifier on-line study, special
Sign is, the step 4.1 specifically,
Step 4.1.1, calculation formula (1) the progress essential characteristic that similarity measure values are merged using normalized essential characteristic are melted
Close similarity measure values dHWDCalculating,
Step 4.1.2 measures single features using histogram intersection method, obtains the measuring similarity value of each feature,
Histogram intersection method formula is：
Wherein, M and N is respectively the histogram of original normal insulation subgraph p and input sample q same dimensions, and d (M, N) is indicated
The normalization of histogram intersection is as a result, its value is closer to 1, then two Histogram Matchings is better；If image p and image q return
It is 1 that one change histogram, which exactly matches then histogram intersection institute value, and complete mismatch value is 0.
5. the insulator exception automatic testing method according to claim 3 based on integrated classifier on-line study, special
Sign is that the step 4.2 obtains insulation sub-connected domain specifically, the image concentrated to training sample is split processing, obtains
It obtains each connected domain center and indicates sub-pieces center, acquire the distance between adjacent sub-pieces of normalization：
Wherein, Δ l indicates adjacent insulator distance between commutator segments,L indicates insulator chain overall length
Degree；(X1,Y1),(X2,Y2) be respectively two neighboring connected domain center point coordinate.
6. the insulator exception automatic testing method according to claim 3 based on integrated classifier on-line study, special
Sign is, the step 4.3 specifically,
Step 4.3.1 carries out piecemeal to image first with adaptive multi-thresholding algorithm, and determines that the best of segmented areas is getted over
Point；Then use linear membership function by image pixel values transformation for fuzzy field, fuzzy pair proposed by getting over calculating
The relevant parameter of bent tangent function carries out nonlinear images enhancing to each segmented areas application Retinex algorithm of image；
Step 4.3.2 calls the binary segmentation that Canny operators carry out image to obtain the potential area edge of local anomaly, is carried out to it
The maximum connected domain of Retention area, length after dilation operation calculates two features of its circularity R and length-width ratio D, as circularity R
Less than R0And length-width ratio D is more than D0When judgement artwork in there are crackle, circularity R calculates as follows:Wherein s is
Connected domain area, C are connected domain perimeter；When region is round, R values are up to 1；
Length-width ratio D calculates as follows:Wherein, L is the length of connected domain, that is, surrounds the length of the minimum rectangle of connected domain；
Obtained normalization marginal information is denoted as：IRD=(R≤R0)&&(L≤L0)。
7. the insulator exception automatic testing method according to claim 3 based on integrated classifier on-line study, special
Sign is, the step 4.4 specifically,
Step 4.4.1, to handling obtained marginal information I through step 4.3RDObtained image img after screening is weighted flat
Equal gray processing, and average gray is weighted to image img by formula (3), obtain weighted average gray level image
Grays, specific computational methods are as follows：
The gray value after weighted average gray processing at coordinate points (x, y) is calculated using formula (3),
Grays (x, y)=(WrR(x,y)+WgG(x,y)+WbB(x,y))/3 (3)
Wherein, Wr=0.299, Wg=0.587, Wb=0.114, grays (x, y) indicate weighted average gray processing after coordinate points (x,
Y) gray value at place,
Step 4.4.2, using step 4.4.1 calculating as a result, formula (4) is pushed away to obtain by formula (3), according to formula (4) calculating ash
Longitudinal gray value of degreeization image：
graysc_y(x)=grays (x, y=c) 1≤c≤Wcolumn (4)
Wherein, c indicates the row coordinate of longitudinal hatching line position, WcolumnIndicate the width of row；
Obtain ▽ graysc_y(x), that is, normalized longitudinal gray-value variation rate is indicated.
8. the insulator exception automatic testing method according to claim 1 based on integrated classifier on-line study, special
Sign is that the specific step of the step 5 is：
Step 5.1, by normalized image feature I { d (M, N), Il,IRD,▽graysc_y(x) } weak point is inputted as sample attribute
Class device, Weak Classifier algorithm by the weights based on training sample set before larger 80% carry out structure grader operation, generate
Corresponding decision tree ht:U → V,
Step 5.2, assess single classifier performance, in total training sample other than screening the sample for being used as training all samples
This is all by the test sample as grader caused by assessment epicycle iteration；The error rate h of the base gradert:erriBy formula
(5) it can be calculated：
Wherein, ωt(i) weight of i-th of sample in t wheels is indicated；
Step 5.3, weight being adjusted according to error rate, promotes the weight of misjudgement sample, specific method is,
It enablesThen,
Wherein, αtThe Weak Classifier h generated for t wheelstEvaluation points, classification accuracy rate is higher, value it is bigger, it is on the contrary then smaller；
ZtIt is a normalization constant；ωt+1(i) it is the weight of i-th group of data of t+1 wheels；
The weight for distinguishing correct sample is reduced with this, will more consider that these are misjudged in this way in training later
Sample；
On the basis of sample weights after the adjustment, repeats step 5.1 and 5.2 and be trained the new grader of generation again, i.e., it is several
A general Weak Classifier of detectability.
9. the insulator exception automatic testing method according to claim 1 based on integrated classifier on-line study, special
Sign is that the step 9 is specially：
Step 9.1, it will be added in training sample set S centered on the inconsistent sample of step 8 acquired results, utilize kNN neighbours
Thought, the essential characteristic fusion similarity measure values that sample image is obtained according to the method for step 4.1 find out it away from minimum k (k
=m) the maximum a kind of reference category as the sample of quantity in a sample, it is matched with above two result, takes probability
High result as the sample classification and using the sample as on-line training sample replace original training sample collection correspond to classification
A minimum sample of middle weight, to constitute new training sample set S '；
Step 9.2, return to step 5, the weight of training sample after initialization update, according to the setting training time of off-line training empirical value
Number trains Weak Classifier, counts its number correctly or incorrectly classified, simplified Weak Classifier error rate is found out, with this again
Update sample weights；
Step 9.3, according to step 6 strong classifier generating process, updated strong classifier h is obtainedn fin(x), wherein n is represented more
New number, takes 0,1,2 ....
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810386292.5A CN108765373B (en) | 2018-04-26 | 2018-04-26 | Insulator abnormity automatic detection method based on integrated classifier online learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810386292.5A CN108765373B (en) | 2018-04-26 | 2018-04-26 | Insulator abnormity automatic detection method based on integrated classifier online learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108765373A true CN108765373A (en) | 2018-11-06 |
| CN108765373B CN108765373B (en) | 2022-03-22 |
Family
ID=64011895
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810386292.5A Active CN108765373B (en) | 2018-04-26 | 2018-04-26 | Insulator abnormity automatic detection method based on integrated classifier online learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108765373B (en) |
Cited By (23)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109489977A (en) * | 2018-12-28 | 2019-03-19 | 西安工程大学 | Method for Bearing Fault Diagnosis based on KNN-AdaBoost |
| CN109583904A (en) * | 2018-11-30 | 2019-04-05 | 深圳市腾讯计算机系统有限公司 | Training method, impaired operation detection method and the device of abnormal operation detection model |
| CN110110749A (en) * | 2019-03-29 | 2019-08-09 | 广州思德医疗科技有限公司 | Image processing method and device in a kind of training set |
| CN110503215A (en) * | 2019-08-28 | 2019-11-26 | 国网天津市电力公司 | Method based on image recognition technology identification equipment state, failure and processing exception |
| CN110598737A (en) * | 2019-08-06 | 2019-12-20 | 深圳大学 | Online learning method, device, equipment and medium of deep learning model |
| CN110889435A (en) * | 2019-11-04 | 2020-03-17 | 国网河北省电力有限公司检修分公司 | Insulator evaluation classification method and device based on infrared image |
| CN111242923A (en) * | 2020-01-13 | 2020-06-05 | 国网湖北省电力有限公司宜昌供电公司 | Canny operator-based insulator crack detection method based on maximum entropy threshold optimization |
| CN111325224A (en) * | 2018-12-13 | 2020-06-23 | 数优(苏州)人工智能科技有限公司 | Computer-readable storage medium, input data checking method, and computing device |
| CN111444857A (en) * | 2020-03-30 | 2020-07-24 | 国网河北省电力有限公司沧州供电分公司 | Unmanned aerial vehicle inspection data processing method |
| CN111624931A (en) * | 2020-06-18 | 2020-09-04 | 山东山大世纪科技有限公司 | Industrial park electricity utilization internet intelligent operation and maintenance management and control system and method |
| CN111696060A (en) * | 2020-06-08 | 2020-09-22 | 国网新疆电力有限公司乌鲁木齐供电公司 | Method and system for online monitoring hydrophobicity of composite insulator |
| CN111799697A (en) * | 2020-07-08 | 2020-10-20 | 云南电网有限责任公司电力科学研究院 | Self-explosion simulation method and system for glass insulator of power transmission line |
| CN112598234A (en) * | 2020-12-14 | 2021-04-02 | 广东电网有限责任公司广州供电局 | Low-voltage transformer area line loss abnormity analysis method, device and equipment |
| CN113252695A (en) * | 2021-06-16 | 2021-08-13 | 上海启迪睿视智能科技有限公司 | Plastic packaging film defect detection method and detection device based on image processing |
| CN113537353A (en) * | 2021-07-19 | 2021-10-22 | 天津工业大学 | Small sample learning method for insulator image data |
| CN113591873A (en) * | 2021-05-26 | 2021-11-02 | 东南大学 | Flame image classification method based on ensemble learning |
| CN113657460A (en) * | 2021-07-28 | 2021-11-16 | 上海影谱科技有限公司 | Boosting-based attribute identification method and device |
| CN113935440A (en) * | 2021-12-15 | 2022-01-14 | 武汉格蓝若智能技术有限公司 | Iterative evaluation method and system for error state of voltage transformer |
| CN114049507A (en) * | 2021-11-19 | 2022-02-15 | 国网湖南省电力有限公司 | Distribution network line insulator defect identification method, equipment and medium based on twin network |
| CN114758257A (en) * | 2022-04-25 | 2022-07-15 | 汇德晟新电气科技(扬州)有限公司 | CART algorithm-based power transmission line degraded insulator identification method |
| CN114792299A (en) * | 2021-01-25 | 2022-07-26 | 山东信通电子股份有限公司 | Method and device for detecting abnormal state of power transmission line |
| CN117934481A (en) * | 2024-03-25 | 2024-04-26 | 国网浙江省电力有限公司宁波供电公司 | Power transmission cable state identification processing method and system based on artificial intelligence |
| CN117934458A (en) * | 2024-03-21 | 2024-04-26 | 武汉市豪迈电力自动化技术有限责任公司 | Smart grid safety protection method based on short shared power wireless communication |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8391592B2 (en) * | 2006-06-13 | 2013-03-05 | Sri International | System and method for detection of multi-view/multi-pose objects |
| CN106127756A (en) * | 2016-06-21 | 2016-11-16 | 西安工程大学 | A kind of insulator recognition detection method based on multicharacteristic information integration technology |
| CN106250920A (en) * | 2016-07-26 | 2016-12-21 | 国网福建省电力有限公司 | The insulator state detection merged based on multicharacteristic information and diagnostic method |
| CN106778650A (en) * | 2016-12-26 | 2017-05-31 | 深圳极视角科技有限公司 | Scene adaptive pedestrian detection method and system based on polymorphic type information fusion |
-
2018
- 2018-04-26 CN CN201810386292.5A patent/CN108765373B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8391592B2 (en) * | 2006-06-13 | 2013-03-05 | Sri International | System and method for detection of multi-view/multi-pose objects |
| CN106127756A (en) * | 2016-06-21 | 2016-11-16 | 西安工程大学 | A kind of insulator recognition detection method based on multicharacteristic information integration technology |
| CN106250920A (en) * | 2016-07-26 | 2016-12-21 | 国网福建省电力有限公司 | The insulator state detection merged based on multicharacteristic information and diagnostic method |
| CN106778650A (en) * | 2016-12-26 | 2017-05-31 | 深圳极视角科技有限公司 | Scene adaptive pedestrian detection method and system based on polymorphic type information fusion |
Non-Patent Citations (2)
| Title |
|---|
| TOUFIQ PARAG ET AL.: "Boosting adaptive linear weak classifiers for online learning and tracking", 《2008 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION》 * |
| 苗向鹏: "基于图像处理的接触网绝缘子识别与破损检测", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (29)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109583904B (en) * | 2018-11-30 | 2023-04-07 | 深圳市腾讯计算机系统有限公司 | Training method of abnormal operation detection model, abnormal operation detection method and device |
| CN109583904A (en) * | 2018-11-30 | 2019-04-05 | 深圳市腾讯计算机系统有限公司 | Training method, impaired operation detection method and the device of abnormal operation detection model |
| CN111325224A (en) * | 2018-12-13 | 2020-06-23 | 数优(苏州)人工智能科技有限公司 | Computer-readable storage medium, input data checking method, and computing device |
| CN109489977A (en) * | 2018-12-28 | 2019-03-19 | 西安工程大学 | Method for Bearing Fault Diagnosis based on KNN-AdaBoost |
| CN110110749A (en) * | 2019-03-29 | 2019-08-09 | 广州思德医疗科技有限公司 | Image processing method and device in a kind of training set |
| CN110598737A (en) * | 2019-08-06 | 2019-12-20 | 深圳大学 | Online learning method, device, equipment and medium of deep learning model |
| CN110598737B (en) * | 2019-08-06 | 2023-02-24 | 深圳大学 | Online learning method, device, equipment and medium of deep learning model |
| CN110503215A (en) * | 2019-08-28 | 2019-11-26 | 国网天津市电力公司 | Method based on image recognition technology identification equipment state, failure and processing exception |
| CN110889435A (en) * | 2019-11-04 | 2020-03-17 | 国网河北省电力有限公司检修分公司 | Insulator evaluation classification method and device based on infrared image |
| CN111242923A (en) * | 2020-01-13 | 2020-06-05 | 国网湖北省电力有限公司宜昌供电公司 | Canny operator-based insulator crack detection method based on maximum entropy threshold optimization |
| CN111444857A (en) * | 2020-03-30 | 2020-07-24 | 国网河北省电力有限公司沧州供电分公司 | Unmanned aerial vehicle inspection data processing method |
| CN111444857B (en) * | 2020-03-30 | 2022-11-15 | 国网河北省电力有限公司沧州供电分公司 | Unmanned aerial vehicle inspection data processing method |
| CN111696060A (en) * | 2020-06-08 | 2020-09-22 | 国网新疆电力有限公司乌鲁木齐供电公司 | Method and system for online monitoring hydrophobicity of composite insulator |
| CN111624931A (en) * | 2020-06-18 | 2020-09-04 | 山东山大世纪科技有限公司 | Industrial park electricity utilization internet intelligent operation and maintenance management and control system and method |
| CN111799697B (en) * | 2020-07-08 | 2021-09-03 | 云南电网有限责任公司电力科学研究院 | Self-explosion simulation method and system for glass insulator of power transmission line |
| CN111799697A (en) * | 2020-07-08 | 2020-10-20 | 云南电网有限责任公司电力科学研究院 | Self-explosion simulation method and system for glass insulator of power transmission line |
| CN112598234A (en) * | 2020-12-14 | 2021-04-02 | 广东电网有限责任公司广州供电局 | Low-voltage transformer area line loss abnormity analysis method, device and equipment |
| CN114792299A (en) * | 2021-01-25 | 2022-07-26 | 山东信通电子股份有限公司 | Method and device for detecting abnormal state of power transmission line |
| CN113591873A (en) * | 2021-05-26 | 2021-11-02 | 东南大学 | Flame image classification method based on ensemble learning |
| CN113252695B (en) * | 2021-06-16 | 2021-09-28 | 上海启迪睿视智能科技有限公司 | Plastic packaging film defect detection method and detection device based on image processing |
| CN113252695A (en) * | 2021-06-16 | 2021-08-13 | 上海启迪睿视智能科技有限公司 | Plastic packaging film defect detection method and detection device based on image processing |
| CN113537353B (en) * | 2021-07-19 | 2022-09-20 | 天津工业大学 | Small sample learning method for insulator image data |
| CN113537353A (en) * | 2021-07-19 | 2021-10-22 | 天津工业大学 | Small sample learning method for insulator image data |
| CN113657460A (en) * | 2021-07-28 | 2021-11-16 | 上海影谱科技有限公司 | Boosting-based attribute identification method and device |
| CN114049507A (en) * | 2021-11-19 | 2022-02-15 | 国网湖南省电力有限公司 | Distribution network line insulator defect identification method, equipment and medium based on twin network |
| CN113935440A (en) * | 2021-12-15 | 2022-01-14 | 武汉格蓝若智能技术有限公司 | Iterative evaluation method and system for error state of voltage transformer |
| CN114758257A (en) * | 2022-04-25 | 2022-07-15 | 汇德晟新电气科技(扬州)有限公司 | CART algorithm-based power transmission line degraded insulator identification method |
| CN117934458A (en) * | 2024-03-21 | 2024-04-26 | 武汉市豪迈电力自动化技术有限责任公司 | Smart grid safety protection method based on short shared power wireless communication |
| CN117934481A (en) * | 2024-03-25 | 2024-04-26 | 国网浙江省电力有限公司宁波供电公司 | Power transmission cable state identification processing method and system based on artificial intelligence |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108765373B (en) | 2022-03-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108765373A (en) | A kind of insulator exception automatic testing method based on integrated classifier on-line study | |
| Davari et al. | Intelligent diagnosis of incipient fault in power distribution lines based on corona detection in UV-visible videos | |
| CN111242123B (en) | Power equipment fault diagnosis method based on infrared image | |
| Liu et al. | Automated visual inspection system for bogie block key under complex freight train environment | |
| CN111784633B (en) | Insulator defect automatic detection algorithm for electric power inspection video | |
| Ibrahim et al. | Application of machine learning to evaluate insulator surface erosion | |
| CN113344475B (en) | Transformer bushing defect identification method and system based on sequence modal decomposition | |
| CN108537170A (en) | A kind of power equipment firmware unmanned plane inspection pin missing detection method | |
| CN111080691A (en) | Infrared hot spot detection method and device for photovoltaic module | |
| KR101797400B1 (en) | Method and apparatus for diagnosing fault based on probabilistic density | |
| CN106951863B (en) | Method for detecting change of infrared image of substation equipment based on random forest | |
| CN103824092A (en) | Image classification method for monitoring state of electric transmission and transformation equipment on line | |
| CN113065484A (en) | Insulator contamination state assessment method based on ultraviolet spectrum | |
| CN114724042B (en) | Automatic detection method for zero-value insulator in power transmission line | |
| Huang et al. | Study on hydrophobicity detection of composite insulators of transmission lines by image analysis | |
| Zhang et al. | Detection and condition assessment of icicle bridging for suspension glass insulator by image analysis | |
| CN113298178A (en) | Transformer substation high-voltage equipment fault identification method based on thermal infrared image | |
| CN115994907A (en) | Intelligent processing system and method for comprehensive information of food detection mechanism | |
| Tomaszewski et al. | Detection of power line insulators on digital images with the use of laser spots | |
| Chatterjee et al. | Recognition of hydrophobicity class of polymeric insulators employing residual morphological neural network and granulometry-based image analysis | |
| Iruansi et al. | Classification of power-line insulator condition using local binary patterns with support vector machines | |
| CN111325708B (en) | Transmission line detection method and server | |
| CN112132226A (en) | Method for diagnosing defects of +/-1100 kV composite insulator | |
| Li et al. | Detection of small size defects in belt layer of radial tire based on improved faster r-cnn | |
| Taib et al. | Thermal imaging for qualitative-based measurements of thermal anomalies in electrical components |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |