CN107292873A - A kind of close degree detecting method of porcelain insulator ash based on color character - Google Patents
A kind of close degree detecting method of porcelain insulator ash based on color character Download PDFInfo
- Publication number
- CN107292873A CN107292873A CN201710515376.XA CN201710515376A CN107292873A CN 107292873 A CN107292873 A CN 107292873A CN 201710515376 A CN201710515376 A CN 201710515376A CN 107292873 A CN107292873 A CN 107292873A
- Authority
- CN
- China
- Prior art keywords
- mrow
- msub
- feature
- close degree
- msup
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/136—Segmentation; Edge detection involving thresholding
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of close degree detecting method of porcelain insulator ash based on color character, coloured image Channel-shifted is obtained filtering segmentation with two-dimentional minimum error method combining form after H components and extracts insulator card region, then the average of six passages in card region is extracted, maximum, minimum value, extreme difference, variance, gray scale anisotropy, 7 characteristic quantities of gray level entropy simultaneously filter out the strong feature of classification capacity with Fisher criterion functions as grey close degree differentiation feature, finally it regard the differentiation feature of training set as input, the close degree of ash is trained as output to mind evolutionary MEA Optimized BP Neural Networks, simulation and prediction and judging nicety rate are carried out with test set data, the noncontact of insulator contamination grade can be realized, online efficient detection.A kind of close degree detecting method of porcelain insulator ash based on color character of the present invention, solve existing grey close degree detecting method can not on-line checking the problem of.
Description
Technical field
The invention belongs to transmission line of electricity technical field of image processing, and in particular to a kind of porcelain insulation based on color character
The grey close degree detecting method of son.
Background technology
Insulator is that support and the important component of insulating effect, its working condition direct relation are played in ultra-high-tension power transmission line
To the safe and stable operation of whole power system.Due to the long term by pollutants such as dust in air, insulator surface
Pollution layer can be formed.Pollution layer, which absorbs moisture, when air humidity is larger causes insulator external insulating strength to be remarkably decreased, and easily sends out
Raw pollution flashover accident, causes large-area power-cuts.It is how accurate, easy, to be reliably achieved insulator contamination detection be that power department is closed
The major issue of the heart, is also the study hotspot of current academia.
At present, insulator contamination detection method mainly has equivalent salt deposit density method, leakage current method and IR thermometry etc..
Equivalent salt deposit density method needs to have a power failure due to filth sampling, and operating process is complicated, it is difficult to realize the efficient, online of gradation for surface pollution
Detection.Because the leakage current under low environment damp condition and fever phenomenon be not obvious, therefore leakage current method and infrared measurement of temperature
Method both approaches need to use under the conditions of higher levels of humidity.Have and need not stop using visible images detection insulator contamination
Electricity, non-cpntact measurement, not by the advantage of the such environmental effects such as temperature and humidity, the difference of the close degree of insulator surface ash is in figure
The difference of card regional color is shown as on picture, so can be for judgement according to the difference of insulator card area image color
The close degree of insulator ash.
The content of the invention
It is an object of the invention to provide a kind of close degree detecting method of porcelain insulator ash based on color character, solve
Existing grey close degree detecting method can not noncontact, online efficient detection the problem of.
The technical solution adopted in the present invention is, a kind of close degree detecting side of porcelain insulator ash based on color character
Method, coloured image Channel-shifted is obtained filtering segmentation with two-dimentional minimum error method combining form after H components extracts insulation
Sub-disk face region, then the average of extraction six passages in card region, maximum, minimum value, extreme difference, variance, gray scale are respectively to different
Property, 7 characteristic quantities of gray level entropy and the strong feature of classification capacity filtered out with Fisher criterion functions as grey close degree differentiate special
Levy, finally using the differentiation feature of training set as input, grey close degree optimizes BP nerves to mind evolutionary MEA as output
Network is trained, and simulation and prediction and judging nicety rate are carried out with test set data, it is possible to achieve the close degree of insulator ash it is non-
Contact, online efficient detection.
The features of the present invention is also resided in：
A kind of close degree detecting method of porcelain insulator ash based on color character, specifically implements according to following steps：
Step 1：By the insulator coloured image channel decomposition collected and change, with two-dimentional minimum error method to H components
Carry out segmentation and extract insulation subregion, the hole between interfered cell domain and insulator is then eliminated using morphologic filtering；
Step 2：The averages of six passages in insulator card region, maximum, minimum value, extreme difference, variance, ash are extracted respectively
Anisotropy, 7 features of gray level entropy are spent as primitive character amount, and classification capacity is then filtered out using Fisher criterion functions
Stronger feature differentiates feature as grey close degree；
Step 3：It regard the differentiation feature of the N number of training sample obtained in step 2 as mind evolutionary optimization BP nerves
The input of network, grey close degree is as output, and training obtains the close degree of visible ray ash and differentiates neutral net；
Step 4：Emulation testing is carried out to the MEA Optimized BP Neural Networks obtained in step 3, by the M obtained in step 2
The differentiation feature of test sample is as input, the close degree of differentiation ash exported.
Two-dimentional minimum error method formula is as follows in step 1：
Wherein, P0(s, t), P1(s, t) represents prior probability, δ00(s, t), δ01(s, t), δ10(s, t), δ11(s, t) is represented
Variance of Normal Distribution, ρ0(s, t), ρ1(s, t) represents coefficient correlation；
Optimal threshold is obtained when taking minimum value：
Image has some interfered cell domains after splitting in step 1, and has certain hole between sub-pieces, after segmentation
Region carry out connected component labeling, by the way that white pixel in bianry image is marked, allow each single connected region shape
Into an identified block, the geometric parameters such as profile, boundary rectangle, barycenter, the area of these blocks are further obtained, pass through area
Parameter removes interfered cell domain, filters out insulator region, then image is handled using morphologic filtering, obtain
Clean insulator card region.
Step 2 is specially：
It is six passages of R, G, B, H, S, I by insulator card Region Decomposition, average, the maximum of six passages is extracted respectively
Value, minimum value, extreme difference, variance, gray scale anisotropy, 7 features of gray level entropy can obtain 42 altogether as primitive character amount
Individual characteristic quantity, is then screened using Fisher criterion functions, and Fisher criterion functions screening principle is that classification capacity is strong
Feature variance within clusters should be as far as possible small, and inter-class variance should try one's best greatly, and functional value is bigger, and the separability for representing this feature is better；
Define a shared N number of sample in data set Ω and be belonging respectively to n class L1,L2,...Ln, N is included per classiIndividual sample,WithThe respectively inter-class variance and variance within clusters of the jth dimensional feature of sample, expression formula is as follows：
In above formula,Represent the average of the i-th class sample jth dimensional feature, mjRepresent the equal of whole sample jth dimensional features
Value, ajRepresent sample a jth dimensional feature；
The Fisher criterion functions of single feature are：
The Fisher criterion function values F of certain dimensional feature is bigger, represents that the separability of the dimensional feature is better；
Screening F values maximum three characteristic quantity S averages, S maximums and S variances differentiate that feature is come to ash as grey close degree
Close degree is classified.
Step 3 is specially：
Step 3.1：Input and the output data of N number of training sample are obtained, and input data is normalized；
Step 3.2：According to the topological structure 3-7-5 of BP neural network, solution space is mapped to space encoder, code length
S=S1 × S2+S2 × S3+S2+S3, wherein S1 are input layer number, and S2 is node in hidden layer, and S3 is output node layer
Number, code length is the total number of weights and threshold value；
Step 3.3：It is 20 to set iterations, and Population Size is 200, winning sub- Population Size and interim sub- Population Size
It is 5, defines a scoring function val=1/mse (T-Tout), wherein T is desired output, ToutFor output layer after each iteration
Output valve, SE is mean square error；The 5 winning individuals and 5 temporary individuals of highest scoring are filtered out and with this according to score
Centered on produce 5 winning sub- populations and 5 interim sub- populations；
Step 3.4：Operation similartaxis is performed inside each sub-group until the subgroup body maturation, is calculated in the sub-group most
Excellent individual score, the score of optimum individual in each sub-group is puted up on global advertisement plate, performed between sub-group different
Change operation, so as to obtain global optimum's individual and its score；
Step 3.5：If being unsatisfactory for iteration termination condition, return to step 3.4 is continued executing with, continuous iteration, until iteration knot
Optimum individual is exported after beam and is decoded, the initial weight and threshold value of BP neural network is produced, is then produced using in step 3.1
Raw input and output data is trained to BP neural network, is obtained grey close degree and is differentiated neutral net.
The beneficial effects of the invention are as follows：Porcelain insulator ash close degree detecting method of the present invention based on color character, will
The insulation subgraph collected is converted to after H component images, it is contemplated that minimum error method influenceed small, precision high by target sizes and
Fireballing advantage, two-dimensional histogram can make full use of image information, by the small advantage of noise jamming, be missed using two-dimentional minimum
Poor method carries out segmentation to image and extracts insulation subregion, and combining form filtering is removed between interfered cell domain and insulator
Hole, obtains the insulator card region of complete display, is conducive to follow-up color feature extracted.Calculate respectively in every grade of image
The average of six passages, maximum, minimum value, extreme difference, variance, gray scale anisotropy, 7 features of gray level entropy are used as primitive character
Amount, feature is differentiated using the stronger feature of Fisher criterion selection sort abilities as grey close degree.Mind evolutionary (MEA)
Optimized BP Neural Network can remember the evolution information of a more than generation, instruct convergent and alienation to be carried out towards favourable direction, profit
Optimize the initial weight and threshold value in BP neutral nets with its global search, have than single BP neural network precision of prediction
Large increase.
Brief description of the drawings
Fig. 1 is the flow chart of the grey close degree detecting method of the present invention；
Fig. 2 is different grey close degree insulator image three-dimensional characteristic profiles in the grey close degree detecting method of the present invention；
Fig. 3 is MEA Optimized BP Neural Network flow charts in the grey close degree detecting method of the present invention.
Embodiment
The present invention is described in detail with reference to the accompanying drawings and detailed description.
A kind of close degree detecting method of porcelain insulator ash based on color character of the present invention, by coloured image Channel-shifted
Obtain filtering segmentation with two-dimentional minimum error method combining form after H components and extract insulator card region, then extract disk
The average of six passages in face region, maximum, minimum value, extreme difference, variance, gray scale anisotropy, 7 characteristic quantities of gray level entropy are used in combination
Fisher criterion functions filter out the strong feature of classification capacity as grey close degree and differentiate feature, finally that the differentiation of training set is special
Levy as input, grey close degree is trained as output to mind evolutionary MEA Optimized BP Neural Networks, uses test set number
According to progress simulation and prediction and judging nicety rate, it is possible to achieve the noncontact of insulator contamination grade, online efficient detection.
A kind of close degree detecting method of porcelain insulator ash based on color character of the present invention, as shown in figure 1, it is specific according to
Following steps are implemented：
Step 1：By the insulator coloured image channel decomposition collected and change, with two-dimentional minimum error method to H components
Carry out segmentation and extract insulation subregion, the hole between interfered cell domain and insulator is then eliminated using morphologic filtering；
The insulation subgraph collected is converted to after H component images, it is contemplated that the factor of insulator image capturing environment,
Because minimum error method is influenceed that small, precision is high, speed is fast and two-dimensional histogram can make full use of image to believe by target sizes
It is breath, small by noise jamming, therefore use two dimension minimum error method to split the insulation subgraph collected and extract insulator
Region.Two-dimentional minimum error method formula is as follows：
Wherein, P0(s, t), P1(s, t) represents prior probability, δ00(s, t), δ01(s, t), δ10(s, t), δ11(s, t) is represented
Variance of Normal Distribution, ρ0(s, t), ρ1(s, t) represents coefficient correlation；
Optimal threshold is obtained when taking minimum value：
Image has some interfered cell domains after segmentation, and has certain hole between sub-pieces, therefore to the area after segmentation
Domain carries out connected component labeling, by the way that white pixel in bianry image (target) is marked, and allows each single connected region
Form an identified block, it is possible to further obtain the geometric parameters such as profile, boundary rectangle, barycenter, the area of these blocks,
The present invention removes interfered cell domain by area parameters, filters out insulator region, then using morphologic filtering to figure
As being handled, clean insulator card region is obtained.
Step 2：The averages of six passages in insulator card region, maximum, minimum value, extreme difference, variance, ash are extracted respectively
Spend anisotropy, 7 features of gray level entropy as primitive character amount, then using Fisher criterion functions filter out classification capacity compared with
Strong feature differentiates feature as grey close degree；
It is six passages of R, G, B, H, S, I by insulator card Region Decomposition, average, the maximum of six passages is extracted respectively
Value, minimum value, extreme difference, variance, gray scale anisotropy, 7 features of gray level entropy can obtain 42 altogether as primitive character amount
Individual characteristic quantity, is then screened, its general principle is the strong feature variance within clusters of classification capacity using Fisher criterion functions
Should be as far as possible small, inter-class variance should try one's best greatly, and functional value is bigger, and the separability for representing this feature is better.Define in data set Ω altogether
There is N sample to be belonging respectively to n class L1,L2,...Ln, N is included per classiIndividual sample,WithThe respectively jth Wei Te of sample
The inter-class variance and variance within clusters levied, expression formula are as follows：
In above formula,Represent the average of the i-th class sample jth dimensional feature, mjThe average of whole sample jth dimensional features is represented,
ajRepresent sample a jth dimensional feature.The Fisher criterion functions of single feature are：
The Fisher criterion function values F of certain dimensional feature is bigger, represents that the separability of the dimensional feature is better.The present invention is filtered out
F values maximum three characteristic quantity S averages, S maximums and S variances differentiate that feature is divided grey close degree as grey close degree
Class.Different grey close degree insulator image three-dimensional characteristic profiles are as shown in Figure 2, it can be seen that do not have between five close degree of ash
Occur simultaneously, reached preferable separating effect.
Step 3：Optimize the differentiation feature of obtained in step 2 200 training samples as mind evolutionary (MEA)
The input of BP neural network, grey close degree is as output, and training obtains the close degree of visible ray ash and differentiates neutral net, and flow chart is such as
Shown in Fig. 3, specifically implement according to following steps：
Step 3.1：The input of 200 training samples of acquisition and output data, in order to which less variable-difference is larger to model
The influence of performance, input data is normalized；
Step 3.2：According to the topological structure 3-7-5 of BP neural network, solution space is mapped to space encoder, code length
(total number of weights and threshold value) S=S1 × S2+S2 × S3+S2+S3, wherein S1 are input layer number, and S2 is hidden layer section
Points, S3 is output layer nodes；
Step 3.3：It is 20 to set iterations, and Population Size is 200, winning sub- Population Size and interim sub- Population Size
It is 5, defines a scoring function val=1/mse (T-Tout), wherein T is desired output, ToutTo be exported after each iteration
The output valve of layer, SE is mean square error.According to score filter out 5 of highest scoring winning individuals and 5 temporary individuals and with
5 winning sub- populations and 5 interim sub- populations are produced centered on this；
Step 3.4：Operation similartaxis is performed inside each sub-group until the subgroup body maturation, is calculated in the sub-group most
Excellent individual score, the score of optimum individual in each sub-group is puted up on global advertisement plate, performed between sub-group different
Change operation, so as to obtain global optimum's individual and its score；
Step 3.5：If being unsatisfactory for iteration termination condition, return to step 3.4 is continued executing with, continuous iteration, until iteration knot
Optimum individual is exported after beam and is decoded, the initial weight and threshold value of BP neural network is produced, is then produced using in step 3.1
Raw input and output data is trained to BP neural network, is obtained grey close degree and is differentiated neutral net.
Neutral net is differentiated as grey close degree using mind evolutionary (MEA) Optimized BP Neural Network in step 3,
Colony is divided into winning sub-group and interim sub-group by MEA algorithms, is defined convergent and operation dissimilation and is carried out structure optimization, in office
Locally optimal solution is searched in portion space, dissimilation operator search globally optimal solution in whole solution is then recycled, it is global using it
Initial weight and threshold value in search property Optimized BP Neural Network, improve a lot than single BP neural network precision of prediction,
With certain feasibility.
Step 4：Emulation testing is carried out to the MEA Optimized BP Neural Networks obtained in step 3, by obtained in step 2 100
The differentiation feature of individual test sample is as input, the close degree of differentiation ash exported.
Claims (6)
1. a kind of close degree detecting method of porcelain insulator ash based on color character, it is characterised in that by coloured image passage
It is converted to after H components and extracts insulator card region, Ran Houti with two-dimentional minimum error method combining form filtering segmentation
Take the averages of six passages in card region, maximum, minimum value, extreme difference, variance, gray scale anisotropy, 7 characteristic quantities of gray level entropy
And the strong feature of classification capacity is filtered out with Fisher criterion functions as grey close degree differentiate feature, last sentencing training set
Other feature is as input, and grey close degree is trained as output to mind evolutionary MEA Optimized BP Neural Networks, with test
Collect data and carry out simulation and prediction and judging nicety rate, it is possible to achieve the noncontact of insulator contamination grade, online efficient detection.
2. a kind of close degree detecting method of porcelain insulator ash based on color character according to claim 1, its feature
It is, specifically implements according to following steps：
Step 1：By the insulator coloured image channel decomposition collected and change, H components are carried out with two-dimentional minimum error method
Segmentation extracts insulation subregion, then eliminates the hole between interfered cell domain and insulator using morphologic filtering；
Step 2：Average, maximum, minimum value, extreme difference, variance, the gray scale for extracting six passages in insulator card region respectively are each
Then anisotropy, 7 features of gray level entropy filter out classification capacity stronger as primitive character amount using Fisher criterion functions
Feature differentiates feature as grey close degree；
Step 3：It regard the differentiation feature of the N number of training sample obtained in step 2 as mind evolutionary Optimized BP Neural Network
Input, grey close degree is as output, and training obtains the close degree of visible ray ash and differentiates neutral net；
Step 4：Emulation testing is carried out to the MEA Optimized BP Neural Networks obtained in step 3, by M obtained in step 2 test
The differentiation feature of sample is as input, the close degree of differentiation ash exported.
3. a kind of close degree detecting method of porcelain insulator ash based on color character according to claim 2, its feature
It is, two-dimentional minimum error method formula is as follows in the step 1：
<mfenced open = "" close = "">
<mtable>
<mtr>
<mtd>
<mrow>
<mi>J</mi>
<mrow>
<mo>(</mo>
<mi>s</mi>
<mo>,</mo>
<mi>t</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mn>1</mn>
<mo>-</mo>
<msub>
<mi>P</mi>
<mn>0</mn>
</msub>
<mrow>
<mo>(</mo>
<mi>s</mi>
<mo>,</mo>
<mi>t</mi>
<mo>)</mo>
</mrow>
<mi>ln</mi>
<mi> </mi>
<msub>
<mi>P</mi>
<mn>0</mn>
</msub>
<mrow>
<mo>(</mo>
<mi>s</mi>
<mo>,</mo>
<mi>t</mi>
<mo>)</mo>
</mrow>
<mo>-</mo>
<msub>
<mi>P</mi>
<mn>1</mn>
</msub>
<mrow>
<mo>(</mo>
<mi>s</mi>
<mo>,</mo>
<mi>t</mi>
<mo>)</mo>
</mrow>
<mi>ln</mi>
<mi> </mi>
<msub>
<mi>P</mi>
<mn>1</mn>
</msub>
<mrow>
<mo>(</mo>
<mi>s</mi>
<mo>,</mo>
<mi>t</mi>
<mo>)</mo>
</mrow>
<mo>+</mo>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<msub>
<mi>P</mi>
<mn>0</mn>
</msub>
<mrow>
<mo>(</mo>
<mi>s</mi>
<mo>,</mo>
<mi>t</mi>
<mo>)</mo>
</mrow>
<msub>
<mi>ln&delta;</mi>
<mn>00</mn>
</msub>
<mrow>
<mo>(</mo>
<mi>s</mi>
<mo>,</mo>
<mi>t</mi>
<mo>)</mo>
</mrow>
<msub>
<mi>&delta;</mi>
<mn>01</mn>
</msub>
<mrow>
<mo>(</mo>
<mi>s</mi>
<mo>,</mo>
<mi>t</mi>
<mo>)</mo>
</mrow>
<mo>+</mo>
<msub>
<mi>P</mi>
<mn>1</mn>
</msub>
<mrow>
<mo>(</mo>
<mi>s</mi>
<mo>,</mo>
<mi>t</mi>
<mo>)</mo>
</mrow>
<msub>
<mi>ln&delta;</mi>
<mn>10</mn>
</msub>
<mrow>
<mo>(</mo>
<mi>s</mi>
<mo>,</mo>
<mi>t</mi>
<mo>)</mo>
</mrow>
<msub>
<mi>&delta;</mi>
<mn>11</mn>
</msub>
<mrow>
<mo>(</mo>
<mi>s</mi>
<mo>,</mo>
<mi>t</mi>
<mo>)</mo>
</mrow>
<mo>+</mo>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<msub>
<mi>P</mi>
<mn>0</mn>
</msub>
<mrow>
<mo>(</mo>
<mi>s</mi>
<mo>,</mo>
<mi>t</mi>
<mo>)</mo>
</mrow>
<mi>ln</mi>
<msqrt>
<mrow>
<mn>1</mn>
<mo>-</mo>
<msup>
<msub>
<mi>&rho;</mi>
<mn>0</mn>
</msub>
<mn>2</mn>
</msup>
<mrow>
<mo>(</mo>
<mi>s</mi>
<mo>,</mo>
<mi>t</mi>
<mo>)</mo>
</mrow>
</mrow>
</msqrt>
<mo>+</mo>
<msub>
<mi>P</mi>
<mn>1</mn>
</msub>
<mrow>
<mo>(</mo>
<mi>s</mi>
<mo>,</mo>
<mi>t</mi>
<mo>)</mo>
</mrow>
<mi>ln</mi>
<msqrt>
<mrow>
<mn>1</mn>
<mo>-</mo>
<msup>
<msub>
<mi>&rho;</mi>
<mn>1</mn>
</msub>
<mn>2</mn>
</msup>
<mrow>
<mo>(</mo>
<mi>s</mi>
<mo>,</mo>
<mi>t</mi>
<mo>)</mo>
</mrow>
</mrow>
</msqrt>
</mrow>
</mtd>
</mtr>
</mtable>
</mfenced>
Wherein, P0(s, t), P1(s, t) represents prior probability, δ00(s, t), δ01(s, t), δ10(s, t), δ11(s, t) represents normal state
Distribution variance, ρ0(s, t), ρ1(s, t) represents coefficient correlation；
Optimal threshold is obtained when taking minimum value：
<mrow>
<mo>(</mo>
<msup>
<mi>s</mi>
<mo>*</mo>
</msup>
<mo>,</mo>
<msup>
<mi>t</mi>
<mo>*</mo>
</msup>
<mo>)</mo>
<mo>=</mo>
<mi>arg</mi>
<munder>
<mrow>
<mi>m</mi>
<mi>i</mi>
<mi>n</mi>
</mrow>
<mrow>
<mn>0</mn>
<mo><</mo>
<mi>s</mi>
<mo>,</mo>
<mi>t</mi>
<mo><</mo>
<mi>L</mi>
<mo>-</mo>
<mn>1</mn>
</mrow>
</munder>
<mi>J</mi>
<mo>(</mo>
<mi>s</mi>
<mo>,</mo>
<mi>t</mi>
<mo>)</mo>
<mo>.</mo>
</mrow>
4. a kind of close degree detecting method of porcelain insulator ash based on color character according to claim 2, its feature
It is, image there are some interfered cell domains after splitting in the step 1, and has certain hole between sub-pieces, to segmentation
Rear region carries out connected component labeling, by the way that white pixel in bianry image is marked, and allows each single connected region
An identified block is formed, the geometric parameters such as profile, boundary rectangle, barycenter, the area of these blocks is further obtained, passes through face
Product parameter removes interfered cell domain, filters out insulator region, then image is handled using morphologic filtering, obtain
To clean insulator card region.
5. a kind of close degree detecting method of porcelain insulator ash based on color character according to claim 2, its feature
It is, the step 2 is specially：
Be six passages of R, G, B, H, S, I by insulator card Region Decomposition, extract respectively the averages of six passages, maximum,
Minimum value, extreme difference, variance, gray scale anisotropy, 7 features of gray level entropy can obtain 42 altogether as primitive character amount
Characteristic quantity, is then screened using Fisher criterion functions, and Fisher criterion functions screening principle is the strong spy of classification capacity
Levying variance within clusters should be as far as possible small, and inter-class variance should try one's best greatly, and functional value is bigger, and the separability for representing this feature is better；
Define a shared N number of sample in data set Ω and be belonging respectively to n class L1,L2,...Ln, N is included per classiIndividual sample,WithThe respectively inter-class variance and variance within clusters of the jth dimensional feature of sample, expression formula is as follows：
<mrow>
<msubsup>
<mi>S</mi>
<mi>X</mi>
<mi>j</mi>
</msubsup>
<mo>=</mo>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>n</mi>
</munderover>
<mfrac>
<msub>
<mi>N</mi>
<mi>i</mi>
</msub>
<mi>N</mi>
</mfrac>
<msup>
<mrow>
<mo>(</mo>
<msubsup>
<mi>m</mi>
<mi>i</mi>
<mi>j</mi>
</msubsup>
<mo>-</mo>
<msup>
<mi>m</mi>
<mi>j</mi>
</msup>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
</mrow>
<mrow>
<msubsup>
<mi>S</mi>
<mi>Y</mi>
<mi>j</mi>
</msubsup>
<mo>=</mo>
<mfrac>
<mn>1</mn>
<mi>N</mi>
</mfrac>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>n</mi>
</munderover>
<munder>
<mo>&Sigma;</mo>
<mrow>
<mi>a</mi>
<mo>&Element;</mo>
<msub>
<mi>L</mi>
<mi>i</mi>
</msub>
</mrow>
</munder>
<msup>
<mrow>
<mo>(</mo>
<msup>
<mi>a</mi>
<mi>j</mi>
</msup>
<mo>-</mo>
<msubsup>
<mi>m</mi>
<mi>i</mi>
<mi>j</mi>
</msubsup>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
</mrow>
In above formula,Represent the average of the i-th class sample jth dimensional feature, mjRepresent the average of whole sample jth dimensional features, ajTable
This of sample jth dimensional feature；
The Fisher criterion functions of single feature are：
<mrow>
<mi>F</mi>
<mo>=</mo>
<mfrac>
<msubsup>
<mi>S</mi>
<mi>X</mi>
<mi>j</mi>
</msubsup>
<msubsup>
<mi>S</mi>
<mi>Y</mi>
<mi>j</mi>
</msubsup>
</mfrac>
</mrow>
The Fisher criterion function values F of certain dimensional feature is bigger, represents that the separability of the dimensional feature is better；
Screening F values maximum three characteristic quantity S averages, S maximums and S variances differentiate that feature is come to grey close journey as grey close degree
Degree is classified.
6. a kind of close degree detecting method of porcelain insulator ash based on color character according to claim 2, its feature
It is, the step 3 is specially：
Step 3.1：Input and the output data of N number of training sample are obtained, and input data is normalized；
Step 3.2：According to the topological structure 3-7-5 of BP neural network, solution space is mapped to space encoder, code length S=
S1 × S2+S2 × S3+S2+S3, wherein S1 are input layer number, and S2 is node in hidden layer, and S3 is output layer nodes, is compiled
Code length is the total number of weights and threshold value；
Step 3.3：It is 20 to set iterations, and Population Size is 200, and winning sub- Population Size and interim sub- Population Size are
5, define a scoring function val=1/mse (T-Tout), wherein T is desired output, ToutFor after each iteration output layer it is defeated
Go out value, SE is mean square error；The 5 winning individuals and 5 temporary individuals of highest scoring are filtered out and as according to score
The heart produces 5 winning sub- populations and 5 interim sub- populations；
Step 3.4：Operation similartaxis is performed inside each sub-group until the subgroup body maturation, calculates in the sub-group optimal
The score of body, the score of optimum individual in each sub-group is puted up on global advertisement plate, and alienation behaviour is performed between sub-group
Make, so as to obtain global optimum's individual and its score；
Step 3.5：If being unsatisfactory for iteration termination condition, return to step 3.4 is continued executing with, continuous iteration, after iteration terminates
Output optimum individual is simultaneously decoded, and produces the initial weight and threshold value of BP neural network, then using producing in step 3.1
Input and output data are trained to BP neural network, are obtained grey close degree and are differentiated neutral net.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710515376.XA CN107292873A (en) | 2017-06-29 | 2017-06-29 | A kind of close degree detecting method of porcelain insulator ash based on color character |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710515376.XA CN107292873A (en) | 2017-06-29 | 2017-06-29 | A kind of close degree detecting method of porcelain insulator ash based on color character |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107292873A true CN107292873A (en) | 2017-10-24 |
Family
ID=60099173
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710515376.XA Pending CN107292873A (en) | 2017-06-29 | 2017-06-29 | A kind of close degree detecting method of porcelain insulator ash based on color character |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107292873A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108345898A (en) * | 2017-12-31 | 2018-07-31 | 国网山西省电力公司检修分公司 | A kind of novel line insulator Condition assessment of insulation method |
| CN110781901A (en) * | 2019-10-29 | 2020-02-11 | 湖北工业大学 | Instrument ghost character recognition method based on BP neural network prediction threshold |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103411980A (en) * | 2013-07-23 | 2013-11-27 | 同济大学 | External insulation filth status identification method based on visible-light images |
| CN105445283A (en) * | 2016-02-01 | 2016-03-30 | 成都通甲优博科技有限责任公司 | Detection method for filthy conditions of insulator images |
| CN106769030A (en) * | 2016-11-10 | 2017-05-31 | 浙江工业大学 | A kind of bearing state tracking and Forecasting Methodology based on MEA BP neural network algorithms |
-
2017
- 2017-06-29 CN CN201710515376.XA patent/CN107292873A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103411980A (en) * | 2013-07-23 | 2013-11-27 | 同济大学 | External insulation filth status identification method based on visible-light images |
| CN105445283A (en) * | 2016-02-01 | 2016-03-30 | 成都通甲优博科技有限责任公司 | Detection method for filthy conditions of insulator images |
| CN106769030A (en) * | 2016-11-10 | 2017-05-31 | 浙江工业大学 | A kind of bearing state tracking and Forecasting Methodology based on MEA BP neural network algorithms |
Non-Patent Citations (3)
| Title |
|---|
| 张以帅 等: "基于 MEA 优化 BP 神经网络的天然气短期负荷预测", 《自动化与仪表》 * |
| 田治仁 等: "基于彩色可见光图像的绝缘子污秽等级判别", 《电工电能新技术》 * |
| 范九伦 等: "灰度图像最小误差阈值分割法的二维推广", 《自动化学报》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108345898A (en) * | 2017-12-31 | 2018-07-31 | 国网山西省电力公司检修分公司 | A kind of novel line insulator Condition assessment of insulation method |
| CN110781901A (en) * | 2019-10-29 | 2020-02-11 | 湖北工业大学 | Instrument ghost character recognition method based on BP neural network prediction threshold |
| CN110781901B (en) * | 2019-10-29 | 2023-04-28 | 湖北工业大学 | Instrument ghost character recognition method based on BP neural network prediction threshold |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103886344B (en) | A kind of Image Fire Flame recognition methods | |
| CN109034210A (en) | Object detection method based on super Fusion Features Yu multi-Scale Pyramid network | |
| CN109389180A (en) | A power equipment image-recognizing method and inspection robot based on deep learning | |
| CN110322453A (en) | 3D point cloud semantic segmentation method based on position attention and auxiliary network | |
| CN106650913A (en) | Deep convolution neural network-based traffic flow density estimation method | |
| CN109977790A (en) | A kind of video smoke detection and recognition methods based on transfer learning | |
| CN107247938A (en) | A kind of method of high-resolution remote sensing image City Building function classification | |
| CN107316058A (en) | Improve the method for target detection performance by improving target classification and positional accuracy | |
| CN106971152A (en) | A kind of method of Bird's Nest in detection transmission line of electricity based on Aerial Images | |
| CN105825502B (en) | A kind of Weakly supervised method for analyzing image of the dictionary study based on conspicuousness guidance | |
| CN107945153A (en) | A kind of road surface crack detection method based on deep learning | |
| CN105160310A (en) | 3D (three-dimensional) convolutional neural network based human body behavior recognition method | |
| CN109559310A (en) | Power transmission and transformation inspection image quality evaluating method and system based on conspicuousness detection | |
| CN107341480A (en) | A kind of crowd massing detection method of modified PCCNN neural network models | |
| CN107808358A (en) | Image watermark automatic testing method | |
| CN105046197A (en) | Multi-template pedestrian detection method based on cluster | |
| CN106991666A (en) | A kind of disease geo-radar image recognition methods suitable for many size pictorial informations | |
| CN110163213A (en) | Remote sensing image segmentation method based on disparity map and multiple dimensioned depth network model | |
| CN105808665A (en) | Novel hand-drawn sketch based image retrieval method | |
| CN109657569A (en) | More vegetation areas transmission of electricity corridor hidden danger point quick extraction method based on cloud analysis | |
| CN103679214B (en) | Vehicle checking method based on online Class area estimation and multiple features Decision fusion | |
| CN110390673A (en) | Cigarette automatic testing method based on deep learning under a kind of monitoring scene | |
| CN108197731A (en) | It is a kind of based on jointly trained telemetering motor vehicle tail and car test result coherence method | |
| CN103440501A (en) | Scene classification method based on nonparametric space judgment hidden Dirichlet model | |
| CN105046259A (en) | Coronal mass ejection (CME) detection method based on multi-feature fusion |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20171024 |
|
| RJ01 | Rejection of invention patent application after publication |