CN109615611A - A kind of insulator self-destruction defect inspection method based on inspection image - Google Patents
A kind of insulator self-destruction defect inspection method based on inspection image Download PDFInfo
- Publication number
- CN109615611A CN109615611A CN201811376353.6A CN201811376353A CN109615611A CN 109615611 A CN109615611 A CN 109615611A CN 201811376353 A CN201811376353 A CN 201811376353A CN 109615611 A CN109615611 A CN 109615611A
- Authority
- CN
- China
- Prior art keywords
- insulator
- defect
- self
- destruction
- inspection image
- 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
- 239000012212 insulator Substances 0.000 title claims abstract description 238
- 230000007547 defect Effects 0.000 title claims abstract description 212
- 238000007689 inspection Methods 0.000 title claims abstract description 188
- 238000000034 method Methods 0.000 title claims abstract description 58
- 238000001514 detection method Methods 0.000 claims abstract description 69
- 239000011521 glass Substances 0.000 claims abstract description 64
- 238000012549 training Methods 0.000 claims abstract description 53
- 238000001914 filtration Methods 0.000 claims abstract description 39
- 238000013135 deep learning Methods 0.000 claims abstract description 34
- 230000002708 enhancing effect Effects 0.000 claims abstract description 20
- 238000012545 processing Methods 0.000 claims abstract description 18
- 238000009413 insulation Methods 0.000 claims description 35
- 230000002950 deficient Effects 0.000 claims description 13
- 238000012360 testing method Methods 0.000 claims description 13
- 238000009877 rendering Methods 0.000 claims description 9
- 230000000694 effects Effects 0.000 claims description 7
- 239000000463 material Substances 0.000 claims description 4
- 238000005259 measurement Methods 0.000 claims description 4
- 238000004088 simulation Methods 0.000 claims description 4
- 230000000007 visual effect Effects 0.000 claims description 3
- 238000004519 manufacturing process Methods 0.000 abstract description 5
- 238000013527 convolutional neural network Methods 0.000 description 24
- 239000013598 vector Substances 0.000 description 9
- 238000000605 extraction Methods 0.000 description 6
- 230000006870 function Effects 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 4
- PCHJSUWPFVWCPO-UHFFFAOYSA-N gold Chemical compound [Au] PCHJSUWPFVWCPO-UHFFFAOYSA-N 0.000 description 4
- 239000010931 gold Substances 0.000 description 4
- 229910052737 gold Inorganic materials 0.000 description 4
- 238000005070 sampling Methods 0.000 description 4
- 238000004422 calculation algorithm Methods 0.000 description 3
- 238000012937 correction Methods 0.000 description 3
- 230000005540 biological transmission Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 235000013399 edible fruits Nutrition 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 238000012423 maintenance Methods 0.000 description 2
- 230000000306 recurrent effect Effects 0.000 description 2
- 201000004569 Blindness Diseases 0.000 description 1
- 238000012952 Resampling Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000003708 edge detection Methods 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
- 230000003313 weakening effect Effects 0.000 description 1
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
- G06T7/0006—Industrial image inspection using a design-rule based approach
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
-
- 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
- G06T2207/20024—Filtering details
- G06T2207/20032—Median filtering
-
- 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
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
- G06T2207/30164—Workpiece; Machine component
-
- 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
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Investigating Materials By The Use Of Optical Means Adapted For Particular Applications (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The present invention provides a kind of insulator self-destruction defect inspection method based on inspection image, comprising: obtain the shooting of power-line patrolling unmanned plane includes the inspection image to be detected of glass insulator and manually mark a small amount of insulator self-destruction defect;Mass production includes the virtual inspection image of defects of insulator in analogue system;Enhancing processing is filtered to true and virtual image, obtains filtering enhanced inspection image association training set;Faster R-CNN deep learning network, which is based on, using joint training collection training obtains insulator self-destruction defects detection model;Whether detection model can detect the enhanced inspection image to be detected of filtering, include in the inspection image to be detected of glass insulator described in determination including self-destruction defect.Method provided by the invention solves the problems, such as big data training set lazy weight in previous deep learning training;Improve the accuracy rate of insulator self-destruction defects detection.
Description
Technical field
The present invention relates to power grid O&M technical fields, and more particularly, to a kind of insulator based on inspection image
Reveal defect inspection method.
Background technique
Glass insulator is widely used in grid operation, is the emphasis shooting unit of unmanned plane inspection.Glass insulator
Concentration self-destruction be likely to result in line tripping fault, jeopardize route stable operation.
Power-line patrolling unmanned plane has been widely applied in recent years, produces magnanimity inspection image.By manually come interpretation
Inspection image simultaneously detects insulator self-destruction defect, and low efficiency, subjectivity are strong, consistency is poor, significantly limit maintenance work effect
Rate.And the accuracy and omission factor of the method for being detected insulator self-destruction defect automatically based on inspection image existing at present all can not
Meet application demand.
Summary of the invention
The present invention provides a kind of, and the insulator based on inspection image reveals defect inspection method, to solve current insulator
The problem of accuracy and omission factor for revealing defect inspection method are all unable to satisfy application demand.
The insulator of inspection image provided by the invention reveals defect inspection method, comprising:
Obtain the shooting of power-line patrolling unmanned plane include several true inspection images of glass insulator defect and it is artificial
It marks insulator and reveals defect;
It generates virtual inspection image and automatic marking insulator comprising insulator self-destruction defect and reveals defect;
Enhancing processing is filtered to true inspection image, virtual inspection image, obtains filtering enhanced true inspection
Image and virtual inspection image;
According to preset quantity true inspection image more enhanced than combined filter and virtual inspection image, connection is constituted
Close training set;
It is based on Faster R-CNN deep learning network using joint training collection training, determines that insulator reveals defects detection
Model, insulator are revealed the self-destruction belonging to insulator in the true inspection image of determination or virtual inspection shadow of defects detection model and are lacked
When falling into type, it is greater than preset detection threshold value for the confidence level of corresponding self-destruction defect type.
Further, self-destruction defect type include: monolithic self-destruction in glass insulator strings, it is continuous more in glass insulator strings
The self-destruction of fitting connecting pin monolithic and the self-destruction of glass insulator strings fitting connecting pin multi-disc in piece self-destruction, glass insulator strings.
Further, insulator reveals defect inspection method, further includes:
Obtain the shooting of power-line patrolling unmanned plane includes the inspection image to be detected of glass insulator;
Enhancing processing is filtered to inspection image to be detected, obtains filtering enhanced inspection image to be detected;
Defects detection model inspection is revealed using insulator, enhanced inspection image to be detected is filtered, includes with determination
Whether have in the inspection image to be detected of glass insulator includes self-destruction defect.
Further, it is filtered enhancing processing, comprising:
1) inspection image, the inspection image after obtaining median filtering are traversed using four neighborhood median filterings;
2) after using the Sobel boundary operator traversal median filtering with 0 °, 45 °, 90 ° and 135 ° this 4 direction templates
Inspection image, obtain edge inspection image outstanding;
3) four neighborhood intensity contrast methods, to be detected inspection image de-noising outstanding to edge, after obtaining filtering enhancing are utilized
Inspection image.
Further, to be detected inspection image de-noising of the edge after prominent is filtered using four neighborhood intensity contrast methods
The enhanced inspection image to be detected of wave, comprising:
To any pixel point f (x, y), if | f (x, y)-f (x+i, y+j) | > TH, for any i=-1,1;J=-1,1
It sets up, then pixel is noise spot;Wherein, threshold value TH is the constant not less than 20;
The gray value that will determine as the pixel of noise spot is appointed as 255.
Further, virtual inspection image and the self-destruction of automatic marking insulator comprising insulator self-destruction defect is generated to lack
It falls into, comprising:
1) there is the glass insulator and normal glass insulation of different self-destruction defect types using laser scanner measurement
The material object of son obtains three-dimensional geometry data in kind;
2) according to three-dimensional geometry data, it is directed to the sub-pieces of different self-destruction defect types respectively, normal insulation sub-pieces, connects
Fitting establishes three dimensional virtual models, and rendering obtains defective insulation sub-image, normal insulation sub-image is connected with fitting
The image of part;
3) according to preset rule, insulator chain virtual image is generated, in the not same district of insulator chain virtual image
At least one defective insulator is randomly distributed in domain;
4) it takes pictures to the insulator chain virtual image of generation from the simulation of multiple and different angles, the result that will take pictures rendering output is
Picture images format, and record the coordinate letter of the pixel of defective insulator and the sub- region of neighbouring normal insulation
Breath.
Further, manually mark insulator self-destruction defect includes:
Self-destruction in the true inspection image for the glass insulator with different self-destruction defect types that artificial mark obtains
Region where defect area and its neighbouring normal insulation sub-pieces or fitting connecting pin.
Further,
Automatic marking insulator reveals defect
It determines in the virtual mark shadow of the glass insulator with different self-destruction defect types under multiple and different shooting visual angles certainly
The coordinate information of the pixel in the region where quick-fried disordered insulator piece and its neighbouring normal insulation sub-pieces or fitting connecting pin.
Further, further includes:
It include that fitting is had into self-destruction after revealing defect in the inspection image to be detected that determination includes glass insulator
The sub-pieces of defect and its both ends it is neighbouring normal insulation where pixel region minimum circumscribed rectangle be superimposed upon it is to be detected
On inspection image.
Further, it is based on Faster R-CNN deep learning network using joint training collection training, determines insulator certainly
When quick-fried defects detection model,
If being greater than preset detection threshold value for the confidence level of corresponding self-destruction defect type, it is determined that testing result
Effectively, and in inspection image with rectangle frame label insulator the region where defect is revealed;
If being not more than preset detection threshold value for the confidence level of corresponding self-destruction defect type, it is determined that detection knot
Fruit is invalid.
The insulator of inspection image provided by the invention reveals defect inspection method, enhances prominent virtual image using filtering
With the fitness of target geometrical characteristic and feature in real image.Using joint training collection training Faster R-CNN mould
Type solves the problems, such as big data training set lazy weight in previous deep learning training, while reducing extensive artificial mark
The cost of labor of defect target in inspection image.
Detailed description of the invention
By reference to the following drawings, exemplary embodiments of the present invention can be more fully understood by:
Fig. 1 is that the process of the insulator self-destruction defect inspection method based on inspection image of the preferred embodiment for the present invention is shown
It is intended to;
Fig. 2 is that filtering increases in the insulator self-destruction defect inspection method based on inspection image of the preferred embodiment for the present invention
The flow diagram of strong step;
Fig. 3 is that the process of the insulator self-destruction defect inspection method in the preferred embodiment for the present invention based on inspection image is shown
It is intended to.
Specific embodiment
Exemplary embodiments of the present invention are introduced referring now to the drawings, however, the present invention can use many different shapes
Formula is implemented, and is not limited to the embodiment described herein, and to provide these embodiments be at large and fully disclose
The present invention, and the scope of the present invention is sufficiently conveyed to person of ordinary skill in the field.Show for what is be illustrated in the accompanying drawings
Term in example property embodiment is not limitation of the invention.In the accompanying drawings, identical cells/elements use identical attached
Icon note.
Unless otherwise indicated, term (including scientific and technical terminology) used herein has person of ordinary skill in the field
It is common to understand meaning.Further it will be understood that with the term that usually used dictionary limits, should be understood as and its
The context of related fields has consistent meaning, and is not construed as Utopian or too formal meaning.
When carrying out power equipments defect detection based on inspection image, there are two class problems to fail the solution party for having complete always
Case, first is that the missing problem for the inspection device defect image data collection that each platform obtains, second is that insulating from similar normal condition
The self-destruction defective locations of insulator are accurately positioned out in subassembly.Insulator is detected from the inspection image that electric power unmanned plane obtains
Self-destruction defective locations have both the above two classes problem.
Therefore, the glass insulator self-destruction detection based on inspection image is unfolded using depth learning technology has important meaning
Justice.Trained deep learning network is necessarily involved in depth learning technology.In training deep learning network, need largely to pass through
Training set and test set of the insulator self-destruction defect image master sample of mark as deep learning network.
Reveal defect image master sample to obtain the sufficient insulator by mark of quantity, it is necessary to by electric power without
The man-machine self-destruction defect standard image for going to obtain all types of insulators under different shooting angles, and by manually being lacked to these self-destructions
It falls into standard video and carries out information labeling.This two times to work and human cost are all very high.
Another needs the fact that pay close attention to be, although the resolution for the power equipment inspection image that current electric power unmanned plane obtains
Rate is higher and higher, but is limited by shooting distance, and insulator reveals region model of the defect image in whole inspection image
Enclose smaller, characteristic information is easy to be submerged in the environment of surrounding, and is highly prone to the interference of neighbouring normal insulation, therefore,
Even manual work, the insulation sub-goal that self-destruction defect is accurately positioned from inspection image is also highly difficult.
Further, since the incidence of glass insulator self-destruction defect is relatively low in electric system, the glass based on self-destruction defect
It is especially tired will to complete inspection image of the multi-angle high-volume acquisition comprising insulator self-destruction defect for the small number of glass insulator
Difficulty, and this quantity that will lead to target device defect sample is difficult to meet deep learning network training and the data of test needs are wanted
It asks.The lazy weight of defect sample causes the detection model generated over-fitting occur, and causes recognition effect unstable.
Therefore, the depth learning technology to behave oneself best for object detection field, due to by existing neural network capacity
Limitation, for the inspection image of original resolution after process of convolution and down-sampled layer, the lesser insulator of size reveals mesh
Rotating savings loses characteristic information, is more difficult to.
The present invention provides a kind of insulator self-destruction defect inspection method based on inspection image, comprising: obtains power-line patrolling
Unmanned plane shooting includes the inspection image to be detected of glass insulator and manually marks a small amount of insulator self-destruction defect;Imitative
Mass production includes the virtual inspection image of defects of insulator in true system;True and virtual image is filtered at enhancing
Reason, obtains filtering enhanced inspection image association training set;Faster R-CNN depth is based on using joint training collection training
Learning network obtains insulator and reveals defects detection model;Detection model can be to the enhanced inspection image to be detected of filtering
It is detected, whether includes in the inspection image to be detected of glass insulator described in determination including self-destruction defect.
The invention proposes " generate virtual sample-target marks automatically-and form joint training collection-filtering enhancing pretreatment-
This complete technology path of training deep learning detection model ".Method provided by the invention is virtually patrolled using Mass production
Image is examined, solves the problems, such as big data training set lazy weight in previous deep learning training, while being reduced extensive artificial
Mark the cost of labor of defect target in inspection image;It is several that target in prominent virtual image and real image is enhanced using filtering
The fitness of what feature and feature;Using joint training collection training Faster R-CNN model, improves insulator self-destruction and lack
Fall into the accuracy rate of detection.
As shown in Figure 1, the insulator of the inspection image of the embodiment of the present invention reveals defect inspection method, comprising:
S10: obtain power-line patrolling unmanned plane shooting include several true inspection images of glass insulator defect simultaneously
Artificial mark insulator reveals defect;
S20: virtual inspection image and automatic marking insulator comprising insulator self-destruction defect are generated and reveals defect;
S30: being filtered enhancing processing to true inspection image, virtual inspection image, obtains filtering enhanced true
Inspection image and virtual inspection image;
S40: according to preset quantity true inspection image more enhanced than combined filter and virtual inspection image, structure
At joint training collection;
S50: being based on Faster R-CNN deep learning network using the training of joint training collection, determines that insulator reveals defect
Detection model, insulator reveal defects detection model in the true inspection image of determination or virtual inspection shadow belonging to insulator from
When quick-fried defect type, it is greater than preset detection threshold value for the confidence level of corresponding self-destruction defect type.
It is construed as, what power-line patrolling unmanned plane was shot includes several true inspection images of glass insulator defect
In, it include various typical defect types.It is construed as, insulator reveals defects detection model in the true inspection shadow of determination
In picture or virtual inspection shadow when self-destruction defect type belonging to insulator, it is greater than for the confidence level of corresponding self-destruction defect type
Preset detection threshold value;If confidence level is not more than preset detection threshold value, further to self-destruction defect type
Structural parameters or weight parameter are adjusted, until all meeting detection for all inspection images with typical defect type
Confidence level be greater than preset detection threshold value.
When it is implemented, the insulator reveals defect inspection method, further includes:
Obtain the shooting of power-line patrolling unmanned plane includes the inspection image to be detected of glass insulator;
Enhancing processing is filtered to inspection image to be detected, obtains filtering enhanced inspection image to be detected;
Defects detection model inspection is revealed using insulator, enhanced inspection image to be detected is filtered, includes with determination
Whether have in the inspection image to be detected of glass insulator includes self-destruction defect.
Be construed as, include glass insulator inspection image to be detected in may include self-destruction defect, it is also possible to
It does not include self-destruction defect.
The self-destruction defect type that may include in the inspection image to be detected for including glass insulator includes: that glass is exhausted
Monolithic is revealed in edge substring, continuous multi-disc is revealed in glass insulator strings, fitting connecting pin monolithic is revealed in glass insulator strings
It is revealed with glass insulator strings fitting connecting pin multi-disc.
When it is implemented, the insulator reveals defect inspection method, it is filtered enhancing processing, comprising:
1) inspection image, the inspection image after obtaining median filtering are traversed using four neighborhood median filterings;
2) after using the Sobel boundary operator traversal median filtering with 0 °, 45 °, 90 ° and 135 ° this 4 direction templates
Inspection image, obtain edge inspection image outstanding;
3) four neighborhood intensity contrast methods, to be detected inspection image de-noising outstanding to edge, after obtaining filtering enhancing are utilized
Inspection image.
When it is implemented, the insulator reveals defect inspection method, using four neighborhood intensity contrast methods, after prominent to edge
Inspection image de-noising to be detected, obtain filtering enhanced inspection image to be detected, comprising:
To any pixel point f (x, y), if | f (x, y)-f (x+i, y+j) | > TH, for any i=-1,1;J=-1,1
It sets up, then pixel is noise spot;Wherein, threshold value TH is the constant not less than 20;
The gray value that will determine as the pixel of noise spot is appointed as 255.
When it is implemented, the insulator reveals defect inspection method, the virtual inspection comprising insulator self-destruction defect is generated
Image and automatic marking insulator self-destruction defect, comprising:
1) there is the glass insulator and normal glass insulation of different self-destruction defect types using laser scanner measurement
The material object of son obtains three-dimensional geometry data in kind;
2) according to three-dimensional geometry data, it is directed to the sub-pieces of different self-destruction defect types respectively, normal insulation sub-pieces, connects
Fitting establishes three dimensional virtual models, and rendering obtains defective insulation sub-image, normal insulation sub-image is connected with fitting
The image of part;
3) according to preset rule, insulator chain virtual image is generated, in the not same district of insulator chain virtual image
At least one defective insulator is randomly distributed in domain;
4) it takes pictures to the insulator chain virtual image of generation from the simulation of multiple and different angles, the result that will take pictures rendering output is
Picture images format, and record the coordinate letter of the pixel of defective insulator and the sub- region of neighbouring normal insulation
Breath.
When it is implemented, in insulator self-destruction defect inspection method, the artificial insulator self-destruction defect that marks includes:
Self-destruction in the true inspection image for the glass insulator with different self-destruction defect types that artificial mark obtains
Region where defect area and its neighbouring normal insulation sub-pieces or fitting connecting pin.
When it is implemented, in insulator self-destruction defect inspection method, automatic marking insulator self-destruction defect includes:
It determines in the virtual mark shadow of the glass insulator with different self-destruction defect types under multiple and different shooting visual angles certainly
The coordinate information of the pixel in the region where quick-fried disordered insulator piece and its neighbouring normal insulation sub-pieces or fitting connecting pin.
After mark, defect area is expanded, defect combination zone is formd, to be conducive to target detection.Specifically
Ground, defect combination zone include with Types Below:
The collection of pixel where the neighbouring normal insulation sub-pieces of monolithic insulator piece and its both ends with self-destruction defect
It closes;Or
Multi-disc sub-pieces with continuous self-destruction defect and respectively both ends adjacent to multi-disc sub-pieces two panels just
The set of pixel where normal sub-pieces;
Fitting connecting pin, with fitting connection have self-destruction defect first insulator and it is adjacent with first insulator under
The set of pixel where a piece of normal sub-pieces;
Fitting connecting pin, the continuous multi-disc insulator with self-destruction defect being connect with fitting and with multi-disc insulator phase
The set of pixel under adjacent where a piece of normal sub-pieces.
When it is implemented, the insulator reveals defect inspection method, further includes:
It include that fitting is had into self-destruction after revealing defect in the inspection image to be detected that determination includes glass insulator
The sub-pieces of defect and its both ends it is neighbouring normal insulation where pixel region minimum circumscribed rectangle be superimposed upon it is to be detected
On inspection image.
When it is implemented, the insulator reveals defect inspection method,
It is based on Faster R-CNN deep learning network using joint training collection training, determines that insulator reveals defects detection
When model,
If being greater than preset detection threshold value for the confidence level of corresponding self-destruction defect type, it is determined that testing result
Effectively, and in inspection image with rectangle frame label insulator the region where defect is revealed;
If being not more than preset detection threshold value for the confidence level of corresponding self-destruction defect type, it is determined that detection knot
Fruit is invalid.
The insulator self-destruction defect inspection method that the present embodiment proposes is with glass insulator in electric power unmanned plane inspection image
Revealing defect is detection target, relatively enriches this using the insulator of self-destruction defect and the structure feature of adjacent normal insulation
Characteristic establishes combination identification model, and the target to be detected as insulator self-destruction defect;Using virtual engine rendering data with
A small amount of true inspection image constructs training set jointly;Original image data is used to train Faster R- after filtering enhancing
CNN deep learning network generates the detection model that accuracy rate is met the requirements;It is obtained with the detection model of generation processing electric power unmanned plane
Insulator self-destruction defect can be effectively detected out in the inspection image taken.
The insulator self-destruction defect inspection method that the present embodiment proposes improves the accuracy rate of insulator self-destruction detection, thus
Transmission line of electricity maintenance work efficiency can effectively be promoted.
The insulator that the present embodiment described further below proposes reveals defect inspection method.
1) virtual insulator threedimensional model, is made
The selection of 1.1 insulator target areas
Due to having this relative Repeat of detection target in electric inspection process image of the sub-pieces of self-destruction defect smaller,
The present embodiment will test target and be expanded to the normal insulation sub-pieces neighbouring with the sub-pieces and its both ends for revealing defect.
The combination zone that sub-pieces and the neighbouring normal insulation sub-pieces in its both ends with self-destruction defect occupy is in inspection
There is suitable size, signature, and relatively stable, as the detection model generated based on deep learning network in image
Input, can be improved detection accuracy rate.
Specifically, the combination zone that there are the sub-pieces of self-destruction defect to occupy with the neighbouring normal insulation sub-pieces in its both ends
There are following 3 kinds of situations:
In insulator chain, having the region of the sub-pieces of self-destruction defect is self-destruction defective locations;If its left and right sides
Adjacent two panels sub-pieces are normal, then are that this has the combination zone of self-destruction defect with the normal two panels sub-pieces
Boundary.
If there are the sub-pieces that continuous multi-disc has self-destruction defect in insulator chain, then until its left and right sides phase
It is that this has the combination zone of self-destruction defect with the normal two panels sub-pieces when adjacent two panels insulator is normal
Boundary.
If the first insulator in insulator chain Yu fitting junction has self-destruction defect, with the self-destruction insulator and gold
Having connecting pin to a piece of normal sub-pieces under adjacent is that this has boundary of the combination zone of self-destruction defect.
When generating detection model based on deep learning network, each of used training set or detection standard shadow
In decent, including at least one of the combination zone of above 3 seed type.
1.2 insulators reveal defect object definition classification
The defect sample of the present embodiment include monolithic self-destruction in glass insulator strings, in glass insulator strings continuous multi-disc from
The self-destruction of fitting connecting pin monolithic and glass insulator strings fitting connecting pin multi-disc reveal these four types of glass in quick-fried, glass insulator strings
Insulator reveals defect target, and using the glass insulator strings of normal mounting as check sample, carries out subsequent defects detection
Test.
1.3 true insulator part 3 D laser scannings
Insulator and its connector is accurately measured using laser scanner, to obtain the accurate modeling data of insulator.
Specifically, it is revealed with monolithic in laser scanner difference scanning glass insulator chain, is continuous in glass insulator strings
Multi-disc self-destruction, glass insulator strings fitting connecting pin monolithic self-destruction, glass insulator strings fitting connecting pin multi-disc self-destruction etc. this four
Class glass insulator reveals the master sample of defect and the glass insulator strings of the normal mounting as control, to obtain normal
With the accurate three-dimensional geological information of defective all kinds of insulators.
The modeling of 1.4 virtual samples
Specifically, when virtual sample models, using the 3DMax software with stronger modeling function and rendering capability.Input
Using the accurate three-dimensional information for each component of insulator that laser scanner obtains, select Polygon to sub-pieces and correspondence
Connection gold utensil modeled, can be obtained accurate three-dimensional insulation sub-portfolio and insulator expose oneself defect model.
Data processing and three-dimensional modeling can also be carried out using Geomagic software.And it is allowed using Unity3D render component
Model achievees the effect that close to reality.
In addition, revealing defect to insulator, then design function is needed, so that four class defects, in virtual insulator chain
Different zones are randomly generated.
It is construed as, in an effective virtual image sample, including at least at least one in above 4 kinds of defect types
Class.
1.5 generate the virtual inspection image sample of the self-destruction defect containing insulator
In unmanned plane electric inspection process, unmanned plane would generally fly along overhead transmission line, in defined job position
Photographic subjects equipment routing inspection image.When generating virtual inspection image sample, according to the shooting feature of unmanned plane inspection to virtual
Virtual camera in scene is controlled, to simulate the shooting angle in true unmanned plane electric inspection process, makes to generate virtual
Sample Maximal limit close to true unmanned plane inspection image.
Specifically, by camera (Camera) function in Unity3D game engine, simulation bat is carried out to insulator chain
According to, and the result rendering that will take pictures exports as general picture images format;And record defects of insulator position in virtual image
Pixel coordinate information and the markup information file for generating corresponding insulator self-destruction position, wherein remember in the markup information file
Record has the defect attribute of insulator self-destruction defect and four apex pixel coordinate values of self-destruction position place rectangular area.
The method of above-mentioned generation virtual image and markup information had both solved insulator in inspection image and has revealed defect sample
Rare problem, and the workload manually marked is greatly reduced.
The above virtual image insulator is revealed into defect sample and the real image insulator that is obtained by electric power unmanned plane from
Quick-fried defect sample is combined according to set proportion, training when collectively as based on deep learning network generation detection model
Collection or test set.
2), images filter enhances
Insulator self-destruction defect target has geometric profile feature relatively outstanding, and the present embodiment introduces filtering enhancing step
Suddenly, the edge gradient information of insulator self-destruction defect target in inspection image is further enhanced, while weakening ambient noise to inspection
The influence of result is surveyed, to increase the detection probability of insulator self-destruction defect target.Specific steps such as Fig. 2 institute of images filter enhancing
Show.
2.1. image median filter
Edge and noise in image show as radio-frequency component in a frequency domain.Place is filtered to image before testing
Reason, to reduce influence of the noise to edge detection.
Preferably, using median filtering.When median filtering, with the window point by point scanning on the image of a fixed size,
Pixel in window is pressed the big minispread of gray scale, and gray value of the gray value for taking gray scale placed in the middle as the pixel of window center.
Specifically, the inspection image for being M × N for a width resolution ratio remembers the gray scale of each pixel in inspection image
Value is f (x, y), is traversed using four neighborhood median filterings to image, after median filtering, remembers each pixel in inspection image
The gray value of point is G (x, y):
G (x, y)=MED { f (x-1, y), f (x, y-1), f (x, y+1), f (x+1, y), f (x, y) }
Wherein, MED is the function for taking intermediate value.
2.2. increase Sobel direction template
There are many directions at the edge of image, and other than both horizontally and vertically, there are also other edge directions, such as 45 °
Deng.
In order to increase the precision at pixel detection edge, it is by 2 increases by the direction template of classical Sobel boundary operator
4, it is respectively as follows:
Each width inspection image is successively traversed respectively with above 4 direction templates.Such as, in re-sampling operations, to each
A pixel f (x, y), the resampling value of gray scale are pixel value in 3 × 3 regions put centered on it according to template above
Weight calculate after gray value.
2.3 eliminate noise spot
Insulator reveals all potential target in defect image, and edge has direction and amplitude the two features.It hangs down
It is directly moved towards in edge, grey scale pixel value variation is violent;It is moved towards along edge, grey scale pixel value variation is gentle, i.e., along marginal point
Direction can find the other edge point of sum of the grayscale values direction difference very little.
Due to noise be it is random, any noise pixel point is difficult to find sum of the grayscale values direction difference very little along edge direction
Another noise spot.
The present embodiment utilizes this basic thought, and actual edge point is distinguished with noise spot.Specifically, for a width
Specific digital picture, after detecting marginal point using 4 direction template Sobel operators, to any pixel point f (x, y) if |
F (x, y)-f (x+i, y+j) | > TH, for any i=0, -1,1;J=0, -1,1 sets up, then can determine whether that the pixel is to make an uproar
Sound point;Preferably, the value of threshold value TH is 30.
Then, then from image, noise spot f (x, y) is removed.Specifically, the gray value of the point is assigned a value of 255,
Noise spot is turned into white background.
By operating above, weakened the color and background texture of object, at the same the profile of prominent object and
The feature of shape can avoid deep learning decision rule from occurring over-fitting at virtual image details existing to a certain extent
As.
The above filtering processing has not been changed the resolution ratio of original inspection image, therefore inspection image to be detected enhances through filtering
After processing through based on the detection model of deep learning network detection after, the pixel coordinate of the target area of output in original shadow
Pixel coordinate as in is identical;Therefore, in output test result, the rectangle marked frame that insulator can be revealed to region is made
It is being shown in raw video for map overlay.
3), Faster R-CNN deep learning network detection insulator reveals defect
The present embodiment is used in the widely applied Faster R-CNN deep learning network in deep learning field, Faster
R-CNN deep learning network has significant advantage in the accuracy rate and computational efficiency of target detection.
The extracted region network that Faster R-CNN deep learning network is connected in parallel by shared weight layer and thereafter
(Region proposal network, hereinafter referred to as RPN) and target detection network (Fast R-CNN, hereinafter referred to as FRCN)
Composition.
Faster R-CNN deep learning network in the present embodiment uses 16 layers of VGG16 network in shared weight layer.
Extracted region network RPN is the full convolutional network with multiple convolutional layer structures.RPN is by judging last
Each pixel position pixel belongs to the probability of target to be detected in layer feature, provides the picture of candidate region to FRCN network
Plain coordinate initial value is for detection.FRCN is divided into two parallel external frame Recurrent networks and target category score value network again.
Utilize a small amount of true inspection image and the virtual inspection image of generation and its corresponding defect area markup information
Constitute, training set trains Faster R-CNN deep learning network, and obtain insulator self-destruction defects detection model, the detection
Model reveals defects detection for insulator, and output meets the insulator self-destruction defect target area of probabilistic confidence threshold condition
Pixel coordinate.
After being further processed, self-destruction defect target area can be highlighted in a manner of rectangle frame in inspection image.
The process of insulator self-destruction defects detection specifically: input inspection image to be detected, pass through Faster R-CNN
Deep learning network share weight layer successively carries out convolution operation pixel-by-pixel to input image using VGG16 network, generates and corresponds to
All feature vectors of all pixels in layer;Feature vector is inputted to parallel extracted region network RPN and target detection network
The insulator generated after the feature vector and training of all potential insulator self-destruction target area pixels is revealed defect mark by FRCN
Quasi- feature vector compares, and solves through loop iteration, and matching probability is confidence score value;It determines and is greater than preassigned threshold value
Insulator self-destruction defect area pixel point coordinate be the target area detected.The target area has one and corresponding sets
Confidence score value.
When it is implemented, the detection model detection insulator self-destruction defect based on Faster R-CNN deep learning network
Specific steps are as shown in Figure 3.
3.1 input inspection video generations and extraction target signature to be detected
Original inspection image is input to Faster R-CNN and shares weight layer, before shared weight layer VGG16 network
13 layers of multilayer convolution sum down-sampling processing;Treated, and original inspection image can reflect the implicit letter of some non-characterizations in figure
Breath, and eliminated the interference of unrelated images information.
In the present embodiment, in the VGG16 network of shared weight layer, setting convolution kernel is the sliding window of 3 × 3 pixels.
For each sliding window, multidimensional characteristic vectors can be generated, be denoted as k;Then for every layer of w × h rule of convolutional neural networks
The characteristic image of lattice is correspondingly generated w × h × k feature vector.
The sliding window sequentially traverses through original input picture, and layer-by-layer down-sampling, the feature vector input area that will acquire
Characteristic of field extracts network RPN.
Specifically, what convolution algorithm extracted is different feature, some are distribution of color, some are textural characteristics, some
It is boundary characteristic, some are corner features etc.;This extraction process is blindness.
These features tentatively extracted can obtain the stronger feature of ability to express after several convolutional layers below.Convolution
The result of operation can regard the process of template matching degree calculating as.
Specifically, for inspection image, the 512 full connection features of dimension can be generated by standard VGG16 convolutional neural networks
Vector.
3.2 depth excavate image feature information
The full connection features vectors input of the 512 of generation dimensions had into external frame Recurrent networks and target category score value network
The full articulamentum of the two branches of FRCN by full articulamentum integration characteristics and transfers to classifier to carry out classification processing, and generation is sentenced
It establishes rules then, the detection foundation as subsequent insulation sub-goal.
The advantage for realizing full connection processing by the way of convolution can be handled so that the size of input image does not limit
The inspection original image of arbitrary resolution.
In the detection model based on Faster R-CNN deep learning network, prediction interval RPN and determine layer FRCN this two
The effect difference of a branch is as follows:
1, RPN: for determine insulator self-destruction defect Objective extraction feature central point pixel coordinate x, y and the high w of width,
h;
2, FRCN: for determining that this feature belongs to non-targeted background in insulation sub-goal or picture to be identified.Using sliding
The processing mode of window guarantees whole feature spaces of traversal two above branch association convolutional layer, convolution results input area
Feature extraction network RPN carries out next round loop iteration.
The above process is realized by following rule iterative calculation.
1, training Region Feature Extraction network, initial model are initialized using ImageNet pre-training model, with
The layer parameter shared in VGG16 network can train the parameter in obtained model with direct copying through ImageNet;
2, it with Region Feature Extraction network RPN and target category score value network FRCN, alternately trains, extracts insulator self-destruction
Pixel region feature.
3, each layer loss function uses the Gaussian Profile with standard deviation=0.01.
4, it initializes and shares convolutional layer, iteration executes 1,2,3, and iteration 50000 times, training terminates.
5, black box decision rule is generated, each pixel region of original input picture is generated according to above-mentioned calculating
Black box formula decision rule, specially the black box feature calculation of area pixel, confidence value value range are by 0 to 1;Work as confidence level
When being 1, it is denoted as exact matching.
Confidence score, which calculates, in the present embodiment uses Faster R-CNN algorithm source code, not altered.
3.3 positioning defects of insulator target pixel regions
A, by the collected Feature Dimension Reduction of different types of window to fixed dimension.
After the inspection image to be detected of input is carried out convolution operation down-sampling, input classification layer is generated using in 3.2
Decision rule, classification layer provide include in sliding window target confidence score, to preset size be 0~1 it
Between confidence threshold value, wherein the high rectangular pixel area of score as positive sample, score it is low be considered negative sample, and give up
It abandons.
If previously given confidence threshold value is given 0.9, then when confidence level is more than or equal to 0.9, it is believed that testing result has
Effect, and further display the target area detected;If being lower than 0.9, then it is assumed that testing result mistake.
B, when being greater than confidence threshold value in result judgement pixel region that layer provides of classifying, it is believed that in the pixel region
There is insulator to reveal defect target, needs the further regression correction of position frame to target, the rectangle frame for exporting recognition result
Region is accurately bonded the minimum circumscribed rectangle of insulator self-destruction defect target pixel region.
Degree of overlapping judgment threshold in position regression correction calculating is calculated by IOU, and two frame intersection area area of IOU=/
Two frame union areas.When two its values of pixel region area degree of overlapping of output are greater than 0.3, it is believed that two above region may
It in the presence of output is repeated, needs to re-start recurrence and calculates, eliminate extra repetition output.
If corresponding insulator self-destruction target pixel region and the target that is originally inputted in inspection image in input picture
The degree of overlapping of real estate is more than or equal to setting IOU threshold value (e.g., 0.5), then determines that the pixel region has insulation sub-goal, is arranged
Area label is 1;If Duplication is less than IOU threshold value (e.g., 0.5), then setting area label is 0, it is believed that the area pixel is non-
The background area of target.
C, the region for being 1 for label, find map back in the coordinate and image of input picture true coordinates of targets it
Between mapping relations, complete return position fixing process, determine insulator area pixel coordinate position.The region that label is 0 is considered
Wrong identification is not involved in subsequent arithmetic.
D, by regression correction, classify to target, and use multitask loss function frame regression algorithm, obtain
The coordinate of the exact boundary frame of target in down-sampled images restores later to inspection image original resolution step by step, exports original
The pixel coordinate on four vertex of minimum circumscribed rectangle of insulator self-destruction defect target area, draws under inspection image pixel coordinate system
Rectangle processed is simultaneously highlighted, for manual examination and verification.
Defect inspection method is revealed to the insulator based on inspection image below, is completely illustrated stage by stage.
1. insulator and defect equipment three-dimensional modeling
Using the accurate dimension of the small fitting such as vernier caliper measurement connection gold utensil, buckle, generated using polygon modeling vertical
Body Model;
Laser scanner scans normal glass insulator chain various kinds of equipment component and four class disordered insulators reveal defect
3 D stereo information, and modeled in virtual emulation software.
2. generating virtual sample
Defect sample deficiency problem is revealed for insulator in unmanned plane inspection image, according to components such as the insulators obtained
Three-dimensional information virtually display software in Mass production inspection image.Pixel resolution can be arbitrarily arranged, and resolution ratio is recommended to exist
Arbitrary resolution between 6000 × 4000 to 1024 × 768.It include all kinds of insulators of Program Generating in virtual image sample
Defect target is revealed, recommends the quantity of defects of insulator in every virtual sample inspection image to be no less than 3, in insulator chain two
Each normal insulation sub-pieces installation site generates at random between the connection gold utensil of end.Virtual image sample insulator reveals the label letter of defect
Breath is by programming automatic generation.
3. the artificial a small amount of true inspection image of mark
Artificial mark opens true inspection image comprising insulator self-destruction defect 500, wherein all types of insulators reveal defect
Quantity is balanced as far as possible, the minimum circumscribed rectangle of all kinds of insulator self-destructions defect target area is marked, as insulator in training set
The standard target region of defect, and synchronize corresponding generation markup information.
4. forming training set
The training set that the deep learning of the present embodiment uses is made of the virtual sample generated and a small amount of true inspection image.
Training set quantity recommends 2000 width virtual sample image of Mass production and 500 true inspection images, virtually with authentic specimen number
Amount is than being about 4:1.The above quantity is able to satisfy the requirement of deep learning training set data amount, and can be with balance training Faster R-CNN
The precision of model identification and the cost manually marked.
Wherein, the resolution ratio of true inspection image covers 6000 × 4000 to 1024 × 768 existing each inspection platform mainstreams
Resolution ratio, and it is attached with line information (such as route name, voltage class, affiliated company, provinces and cities).
5. filtering enhancing
Using filtering enhance pretreatment training set virtual and true inspection image, as far as possible enhancing virtual image and really
The weight of target geometric profile feature in image, and enhance the fitness of virtual image and real image.
6. training Faster R-CNN model
Training set is inputted into Faster R-CNN deep learning network with corresponding markup information, training insulator self-destruction lacks
Fall into detection model.
The present invention is described by reference to a small amount of embodiment above.However, it is known in those skilled in the art,
As defined by subsidiary Patent right requirement, in addition to the present invention other embodiments disclosed above equally fall in this hair
In bright range.
Normally, all terms used in the claims are all solved according to them in the common meaning of technical field
It releases, unless in addition clearly being defined wherein.All references " one/described/be somebody's turn to do [device, component etc.] " are all opened ground
At least one example being construed in described device, component etc., unless otherwise expressly specified.Any method disclosed herein
Step need not all be run with disclosed accurate sequence, unless explicitly stated otherwise.
Claims (10)
1. a kind of insulator based on inspection image reveals defect inspection method characterized by comprising
Obtain including several true inspection images of glass insulator defect and manually marking for power-line patrolling unmanned plane shooting
Insulator reveals defect;
It generates virtual inspection image and automatic marking insulator comprising insulator self-destruction defect and reveals defect;
Enhancing processing is filtered to the true inspection image, the virtual inspection image, obtains filtering enhanced true
Inspection image and virtual inspection image;
According to preset quantity true inspection image more enhanced than combined filter and virtual inspection image, joint instruction is constituted
Practice collection;
It is based on Faster R-CNN deep learning network using the joint training collection training, determines that insulator reveals defects detection
Model, the insulator self-destruction defects detection model is in determining the true inspection image or virtual inspection shadow belonging to insulator
Self-destruction defect type when, be greater than preset detection threshold value for the confidence level of corresponding self-destruction defect type.
2. insulator as described in claim 1 reveals defect inspection method, which is characterized in that
The self-destruction defect type includes: monolithic self-destruction in glass insulator strings, continuous multi-disc self-destruction, glass in glass insulator strings
The self-destruction of fitting connecting pin monolithic and the self-destruction of glass insulator strings fitting connecting pin multi-disc in glass insulator chain.
3. insulator as described in claim 1 reveals defect inspection method, which is characterized in that further include:
Obtain the shooting of power-line patrolling unmanned plane includes the inspection image to be detected of glass insulator;
Enhancing processing is filtered to the inspection image to be detected, obtains filtering enhanced inspection image to be detected;
Defects detection model inspection, the enhanced inspection image to be detected of filtering, with determination are revealed using the insulator
It is described include glass insulator inspection image to be detected in whether include self-destruction defect.
4. insulator as described in claim 1 reveals defect inspection method, which is characterized in that
It is described to be filtered enhancing processing, comprising:
1) inspection image, the inspection image after obtaining median filtering are traversed using four neighborhood median filterings;
2) after traversing the median filtering using the Sobel boundary operator with 0 °, 45 °, 90 ° and 135 ° this 4 direction templates
Inspection image, obtain edge inspection image outstanding;
3) four neighborhood intensity contrast methods, to be detected inspection image de-noising outstanding to the edge, after obtaining filtering enhancing are utilized
Inspection image.
5. insulator as described in claim 1 reveals defect inspection method, which is characterized in that
It is described to utilize four neighborhood intensity contrast methods, to be detected inspection image de-noising of the edge after prominent, obtains filtering and increase
Inspection image to be detected after strong, comprising:
To any pixel point f (x, y), if | f (x, y)-f (x+i, y+j) | > TH, for any i=-1,1;J=-1,1 at
Vertical, then the pixel is noise spot;Wherein, threshold value TH is the constant not less than 20;
The gray value that will determine as the pixel of noise spot is appointed as 255.
6. insulator as described in claim 1 reveals defect inspection method, which is characterized in that
It is described to generate the virtual inspection image comprising insulator self-destruction defect and automatic marking insulator self-destruction defect, comprising:
1) glass insulators and normal glass insulator of the laser scanner measurement with different self-destruction defect types are utilized
Material object obtains the three-dimensional geometry data of the material object;
2) according to the three-dimensional geometry data, it is directed to the sub-pieces of different self-destruction defect types respectively, normal insulation sub-pieces, connects
Fitting establishes three dimensional virtual models, and rendering obtains defective insulation sub-image, normal insulation sub-image is connected with fitting
The image of part;
3) according to preset rule, insulator chain virtual image is generated, in the not same district of the insulator chain virtual image
At least one defective insulator is randomly distributed in domain;
4) it takes pictures to the insulator chain virtual image of generation from the simulation of multiple and different angles, the result that will take pictures rendering exports as picture
Image format, and record the coordinate information of the pixel of defective insulator and the sub- region of neighbouring normal insulation.
7. insulator as described in claim 1 reveals defect inspection method, which is characterized in that
The artificial mark insulator reveals defect
Self-destruction defect in the true inspection image for the glass insulator with different self-destruction defect types that artificial mark obtains
Region where region and its neighbouring normal insulation sub-pieces or fitting connecting pin.
8. insulator as described in claim 1 reveals defect inspection method, which is characterized in that
The automatic marking insulator reveals defect
Determine that self-destruction lacks in the virtual mark shadow of the glass insulator with different self-destruction defect types under multiple and different shooting visual angles
Fall into the coordinate information of the pixel in the region where sub-pieces and its neighbouring normal insulation sub-pieces or fitting connecting pin.
9. insulator as claimed in claim 3 reveals defect inspection method, which is characterized in that further include:
It include that fitting is had into self-destruction after self-destruction defect in the inspection image to be detected for including glass insulator described in the determination
The minimum circumscribed rectangle of pixel region where neighbouring normal insulation of the sub-pieces of defect and its both ends be superimposed upon it is described to
It detects on inspection image.
10. insulator as described in claim 1 reveals defect inspection method, which is characterized in that
It is based on Faster R-CNN deep learning network using the joint training collection training, determines that insulator reveals defects detection
When model,
If being greater than preset detection threshold value for the confidence level of corresponding self-destruction defect type, it is determined that testing result has
Effect, and with the region where rectangle frame label insulator self-destruction defect in inspection image;
If being not more than preset detection threshold value for the confidence level of corresponding self-destruction defect type, it is determined that testing result without
Effect.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811376353.6A CN109615611B (en) | 2018-11-19 | 2018-11-19 | Inspection image-based insulator self-explosion defect detection method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811376353.6A CN109615611B (en) | 2018-11-19 | 2018-11-19 | Inspection image-based insulator self-explosion defect detection method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109615611A true CN109615611A (en) | 2019-04-12 |
CN109615611B CN109615611B (en) | 2023-06-27 |
Family
ID=66004524
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811376353.6A Active CN109615611B (en) | 2018-11-19 | 2018-11-19 | Inspection image-based insulator self-explosion defect detection method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109615611B (en) |
Cited By (20)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110120091A (en) * | 2019-04-28 | 2019-08-13 | 深圳供电局有限公司 | Electric inspection process image pattern production method, device and computer equipment |
CN110147777A (en) * | 2019-05-24 | 2019-08-20 | 合肥工业大学 | A kind of insulator category detection method based on depth migration study |
CN110148117A (en) * | 2019-04-22 | 2019-08-20 | 南方电网科学研究院有限责任公司 | Power equipments defect recognition methods, device and storage medium based on electric power image |
CN110222683A (en) * | 2019-06-11 | 2019-09-10 | 云南电网有限责任公司曲靖供电局 | A kind of quick-fried defect recognition localization method of electric transmission line isolator component based on depth convolutional neural networks |
CN110503029A (en) * | 2019-08-21 | 2019-11-26 | 云南电网有限责任公司电力科学研究院 | A kind of detection method of transmission line of electricity glass insulator state |
CN110751642A (en) * | 2019-10-18 | 2020-02-04 | 国网黑龙江省电力有限公司大庆供电公司 | Insulator crack detection method and system |
CN110852988A (en) * | 2019-09-27 | 2020-02-28 | 广东电网有限责任公司清远供电局 | Method, device and equipment for detecting self-explosion of insulator string and storage medium |
CN111182491A (en) * | 2019-12-31 | 2020-05-19 | 淮安中科晶上智能网联研究院有限公司 | Radio frequency tomography-based equipment-free target positioning method and device |
CN111220619A (en) * | 2019-12-05 | 2020-06-02 | 河海大学常州校区 | Insulator self-explosion detection method |
CN111709918A (en) * | 2020-06-01 | 2020-09-25 | 深圳市深视创新科技有限公司 | Product defect classification method combining multiple channels based on deep learning |
CN111784692A (en) * | 2020-08-11 | 2020-10-16 | 国网内蒙古东部电力有限公司 | Method and device for detecting insulator defects in power system and electronic equipment |
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 |
CN112634254A (en) * | 2020-12-29 | 2021-04-09 | 北京市商汤科技开发有限公司 | Insulator defect detection method and related device |
CN112837315A (en) * | 2021-03-05 | 2021-05-25 | 云南电网有限责任公司电力科学研究院 | Transmission line insulator defect detection method based on deep learning |
CN113393453A (en) * | 2021-06-28 | 2021-09-14 | 北京百度网讯科技有限公司 | Method, apparatus, device, medium and product for detecting self-bursting insulators |
CN113469997A (en) * | 2021-07-19 | 2021-10-01 | 京东科技控股股份有限公司 | Method, device, equipment and medium for detecting plane glass |
CN113538631A (en) * | 2020-04-20 | 2021-10-22 | 贤智研究株式会社 | Computer program, method and apparatus for generating virtual defect image by artificial intelligence model generated based on user input |
CN113884500A (en) * | 2021-10-12 | 2022-01-04 | 国家电网有限公司 | Porcelain insulator defect detection method based on ultraviolet imaging |
CN115063618A (en) * | 2022-08-17 | 2022-09-16 | 成都数之联科技股份有限公司 | Defect positioning method, system, equipment and medium based on template matching |
CN117541582A (en) * | 2024-01-09 | 2024-02-09 | 山东海纳智能装备科技股份有限公司 | IGBT insulation quality detection method for high-frequency converter |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106056088A (en) * | 2016-06-03 | 2016-10-26 | 西安电子科技大学 | Single-sample face recognition method based on self-adaptive virtual sample generation criterion |
US20170177997A1 (en) * | 2015-12-22 | 2017-06-22 | Applied Materials Israel Ltd. | Method of deep learining-based examination of a semiconductor specimen and system thereof |
CN107451997A (en) * | 2017-07-31 | 2017-12-08 | 南昌航空大学 | A kind of automatic identifying method of the welding line ultrasonic TOFD D scanning defect types based on deep learning |
CN107918761A (en) * | 2017-10-19 | 2018-04-17 | 九江学院 | A kind of single sample face recognition method based on multiple manifold kernel discriminant analysis |
CN108230391A (en) * | 2017-12-13 | 2018-06-29 | 深圳市鸿益达供应链科技有限公司 | Intelligent identification method |
CN108596886A (en) * | 2018-04-17 | 2018-09-28 | 福州大学 | Aerial Images insulator based on deep learning falls piece fault rapid detecting method |
-
2018
- 2018-11-19 CN CN201811376353.6A patent/CN109615611B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20170177997A1 (en) * | 2015-12-22 | 2017-06-22 | Applied Materials Israel Ltd. | Method of deep learining-based examination of a semiconductor specimen and system thereof |
CN106056088A (en) * | 2016-06-03 | 2016-10-26 | 西安电子科技大学 | Single-sample face recognition method based on self-adaptive virtual sample generation criterion |
CN107451997A (en) * | 2017-07-31 | 2017-12-08 | 南昌航空大学 | A kind of automatic identifying method of the welding line ultrasonic TOFD D scanning defect types based on deep learning |
CN107918761A (en) * | 2017-10-19 | 2018-04-17 | 九江学院 | A kind of single sample face recognition method based on multiple manifold kernel discriminant analysis |
CN108230391A (en) * | 2017-12-13 | 2018-06-29 | 深圳市鸿益达供应链科技有限公司 | Intelligent identification method |
CN108596886A (en) * | 2018-04-17 | 2018-09-28 | 福州大学 | Aerial Images insulator based on deep learning falls piece fault rapid detecting method |
Cited By (27)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110148117A (en) * | 2019-04-22 | 2019-08-20 | 南方电网科学研究院有限责任公司 | Power equipments defect recognition methods, device and storage medium based on electric power image |
CN110120091A (en) * | 2019-04-28 | 2019-08-13 | 深圳供电局有限公司 | Electric inspection process image pattern production method, device and computer equipment |
CN110147777A (en) * | 2019-05-24 | 2019-08-20 | 合肥工业大学 | A kind of insulator category detection method based on depth migration study |
CN110222683A (en) * | 2019-06-11 | 2019-09-10 | 云南电网有限责任公司曲靖供电局 | A kind of quick-fried defect recognition localization method of electric transmission line isolator component based on depth convolutional neural networks |
CN110503029A (en) * | 2019-08-21 | 2019-11-26 | 云南电网有限责任公司电力科学研究院 | A kind of detection method of transmission line of electricity glass insulator state |
CN110852988A (en) * | 2019-09-27 | 2020-02-28 | 广东电网有限责任公司清远供电局 | Method, device and equipment for detecting self-explosion of insulator string and storage medium |
CN110751642A (en) * | 2019-10-18 | 2020-02-04 | 国网黑龙江省电力有限公司大庆供电公司 | Insulator crack detection method and system |
CN111220619A (en) * | 2019-12-05 | 2020-06-02 | 河海大学常州校区 | Insulator self-explosion detection method |
CN111182491A (en) * | 2019-12-31 | 2020-05-19 | 淮安中科晶上智能网联研究院有限公司 | Radio frequency tomography-based equipment-free target positioning method and device |
CN111182491B (en) * | 2019-12-31 | 2022-03-22 | 淮安中科晶上智能网联研究院有限公司 | Radio frequency tomography-based equipment-free target positioning method and device |
CN113538631A (en) * | 2020-04-20 | 2021-10-22 | 贤智研究株式会社 | Computer program, method and apparatus for generating virtual defect image by artificial intelligence model generated based on user input |
CN111709918A (en) * | 2020-06-01 | 2020-09-25 | 深圳市深视创新科技有限公司 | Product defect classification method combining multiple channels based on deep learning |
CN111709918B (en) * | 2020-06-01 | 2023-04-18 | 深圳市深视创新科技有限公司 | Product defect classification method combining multiple channels based on deep learning |
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 |
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 |
CN111784692A (en) * | 2020-08-11 | 2020-10-16 | 国网内蒙古东部电力有限公司 | Method and device for detecting insulator defects in power system and electronic equipment |
CN112634254A (en) * | 2020-12-29 | 2021-04-09 | 北京市商汤科技开发有限公司 | Insulator defect detection method and related device |
CN112837315A (en) * | 2021-03-05 | 2021-05-25 | 云南电网有限责任公司电力科学研究院 | Transmission line insulator defect detection method based on deep learning |
CN112837315B (en) * | 2021-03-05 | 2023-11-21 | 云南电网有限责任公司电力科学研究院 | Deep learning-based transmission line insulator defect detection method |
CN113393453A (en) * | 2021-06-28 | 2021-09-14 | 北京百度网讯科技有限公司 | Method, apparatus, device, medium and product for detecting self-bursting insulators |
CN113469997A (en) * | 2021-07-19 | 2021-10-01 | 京东科技控股股份有限公司 | Method, device, equipment and medium for detecting plane glass |
CN113469997B (en) * | 2021-07-19 | 2024-02-09 | 京东科技控股股份有限公司 | Method, device, equipment and medium for detecting plane glass |
CN113884500A (en) * | 2021-10-12 | 2022-01-04 | 国家电网有限公司 | Porcelain insulator defect detection method based on ultraviolet imaging |
CN115063618A (en) * | 2022-08-17 | 2022-09-16 | 成都数之联科技股份有限公司 | Defect positioning method, system, equipment and medium based on template matching |
CN115063618B (en) * | 2022-08-17 | 2022-11-11 | 成都数之联科技股份有限公司 | Defect positioning method, system, equipment and medium based on template matching |
CN117541582A (en) * | 2024-01-09 | 2024-02-09 | 山东海纳智能装备科技股份有限公司 | IGBT insulation quality detection method for high-frequency converter |
CN117541582B (en) * | 2024-01-09 | 2024-04-19 | 山东海纳智能装备科技股份有限公司 | IGBT insulation quality detection method for high-frequency converter |
Also Published As
Publication number | Publication date |
---|---|
CN109615611B (en) | 2023-06-27 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109615611A (en) | A kind of insulator self-destruction defect inspection method based on inspection image | |
CN108596101B (en) | Remote sensing image multi-target detection method based on convolutional neural network | |
CN107742093B (en) | Real-time detection method, server and system for infrared image power equipment components | |
CN106356757B (en) | A kind of power circuit unmanned plane method for inspecting based on human-eye visual characteristic | |
WO2018214195A1 (en) | Remote sensing imaging bridge detection method based on convolutional neural network | |
CN110415342A (en) | A kind of three-dimensional point cloud reconstructing device and method based on more merge sensors | |
CN109598794B (en) | Construction method of three-dimensional GIS dynamic model | |
CN103988226B (en) | Method for estimating camera motion and for determining real border threedimensional model | |
CN102507592B (en) | Fly-simulation visual online detection device and method for surface defects | |
CN109003253A (en) | Neural network point cloud generates system | |
CN110263705A (en) | Towards two phase of remote sensing technology field high-resolution remote sensing image change detecting method | |
CN109711288A (en) | Remote sensing ship detecting method based on feature pyramid and distance restraint FCN | |
CN110288586A (en) | A kind of multiple dimensioned transmission line of electricity defect inspection method based on visible images data | |
CN109190473A (en) | The application of a kind of " machine vision understanding " in remote monitoriong of electric power | |
CN110148162A (en) | A kind of heterologous image matching method based on composition operators | |
KR20110127202A (en) | Fusion of a 2d electro-optical image and 3d point cloud data for scene interpretation and registration performance assessment | |
CN108648169A (en) | The method and device of high voltage power transmission tower defects of insulator automatic identification | |
CN113435282B (en) | Unmanned aerial vehicle image ear recognition method based on deep learning | |
CN109636927B (en) | System and method for training and identifying aircraft attitude measurement algorithm | |
CN110427797A (en) | A kind of three-dimensional vehicle detection method based on geometrical condition limitation | |
CN110532914A (en) | Building analyte detection method based on fine-feature study | |
CN112907614A (en) | Yoov 5-segnet insulator string contour extraction method based on depth feature fusion | |
CN109684910A (en) | A kind of method and system of network detection transmission line of electricity ground surface environment variation | |
CN109684914A (en) | Based on unmanned plane image intelligent identification Method | |
CN108052909A (en) | A kind of thin fibrous cap patch automatic testing method and device based on cardiovascular OCT images |
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 |