CN105976383A - Power transmission equipment fault diagnosis method based on limit learning machine image recognition - Google Patents
Power transmission equipment fault diagnosis method based on limit learning machine image recognition Download PDFInfo
- Publication number
- CN105976383A CN105976383A CN201610318887.8A CN201610318887A CN105976383A CN 105976383 A CN105976383 A CN 105976383A CN 201610318887 A CN201610318887 A CN 201610318887A CN 105976383 A CN105976383 A CN 105976383A
- Authority
- CN
- China
- Prior art keywords
- image
- learning machine
- feature
- transmission
- transmission line
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 230000005540 biological transmission Effects 0.000 title claims abstract description 104
- 238000000034 method Methods 0.000 title claims abstract description 50
- 238000003745 diagnosis Methods 0.000 title abstract 2
- 238000012360 testing method Methods 0.000 claims abstract description 19
- 238000012545 processing Methods 0.000 claims abstract description 17
- 238000012549 training Methods 0.000 claims abstract description 16
- 238000000605 extraction Methods 0.000 claims abstract description 15
- 238000006243 chemical reaction Methods 0.000 claims abstract description 6
- 230000005611 electricity Effects 0.000 claims description 38
- 230000006835 compression Effects 0.000 claims description 18
- 238000007906 compression Methods 0.000 claims description 18
- 238000001514 detection method Methods 0.000 claims description 11
- 230000009467 reduction Effects 0.000 claims description 10
- 238000001914 filtration Methods 0.000 claims description 9
- PCHJSUWPFVWCPO-UHFFFAOYSA-N gold Chemical compound [Au] PCHJSUWPFVWCPO-UHFFFAOYSA-N 0.000 claims description 9
- 239000010931 gold Substances 0.000 claims description 9
- 229910052737 gold Inorganic materials 0.000 claims description 9
- 238000003709 image segmentation Methods 0.000 claims description 8
- 239000012212 insulator Substances 0.000 claims description 8
- 230000036961 partial effect Effects 0.000 claims description 6
- 239000004020 conductor Substances 0.000 claims description 4
- 230000007257 malfunction Effects 0.000 claims description 4
- 230000002123 temporal effect Effects 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 claims 1
- 230000008569 process Effects 0.000 abstract description 11
- 230000009466 transformation Effects 0.000 abstract description 5
- 230000002159 abnormal effect Effects 0.000 abstract 1
- 238000007781 pre-processing Methods 0.000 abstract 1
- 230000008859 change Effects 0.000 description 11
- 239000002245 particle Substances 0.000 description 11
- 230000008901 benefit Effects 0.000 description 6
- 230000000007 visual effect Effects 0.000 description 6
- 238000005457 optimization Methods 0.000 description 5
- 238000011161 development Methods 0.000 description 4
- 230000018109 developmental process Effects 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 238000009499 grossing Methods 0.000 description 4
- 238000013459 approach Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 230000011218 segmentation Effects 0.000 description 3
- 230000003044 adaptive effect Effects 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 2
- 230000007797 corrosion Effects 0.000 description 2
- 238000005260 corrosion Methods 0.000 description 2
- 238000009792 diffusion process Methods 0.000 description 2
- 239000000203 mixture Substances 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 230000035945 sensitivity Effects 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 241001544487 Macromiidae Species 0.000 description 1
- 208000031481 Pathologic Constriction Diseases 0.000 description 1
- 230000006978 adaptation Effects 0.000 description 1
- 230000004888 barrier function Effects 0.000 description 1
- 238000005452 bending Methods 0.000 description 1
- 230000015556 catabolic process Effects 0.000 description 1
- 238000002790 cross-validation Methods 0.000 description 1
- 238000003708 edge detection Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000002401 inhibitory effect Effects 0.000 description 1
- 210000002364 input neuron Anatomy 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012806 monitoring device Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 210000004205 output neuron Anatomy 0.000 description 1
- 238000005192 partition Methods 0.000 description 1
- 238000011056 performance test Methods 0.000 description 1
- 230000002829 reductive effect Effects 0.000 description 1
- 230000008439 repair process Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 150000003839 salts Chemical class 0.000 description 1
- 238000000638 solvent extraction Methods 0.000 description 1
- 230000001629 suppression Effects 0.000 description 1
- 230000036962 time dependent Effects 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
- 230000032258 transport Effects 0.000 description 1
- 210000001215 vagina Anatomy 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
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
-
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/30—Noise filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/46—Extracting features or characteristics from the video content, e.g. video fingerprints, representative shots or key frames
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Multimedia (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a power transmission equipment fault diagnosis method based on limit learning machine image recognition, and the method sequentially comprises the following steps: A, collecting training data and testing data, and transmitting the data to an image processing module; B, carrying out the digitalized conversion and preprocessing through employing an image processing module; C, carrying out the extraction of interested features, and then carrying out the inverse transformation of an image after feature extraction to a space domain; D, enabling the image to be classified according to different power transmission devices, respectively carrying out limit learning machine training, and finally obtaining a fault state of a power transmission device with a fault in a power transmission line through output result comparison. The method can carry out the quick classification and automatic filing of a large number of image normal samples and abnormal samples in an image recognition process of power grid equipment based on image recognition, so as to efficiently and accurately recognize the condition of fault equipment in a power grid.
Description
Technical field
The present invention relates to a kind of transmission facility method for diagnosing faults, particularly relate to a kind of based on extreme learning machine image recognition
Transmission facility method for diagnosing faults.
Background technology
Progress and the change of energy development general layout, the socio-economic development degree of dependence day to electric energy along with science and technology
Benefit strengthens, and relies on present information, communicates and control technology actively develops intelligent grid, it is achieved power network development mode changes, and becomes
The common choice of Challenges for Future is responded actively for International Power industry.Following intelligent grid will realize operation of power networks and control
Information-based, intelligent, to improve energy resource structure and utilization ratio, meet the various demands of electric power application, improve electric power transmission
Economy, safety and reliability.For realizing power transmission network intelligence O&M is overhauled, improve power transmission network power supply capacity and safety is steady
Fixed, the intelligent O&M examination and repair system of development is imperative.
For now, transmission facility detecting system mainly uses the technology such as monitoring device measurement collection that circuit carries out event
Barrier monitoring.But owing to the judgement of power transmission network distribution its fault relatively wide has the biggest a part of composition to need artificially to complete, and right
Can't find timely in some hidden danger the most not occurred existed, such as the damaged surfaces such as insulator, gold utensil, and make whole
Individual system lacks predictable, reduces the intelligent level of monitoring system to a great extent.
Summary of the invention
It is an object of the invention to provide a kind of transmission facility method for diagnosing faults based on extreme learning machine image recognition, energy
Enough based on image recognition sample normal to great amount of images and Fast Classification of exceptional sample in grid equipment image recognition processes
And automatic sorting, to identify faulty equipment situation in electrical network efficiently and accurately.
The present invention uses following technical proposals:
A kind of transmission facility method for diagnosing faults based on extreme learning machine image recognition, comprises the following steps successively:
A: the transmission line of electricity video image under collection transmission facility is properly functioning is as training data, and gathers power transmission network operation shape
Transmission line of electricity video image under state is as test data, the most respectively by the transmission of electricity under properly functioning of the transmission facility that collects
Transmission line of electricity video image under circuit video image and power transmission network run sends to image processing module；
B: utilize image processing module, respectively to the transmission facility collected transmission line of electricity video image under properly functioning and defeated
Transmission line of electricity video image under operation of power networks is digitally converted, and the digitized image after conversion is carried out pretreatment；
C: utilize image processing module, respectively to the transmission facility collected transmission line of electricity video image under properly functioning and defeated
The feature interested in transmission line of electricity video image under operation of power networks carries out feature extraction, then by after sign is extracted
Spatial domain is changed in image inversion；
D: utilize image processing module, by step C obtains to the transmission facility collected transmission line of electricity under properly functioning
Video image carries out the image after feature extraction and inversion, carries out extreme learning machine instruction respectively according to different transmission facility classification
Practicing, respectively obtain the optimum study weight of fitness value and threshold value, last limit of utilization learning machine carries out testing and diagnosing, by than
Relatively output result obtains existing in transmission line of electricity the malfunction of the transmission facility of fault.
In step A, transmission line of electricity video image and power transmission network under utilizing collection transmission facility of taking photo by plane properly functioning run
Transmission line of electricity video image under state.
In step B, digitized image carries out pretreatment and includes image noise reduction, rim detection, compression of images and image segmentation.
In step B, use filter in spatial domain Denoising Algorithm, transform domain filtering and noise reduction method, partial differential equation Denoising Algorithm, the calculus of variations
Denoising Algorithm, morphology scratch filter Denoising Algorithm and/or medium filtering Denoising Algorithm based on spatial domain carry out noise reduction to image.
In step B, utilize Roberts operator, Prewitt operator, Sobel operator, IsotropicSobel operator and/or
Laplacian operator carries out rim detection to image.
In step B, utilize huffman coding to realize the compression of statistical redundancy degree, utilize orthogonal cosine transform DCT to realize sky
Between the compression of redundancy, utilize differential coding DPCM to realize the compression of temporal redundancy, utilize and realize regarding from calculating cosine transform DCT
The compression of feel redundancy.
In step B, use dividing method based on threshold value to carrying out image segmentation.
In step C, feature interested includes insulator feature, conductor characteristic and gold utensil feature.
In step C, when feature interested is carried out feature extraction, to color characteristic, textural characteristics and/or shape facility
Extract.
The present invention can in grid equipment image recognition processes sample normal to great amount of images and exceptional sample quick
Classification and automatic sorting, to identify faulty equipment situation in electrical network efficiently and accurately.Artificial O&M maintenance environment can be overcome to dislike
Bad cannot complete, overhaul the inaccurate and shortcoming such as not in time, it is achieved the intelligent trouble analysis to transmission line of electricity and auxiliary device thereof,
Such as wire, gold utensil and insulator, find in power transmission network equipment that break stock, gold utensil of power transmission line comes off and insulator breakdown pollution etc. in time
Fault, eliminates potential safety hazard present in power transmission network, it is ensured that the sustainable power supply of power transmission network safety and stability.
Accompanying drawing explanation
Fig. 1 is the schematic flow sheet of the present invention.
Detailed description of the invention
Below in conjunction with drawings and Examples, the present invention is made with detailed description:
As it is shown in figure 1, transmission facility method for diagnosing faults based on extreme learning machine image recognition of the present invention, wrap successively
Include following steps:
A: the transmission line of electricity video image under collection transmission facility is properly functioning is as training data, and gathers power transmission network operation shape
Transmission line of electricity video image under state is as test data, the most respectively by the transmission of electricity under properly functioning of the transmission facility that collects
Transmission line of electricity video image under circuit video image and power transmission network run sends to image processing module.
In the present embodiment, the assisted acquisition equipment such as available cruiser is taken photo by plane, and normally transports gathering transmission facility
Transmission line of electricity video image under Hang and the transmission line of electricity video image under power transmission network running status, effectively reduce artificial collection
Time labor intensity, acquisition quality and collecting efficiency are greatly improved.
B: utilize image processing module, respectively to the transmission facility collected transmission line of electricity video image under properly functioning
Transmission line of electricity video image under running with power transmission network is digitally converted, and the digitized image after conversion is carried out pre-place
Reason.
In the present invention, digitized image carries out pretreatment and includes that image noise reduction, rim detection, compression of images and image divide
Cut；
In the present invention, use one or more in six kinds of noise-reduction methods that image carries out noise reduction:
1. filter in spatial domain denoising
Airspace filter denoising is directly to carry out data operation on original image, processes the gray value of pixel.Common sky
Between area image Denoising Algorithm have neighborhood averaging, medium filtering, low-pass filtering etc..
2. transform domain filtering and noise reduction
Image transform domain denoising method is to convert image, and from transform of spatial domain, image is changed to transform domain, then to transform domain
In conversion coefficient process, then carry out inverse transformation by image from transform domain be transformed into spatial domain reach remove image throat sound
Purpose.By image from transform of spatial domain change to the alternative approach of transform domain include Fourier transform, Walsh-Hadanjard Transform,
Cosine transform, Karhunen-Loeve transformation and wavelet transformation etc..Fourier transform and wavelet transformation are then commonly used in image denoising
Alternative approach.
3. partial differential equation denoising
Partial differential equation (PDE) process mainly for low layer pictures and have realized noise reduction.Partial differential equation have anisotropic spy
Point, applies in image denoising, can well keep edge while removing noise.The application of partial differential equation is main
Two classes can be divided into: a kind of is basic Iteration, by time dependent renewal so that image is to effect to be obtained
Fruit gradually approaches, and this algorithm is represented as the equation of Perona and Malik, and to the follow-up work after its improvement.The party
Method has the biggest selection space when determining diffusion coefficient, to the function of diffusion after having, therefore has while forward direction spreads
There is smoothed image and by edge sharpening ability.
4. calculus of variations denoising
Noise-reduction method based on the calculus of variations, by minimizing energy function on the basis of determining the energy function of image
Work, makes image reach smooth state, and wide variety of full variation TV model is exactly this class.
5. morphology scratch filter denoising
By to having the combination with closed operation of the noise image opening operation to filter noise, first to there being noise image to open fortune
Calculating, optional structure salt matrices is bigger than noise size, thus the result of opening operation is to be removed by background noise；Again to back
The image obtained carries out closed operation, is removed by the noise on image.Understanding accordingly, the image type that the method is suitable for is in image
Object size the biggest, and there is no minor detail, can be preferably to this kind of scene image partition effect.
6. medium filtering denoising based on spatial domain
Median filter belongs to Nonlinear Smoothing Filter, and ultimate principle is that the value of any in digital picture or Serial No. is used
The intermediate value replacement of each point in this vertex neighborhood.If (x y) represents that (filter window is A to Pixel of Digital Image point for x, gray value y) to f
Median filter can be defined as:
；
When number n at digital picture or Serial No. midpoint is odd number,,...,Intermediate value be exactly by numerical values recited
Order is in the value of centre；When n is even number, the meansigma methods of two values that our definition is in centre by numerical values recited order is
Intermediate value.
Rim detection, mainly for detection of line facility image edge, provides basic data for subsequent treatment.Different images
Gray scale is different, and boundary has obvious edge, utilizes this feature can split image.In the present invention, utilize following 5 kinds of detections
One or more in operator carry out rim detection to image:
1.Roberts operator
Roberts operator has that location, edge is accurate and feature to noise-sensitive, it is adaptable to edge is obvious and the less figure of noise
As segmentation.Roberts edge detection operator is a kind of operator utilizing local difference operator to find edge, Robert operator image
After process, result edge is not the most smooth.Through analyzing, owing to Robert operator would generally produce in the region near image border
Raw wider response, therefore the edge image using above-mentioned operator to detect often need to do micronization processes, the precision of location, edge is not very
High.
2.Prewitt operator
Have inhibitory action to noise, the principle of suppression noise is average by pixel, but pixel is averagely equivalent to image
Low-pass filtering, so Prewitt operator is not so good as Roberts operator to the location at edge.
3.Sobel operator
Sobel operator and Prewitt operator are all weighted averages, but Sobel operator is thought, the pixel of neighborhood is to current pixel
The impact produced is not of equal value, so having different weights apart from different pixels, the impact producing operator result is also
Different.In general, distance is the most remote, and the impact of generation is the least.
4.IsotropicSobel operator
Weighted average operator, weights are inversely proportional to the distance of adjoint point and central point, gradient amplitude when along different directions detection edge
Unanimously, it is simply that usually said isotropism.
5.Laplacian operator
Laplacian operator belongs to Second Order Differential Operator, has isotropism, i.e. unrelated with change in coordinate axis direction, and coordinate axes rotates
Rear gradient result is constant.But, Laplacian operator is more sensitive to noise ratio, so, image typically first passes through smoothing processing,
Because smoothing processing is also carried out by template, common partitioning algorithm is all that Laplacian operator and smoothing operator are combined
Get up to generate a new template.
Compression of images can not only remove image spatial domain and statistical redundancy, it is often more important that reduces the redundancy of time domain, will
The determination information that can deduce is removed.The present invention utilizes following four method for compressing image image is compressed:
1. the compression of statistical redundancy degree
For a string data being made up of a lot of numerical value, if some value often occurs, and other value seldom occurs,
The most this just constituted statistical redundancy degree by statistical inhomogeneities, it can be compressed.Concrete grammar is to those
The value often occurred represents by short code character, represents the code character that the value infrequently occurred is long, thus eventually for table
Show total code bit of this burst of data, reduced for the code bit represented by fixed length code character.The present invention uses
Concrete entropy coding method in compression of images is mainly huffman coding, i.e. the code length of a numerical value and this numerical value occurs
Probability is inversely proportional to as much as possible.Although huffman coding compression ratio is the highest, about 1.6:1, but benefit is lossless compress.
2. the compression of spatial redundancies
The value of one width video image adjacent spots is the most close or identical, has spatial coherence, here it is spatial redundancies.
The spatial coherence of image represents that neighbor pixel value changes speed.The energy of picture signal is concentrated mainly near low frequency,
The energy of high-frequency signal is decayed rapidly with the increase of frequency.By frequency domain transform, can be by original image signal DC component
And the coefficient of minority low frequency AC components represents, here it is the method for the orthogonal cosine transform DCT in transition coding.DCT is
The basis of JPEG and MPEG compressed encoding, effectively can compress the spatial redundancies of image.
3. the compression of temporal redundancy
For video sequence, unless occurrence scene switching, otherwise successive frames is the most all continuous print.In front and back two, frame is past
Toward comprising the background identical with present frame and object.Only because the movement of the rotation of camera lens or object makes locus occur
Change.Moving the slowest, the conversion of position is the least.Therefore there is extremely strong dependency in time domain in video sequence.It is being aware of one
After the value of individual pixel, the difference of the value of this pixel and the value of later pixel thereof is utilized just and to obtain a rear pixel
Value.Therefore, do not transmit the value of pixel itself and transmit the difference of itself and former frame corresponding pixel points, also can compressed code effectively
Rate, here it is differential coding DPCM.In actual compressed encoding, DPCM is mainly used in each image subblock after dct transform
The transmission of DC coefficient.For ac coefficient, the value of DCT DC coefficient is very big, and the DCT of the most each frame correspondence sub-block
The value of coefficient is general relatively, and in the case of image does not occurs saltus step, its difference is compared with the value of DC coefficient itself
The least.
4. the compression of visual redundancy degree
Visual redundancy degree is for the visual characteristic of human eye.Heterogeneity due to human eye vision so that human eye regards
Feel insensitive to some spatial frequency.Therefore the content of different frequency composition its importance for human eye system in video
It is different, namely there is frequency domain redundancy.Human eye includes for the visual characteristic of image: to luminance signal chrominance signal
Sensitivity, sensitive to low frequency signal comparison high-frequency signal, sensitive to rest image comparison moving image, and to image level lines
Sensitive with vertical bar comparison oblique line.Therefore, it is included in carrier chrominance signal, some data in image high-frequency signal and moving image
Can not make contributions relative to the definition of human eye to increasing image, and be considered as unnecessary.The present invention utilizes human eye pair
The sensitivity of low-frequency information, and the characteristic insensitive to high-frequency information, carried out the low frequency component after discrete cosine transform finely
Quantify, and high fdrequency component is slightly quantified, carry out again after process quantifying, encoding, can more effectively reduce code check.
Image segmentation can make this characteristic phase of all pixels in the same area by dividing the image into into different regions
With, it is the image procossing committed step that arrives graphical analysis.In the present invention, use dividing method based on threshold value.
After threshold value determines, the gray value of threshold value with pixel is compared, grey scale pixel value is divided into more than threshold value and
Less than threshold value two class, foreground and background two class image segmentation result can be obtained.Threshold segmentation has that calculating is simple, operation efficiency
Advantage higher, fireballing.
Threshold process includes global threshold and adaptive threshold etc..Global threshold refers to that entire image uses same threshold
Value does dividing processing, such as maximum variance between clusters, maximum-entropy automatic threshold etc., it is adaptable to background and prospect have substantially contrast
Image.Sometimes, the contrast of target and background zones of different in the picture is different, and this makes it difficult to a unification
Threshold value target is separated with background, at this moment can be respectively adopted different threshold values according to the local feature of image and split.
When actual treatment, the most dynamically according to certain contiguous range, select every some threshold value located, carry out image segmentation, this
The threshold value of sample is adaptive threshold.
For transmission line of electricity Aerial Images, owing to its background mostly is sky or surface vegetation, by global threshold or self adaptation
Threshold value all can obtain preferable Threshold segmentation result.
C: utilize image processing module, respectively to the transmission facility collected transmission line of electricity video image under properly functioning
The feature interested in transmission line of electricity video image under running with power transmission network carries out feature extraction, then will extract through sign
After image inversion change to spatial domain.
Feature interested includes: insulator feature, conductor characteristic and gold utensil feature；
Insulator major failure feature shows as edge crumbling.When there is edge crumbling phenomenon, the density of marginal point can occur relatively
Big change.
Wire major failure feature distortion is disconnected stock, and when there is this type of fault, conductor width attenuates, gray value step-down, and
And protruding with wire somewhere, the marginal point graded of transmission pressure is bigger.
Gold utensil major failure feature shows as corrosion phenomena, and when there is this type of fault, in target area, color characteristic occurs
Significant change.
In the present invention, mainly color characteristic, textural characteristics and shape facility are extracted, below feature extraction is carried out
Describe in detail, for different faults type, feature extraction can be carried out by one or more of method.
1) color characteristic
Color characteristic is a kind of global characteristics, describes the surface nature of scenery corresponding to image or image-region.General face
Color characteristic is feature based on pixel, and the most all pixels belonging to image or image-region have respective contribution.Due to
Color is insensitive, so color characteristic can not catch in image well to the change such as image or the direction of image-region, size
The local feature of object.During it addition, only use color characteristic inquiry, if data base is very big, often can be by many unwanted figures
As being also retrieved.
The present invention uses color histogram graph expression color characteristic, and its advantage is the shadow not changed by image rotation and translation
Ring, also can not be changed by graphical rule by normalization further and be affected, color generation significant change during gold utensil corrosion, can lead to
Cross color characteristic and judge gold utensil fault.
2) textural characteristics
The significant change on luminance level is there is, just because of ability in these modified-images between neighbouring pixel
Show various texture.Texture is an attribute of image-region, and the texture of a pixel is nonsensical.Cause
This, texture relates to context, relevant with the gray value of the pixel in a spatial neighbors relation, and in other words, texture is with figure
As the spatial distribution of grey scale pixel value depends on the type of texture about. the size of this spatial relationship, or definition texture
The size of primitive.
Texture is that some research worker of homogeneity attribute under certain space scale is more than image resolution ratio are with people
Visual system describes texture: texture does not has uniform brightness, but still can be observed like that by portrait homogeneous region
Arrive.Image texture can be perceived under different scale and different resolution.In different distances and different vision attention journeys
Under degree, texture region all can provide different explanations.Under a normal attention and gauged distance, which give for characterizing
The concept of the macro-rule of particular texture, when the most carefully observing, it can be noted that some homogeneous regions and
Edge, it is plain that they sometimes can constitute texture, and texture is to rely on yardstick.
Owing to texture is the characteristic of a kind of body surface, the essential attribute of object can not be reflected completely, so only
High-level picture material cannot be obtained merely with textural characteristics.Different from color characteristic, textural characteristics is not based on pixel
The feature of point, it needs to carry out statistical computation in the region comprising multiple pixel.As a kind of statistical nature, textural characteristics
Often there is rotational invariance, and have stronger resistivity for noise.
When retrieval has the texture image of the aspect bigger difference such as thickness, density, it is a kind of effective for utilizing textural characteristics
Method.But the when that thickness between texture, density etc. being prone to be more or less the same between the information differentiated, common texture is special
Levy the difference between the texture that the visual sense feeling being difficult to reflect people exactly is different.During wire generation line-broken malfunction, image stricture of vagina
Reason change substantially, can judge whether to break down by texture features.
3) shape facility
Shape is one of basic feature portraying object, the most directly perceived with shape facility difference object.Now with special based on shape
The classification of the Alphabet Gesture levied and identification, moving object classification based on shape facility, image retrieval skill based on shape facility
Art etc..The correctness of Shape Feature Extraction is directly connected to successive image retrieval and the quality identified, therefore studies digitized map
The Shape Feature Extraction of picture is significant.For simple image, can directly use contours extract algorithm, and for more
Complicated image uses the method or with Canny operator the contours extract of target out of first rim detection contours extract again.
Shape facility can describe shape information with not bending moment and skeleton.
Various search methods based on shape facility can relatively efficiently utilize target interested in image to enter
Line retrieval, it is generally the case that shape facility has two class method for expressing, a class is contour feature, and another kind of is provincial characteristics.Image
Contour feature mainly for the external boundary of object, the provincial characteristics of image is then related to whole shape area, obtain
The form parameter closing target is often required to first split image, and the effect of image segmentation influences whether the extraction of shape facility.
During insulator breakage fault, shape substantially deforms, and can judge whether to break down by shape facility.
D: utilize image processing module, by step C obtains to the transmission facility collected transmission of electricity under properly functioning
Circuit video image carries out the image after feature extraction and inversion, carries out extreme learning machine respectively according to different transmission facility classification
Training, respectively obtains the optimum study weight of fitness value and threshold value, and last limit of utilization learning machine carries out testing and diagnosing, passes through
Relatively output result obtains existing in transmission line of electricity the malfunction of the transmission facility of fault.
Extreme learning machine has only to arrange the hidden node number of network, need not adjust network during algorithm performs
Input weights and the biasing of hidden unit, and produce unique optimal solution, therefore have that pace of learning is fast and Generalization Capability good
Advantage.
Quantum particle swarm optimization is a kind of heuristic random searching algorithm, has that speed of searching optimization is fast, algorithm is prone to real
Now with control the advantages such as parameter is few.Quantum particle swarm optimization is applied to input weight and the choosing of threshold value of extreme learning machine
During selecting, significantly more efficient input weight and threshold value can be obtained in span, thus improve the hidden of extreme learning machine
The effectiveness of layer node, simplifies the network structure of extreme learning machine, improves the respond to unknown data.
The present invention uses extreme learning machine algorithm based on quantum particle swarm optimization, is divided into initialization limit study
Machine, training extreme learning machine and test limits learning machine 3 part.
1. initialize extreme learning machine.
First set the network structure of extreme learning machine, be the node numbers of hidden layers mesh L setting extreme learning machine.Obtain defeated
Enter data and output data, input data are normalized pretreatment, if the classification problem of being applied to, total m class, n bar
Input data, arrange input neuron number and equal to output data dimension, obtain equal to input data dimension, output neuron number
Obtain number of training ntr, cross validation sample number nv, test sample number nteWith maximum frequency of training Tmax.
2. training extreme learning machine.
Suitably input weight and threshold value are instructed mainly to use quanta particle swarm optimization evolutionary computation to draw in this stage
Practice extreme learning machine.Record current CPU time as training sart point in time.It specifically comprises the following steps that
1) in the span of input weight and threshold value, according to particle dimension and the position of number of particles random initializtion particle
Information；
2) current group average optimal position and current converging diverging coefficient are calculated；
3) according to fitness function and position, the particle fitness value in current location is calculated；
4) compare particle current location and the fitness value of its individual two positions of history optimal location, update individual history optimum
Position；
5) calculate the positional information of particle, check particle position respectively tie up after updating whether directed overshoot space and do futile searches；
6) iterated conditional judges.If maximum iteration time Tmax that iterations t sets less than user at that time, continue iteration and perform
Algorithm；
7) colony's optimal location of acquisition is converted into input weight and the threshold value of extreme learning machine, calculates according to training dataset
The output weight of network, completes the learning machine process of extreme learning machine, it is thus achieved that current training error；
3. extreme learning machine test phase.
Record current CPU time as test sart point in time.The extreme learning machine model of acquisition is used for test number
Carry out performance test according to collection, according to the input weight of extreme learning machine and threshold value and output weight, calculate in test data set
Actual output.Record current CPU time as test end time point, calculating testing time.By reality output and desired output
Compare, the test error of calculating limit learning machine.The training time of output limit learning machine, the testing time, training error,
Training error variance, test error and test error variance.
Claims (9)
1. a transmission facility method for diagnosing faults based on extreme learning machine image recognition, it is characterised in that include successively with
Lower step:
A: the transmission line of electricity video image under collection transmission facility is properly functioning is as training data, and gathers power transmission network operation shape
Transmission line of electricity video image under state is as test data, the most respectively by the transmission of electricity under properly functioning of the transmission facility that collects
Transmission line of electricity video image under circuit video image and power transmission network run sends to image processing module；
B: utilize image processing module, respectively to the transmission facility collected transmission line of electricity video image under properly functioning and defeated
Transmission line of electricity video image under operation of power networks is digitally converted, and the digitized image after conversion is carried out pretreatment；
C: utilize image processing module, respectively to the transmission facility collected transmission line of electricity video image under properly functioning and defeated
The feature interested in transmission line of electricity video image under operation of power networks carries out feature extraction, then by after sign is extracted
Spatial domain is changed in image inversion；
D: utilize image processing module, by step C obtains to the transmission facility collected transmission line of electricity under properly functioning
Video image carries out the image after feature extraction and inversion, carries out extreme learning machine instruction respectively according to different transmission facility classification
Practicing, respectively obtain the optimum study weight of fitness value and threshold value, last limit of utilization learning machine carries out testing and diagnosing, by than
Relatively output result obtains existing in transmission line of electricity the malfunction of the transmission facility of fault.
Transmission facility method for diagnosing faults based on extreme learning machine image recognition the most according to claim 1, its feature
It is: in step A, the transmission line of electricity video image under utilizing collection transmission facility of taking photo by plane properly functioning and power transmission network running status
Under transmission line of electricity video image.
Transmission facility method for diagnosing faults based on extreme learning machine image recognition the most according to claim 1, its feature
Being: in step B, digitized image carries out pretreatment and includes image noise reduction, rim detection, compression of images and image segmentation.
Transmission facility method for diagnosing faults based on extreme learning machine image recognition the most according to claim 3, its feature
It is: in step B, uses filter in spatial domain Denoising Algorithm, transform domain filtering and noise reduction method, partial differential equation Denoising Algorithm, the calculus of variations to go
Make an uproar method, morphology scratch filter Denoising Algorithm and/or medium filtering Denoising Algorithm based on spatial domain carries out noise reduction to image.
Transmission facility method for diagnosing faults based on extreme learning machine image recognition the most according to claim 3, its feature
Be: in step B, utilize Roberts operator, Prewitt operator, Sobel operator, IsotropicSobel operator and/or
Laplacian operator carries out rim detection to image.
Transmission facility method for diagnosing faults based on extreme learning machine image recognition the most according to claim 3, its feature
It is: in step B, utilizes huffman coding to realize the compression of statistical redundancy degree, utilize orthogonal cosine transform DCT to realize space superfluous
The compression of remaining, utilizes differential coding DPCM to realize the compression of temporal redundancy, and it is superfluous that utilization realizes vision from calculation cosine transform DCT
The compression of remaining.
Transmission facility method for diagnosing faults based on extreme learning machine image recognition the most according to claim 3, its feature
It is: in step B, uses dividing method based on threshold value to carrying out image segmentation.
Transmission facility method for diagnosing faults based on extreme learning machine image recognition the most according to claim 1, its feature
Being: in step C, feature interested includes insulator feature, conductor characteristic and gold utensil feature.
Transmission facility method for diagnosing faults based on extreme learning machine image recognition the most according to claim 8, its feature
It is: when feature interested is carried out feature extraction, color characteristic, textural characteristics and/or shape facility are extracted.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610318887.8A CN105976383A (en) | 2016-05-16 | 2016-05-16 | Power transmission equipment fault diagnosis method based on limit learning machine image recognition |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610318887.8A CN105976383A (en) | 2016-05-16 | 2016-05-16 | Power transmission equipment fault diagnosis method based on limit learning machine image recognition |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN105976383A true CN105976383A (en) | 2016-09-28 |
Family
ID=56992503
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610318887.8A Pending CN105976383A (en) | 2016-05-16 | 2016-05-16 | Power transmission equipment fault diagnosis method based on limit learning machine image recognition |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105976383A (en) |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106780639A (en) * | 2017-01-20 | 2017-05-31 | 中国海洋大学 | Hash coding method based on the sparse insertion of significant characteristics and extreme learning machine |
| CN107080533A (en) * | 2017-05-22 | 2017-08-22 | 北京麦迪克斯科技有限公司 | Physiological information acquisition device and method |
| CN107246959A (en) * | 2017-07-12 | 2017-10-13 | 西安因联信息科技有限公司 | The diagnostic system and method for a kind of equipment fault based on wireless senser |
| CN108037770A (en) * | 2017-11-22 | 2018-05-15 | 国网山东省电力公司济宁供电公司 | Unmanned plane power transmission line polling system and method based on artificial intelligence |
| CN108189043A (en) * | 2018-01-10 | 2018-06-22 | 北京飞鸿云际科技有限公司 | A kind of method for inspecting and crusing robot system applied to high ferro computer room |
| CN109325313A (en) * | 2018-11-01 | 2019-02-12 | 大唐环境产业集团股份有限公司 | Based on improvement quantum telepotation boiler of power plant NOXPrediction model device |
| CN109492807A (en) * | 2018-11-01 | 2019-03-19 | 大唐环境产业集团股份有限公司 | Based on the boiler NO for improving quanta particle swarm optimizationXPrediction model optimization method |
| CN109526119A (en) * | 2018-09-29 | 2019-03-26 | 镇江微芯光子科技有限公司 | A kind of action control sense light |
| CN110287983A (en) * | 2019-05-10 | 2019-09-27 | 杭州电子科技大学 | Based on maximal correlation entropy deep neural network single classifier method for detecting abnormality |
| CN110888025A (en) * | 2019-11-27 | 2020-03-17 | 华东师范大学 | GIS equipment fault judgment method based on machine learning |
| CN110929611A (en) * | 2019-11-12 | 2020-03-27 | 中国地质大学（武汉） | Modeling method of power transmission line icing thickness prediction model based on PR-KELM |
| CN112163573A (en) * | 2020-11-09 | 2021-01-01 | 云南电网有限责任公司迪庆供电局 | Insulator identification method and device |
| CN112464756A (en) * | 2020-11-13 | 2021-03-09 | 上海电力大学 | Image quantification method for insulator defect identification |
| CN112706723A (en) * | 2020-12-28 | 2021-04-27 | 石家庄开发区天远科技有限公司 | Control method and device of vehicle cleaning device and terminal equipment |
| CN112766251A (en) * | 2020-12-30 | 2021-05-07 | 广东电网有限责任公司佛山供电局 | Infrared detection method and system for power transformation equipment, storage medium and computer equipment |
| CN112991297A (en) * | 2021-03-15 | 2021-06-18 | 广东电网有限责任公司清远供电局 | Power transmission line fault diagnosis method and system, electronic equipment and storage medium |
| CN113486731A (en) * | 2021-06-17 | 2021-10-08 | 国网山东省电力公司汶上县供电公司 | Abnormal state monitoring method for power transmission equipment |
| CN114120197A (en) * | 2021-11-27 | 2022-03-01 | 中国传媒大学 | Method for detecting abnormal signals of ultra-high definition video transmitted in 2SI mode |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1822046A (en) * | 2006-03-30 | 2006-08-23 | 上海电力学院 | Infrared and visible light image fusion method based on regional property fuzzy |
| CN103617630A (en) * | 2013-12-18 | 2014-03-05 | 国网重庆潼南县供电有限责任公司 | Infrared image dividing method and system for power system equipment based on wavelet analysis |
| CN104123566A (en) * | 2013-04-24 | 2014-10-29 | 信帧电子技术（北京）有限公司 | Video fuzziness-invariable target tracking method and system |
| CN104483326A (en) * | 2014-12-19 | 2015-04-01 | 长春工程学院 | High-voltage wire insulator defect detection method and high-voltage wire insulator defect detection system based on deep belief network |
| CN104637049A (en) * | 2014-12-16 | 2015-05-20 | 北京航天时代光电科技有限公司 | Automatic detection method for optical fiber coiling quality |
| CN105069807A (en) * | 2015-08-28 | 2015-11-18 | 西安工程大学 | Punched workpiece defect detection method based on image processing |
-
2016
- 2016-05-16 CN CN201610318887.8A patent/CN105976383A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1822046A (en) * | 2006-03-30 | 2006-08-23 | 上海电力学院 | Infrared and visible light image fusion method based on regional property fuzzy |
| CN104123566A (en) * | 2013-04-24 | 2014-10-29 | 信帧电子技术（北京）有限公司 | Video fuzziness-invariable target tracking method and system |
| CN103617630A (en) * | 2013-12-18 | 2014-03-05 | 国网重庆潼南县供电有限责任公司 | Infrared image dividing method and system for power system equipment based on wavelet analysis |
| CN104637049A (en) * | 2014-12-16 | 2015-05-20 | 北京航天时代光电科技有限公司 | Automatic detection method for optical fiber coiling quality |
| CN104483326A (en) * | 2014-12-19 | 2015-04-01 | 长春工程学院 | High-voltage wire insulator defect detection method and high-voltage wire insulator defect detection system based on deep belief network |
| CN105069807A (en) * | 2015-08-28 | 2015-11-18 | 西安工程大学 | Punched workpiece defect detection method based on image processing |
Non-Patent Citations (2)
| Title |
|---|
| 仝卫国: "基于航拍图像的输电线路识别与状态检测方法研究", 《中国博士学位论文全文数据库 信息科技辑》 * |
| 王桢: "基于视频图像处理技术的输电线路杆塔鸟巢识别", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (24)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106780639B (en) * | 2017-01-20 | 2020-08-04 | 中国海洋大学 | Hash coding method based on significance characteristic sparse embedding and extreme learning machine |
| CN106780639A (en) * | 2017-01-20 | 2017-05-31 | 中国海洋大学 | Hash coding method based on the sparse insertion of significant characteristics and extreme learning machine |
| CN107080533A (en) * | 2017-05-22 | 2017-08-22 | 北京麦迪克斯科技有限公司 | Physiological information acquisition device and method |
| CN107246959A (en) * | 2017-07-12 | 2017-10-13 | 西安因联信息科技有限公司 | The diagnostic system and method for a kind of equipment fault based on wireless senser |
| CN107246959B (en) * | 2017-07-12 | 2019-06-04 | 西安因联信息科技有限公司 | A kind of diagnostic system and method for the equipment fault based on wireless sensor |
| CN108037770A (en) * | 2017-11-22 | 2018-05-15 | 国网山东省电力公司济宁供电公司 | Unmanned plane power transmission line polling system and method based on artificial intelligence |
| CN108037770B (en) * | 2017-11-22 | 2021-05-28 | 国网山东省电力公司济宁供电公司 | Unmanned aerial vehicle power transmission line inspection system and method based on artificial intelligence |
| CN108189043A (en) * | 2018-01-10 | 2018-06-22 | 北京飞鸿云际科技有限公司 | A kind of method for inspecting and crusing robot system applied to high ferro computer room |
| CN109526119A (en) * | 2018-09-29 | 2019-03-26 | 镇江微芯光子科技有限公司 | A kind of action control sense light |
| CN109492807A (en) * | 2018-11-01 | 2019-03-19 | 大唐环境产业集团股份有限公司 | Based on the boiler NO for improving quanta particle swarm optimizationXPrediction model optimization method |
| CN109325313A (en) * | 2018-11-01 | 2019-02-12 | 大唐环境产业集团股份有限公司 | Based on improvement quantum telepotation boiler of power plant NOXPrediction model device |
| CN110287983A (en) * | 2019-05-10 | 2019-09-27 | 杭州电子科技大学 | Based on maximal correlation entropy deep neural network single classifier method for detecting abnormality |
| CN110287983B (en) * | 2019-05-10 | 2021-07-09 | 杭州电子科技大学 | Single-classifier anomaly detection method based on maximum correlation entropy deep neural network |
| CN110929611A (en) * | 2019-11-12 | 2020-03-27 | 中国地质大学（武汉） | Modeling method of power transmission line icing thickness prediction model based on PR-KELM |
| CN110888025A (en) * | 2019-11-27 | 2020-03-17 | 华东师范大学 | GIS equipment fault judgment method based on machine learning |
| CN112163573A (en) * | 2020-11-09 | 2021-01-01 | 云南电网有限责任公司迪庆供电局 | Insulator identification method and device |
| CN112464756A (en) * | 2020-11-13 | 2021-03-09 | 上海电力大学 | Image quantification method for insulator defect identification |
| CN112464756B (en) * | 2020-11-13 | 2023-05-02 | 上海电力大学 | Insulator defect identification-oriented image quantization method |
| CN112706723A (en) * | 2020-12-28 | 2021-04-27 | 石家庄开发区天远科技有限公司 | Control method and device of vehicle cleaning device and terminal equipment |
| CN112766251A (en) * | 2020-12-30 | 2021-05-07 | 广东电网有限责任公司佛山供电局 | Infrared detection method and system for power transformation equipment, storage medium and computer equipment |
| CN112991297A (en) * | 2021-03-15 | 2021-06-18 | 广东电网有限责任公司清远供电局 | Power transmission line fault diagnosis method and system, electronic equipment and storage medium |
| CN113486731A (en) * | 2021-06-17 | 2021-10-08 | 国网山东省电力公司汶上县供电公司 | Abnormal state monitoring method for power transmission equipment |
| CN114120197A (en) * | 2021-11-27 | 2022-03-01 | 中国传媒大学 | Method for detecting abnormal signals of ultra-high definition video transmitted in 2SI mode |
| CN114120197B (en) * | 2021-11-27 | 2024-03-29 | 中国传媒大学 | Ultra-high definition video abnormal signal detection method for 2SI mode transmission |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105976383A (en) | Power transmission equipment fault diagnosis method based on limit learning machine image recognition | |
| CN106023185B (en) | A kind of transmission facility method for diagnosing faults | |
| CN108573276B (en) | Change detection method based on high-resolution remote sensing image | |
| CN110264448B (en) | Insulator fault detection method based on machine vision | |
| CN104992223B (en) | Intensive population estimation method based on deep learning | |
| Zhang et al. | Deep convolutional neural networks for forest fire detection | |
| CN112233073A (en) | Real-time detection method for infrared thermal imaging abnormity of power transformation equipment | |
| CN111611874B (en) | Face mask wearing detection method based on ResNet and Canny | |
| CN110930387A (en) | Fabric defect detection method based on depth separable convolutional neural network | |
| CN107392886A (en) | A kind of power equipment image processing method | |
| CN101431664A (en) | Automatic detection method and system for intensity of passenger flow based on video image | |
| CN108564814A (en) | A kind of parking position detection method and device based on image | |
| CN111044570A (en) | Defect identification and early warning method and device for power equipment and computer equipment | |
| CN114360030B (en) | Face recognition method based on convolutional neural network | |
| CN114241511B (en) | Weak supervision pedestrian detection method, system, medium, equipment and processing terminal | |
| CN110276747B (en) | Insulator fault detection and fault rating method based on image analysis | |
| CN108734200B (en) | Human target visual detection method and device based on BING (building information network) features | |
| Zhu et al. | Towards automatic wild animal detection in low quality camera-trap images using two-channeled perceiving residual pyramid networks | |
| CN110415260A (en) | Smog image segmentation and recognition methods based on dictionary and BP neural network | |
| CN106157330A (en) | A kind of visual tracking method based on target associating display model | |
| CN116128820A (en) | Pin state identification method based on improved YOLO model | |
| CN116416237A (en) | Power transmission line defect detection method based on improved YOLOv5 and fuzzy image enhancement | |
| Yu et al. | Design of the transmission line inspection system based on UAV | |
| CN109064444B (en) | Track slab disease detection method based on significance analysis | |
| CN114359167A (en) | Insulator defect detection method based on lightweight YOLOv4 in complex scene |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20160928 |
|
| RJ01 | Rejection of invention patent application after publication |