CN108876788A - Insulator conspicuousness detection method based on the fusion of multiple dimensioned reconstructed error - Google Patents
Insulator conspicuousness detection method based on the fusion of multiple dimensioned reconstructed error Download PDFInfo
- Publication number
- CN108876788A CN108876788A CN201810611259.8A CN201810611259A CN108876788A CN 108876788 A CN108876788 A CN 108876788A CN 201810611259 A CN201810611259 A CN 201810611259A CN 108876788 A CN108876788 A CN 108876788A
- Authority
- CN
- China
- Prior art keywords
- reconstructed error
- super
- pixel
- image block
- pixel image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
-
- 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
Abstract
The invention discloses a kind of insulator conspicuousness detection methods based on the fusion of multiple dimensioned reconstructed error, including：Super-pixel segmentation is carried out to insulation subgraph, extracts the super-pixel image block of insulation subgraph surrounding as background template；Calculate sparse reconstructed error and dense reconstructed error corresponding to each super-pixel image block of insulation subgraph；Using the reconstructed error mechanism of transmission based on image context information, propagate sparse reconstructed error and dense reconstructed error between the super-pixel image block in each class；The reconstructed error obtained in the case of different scale super-pixel segmentation is weighted fusion；Sparse reconstructed error and dense reconstructed error are merged using condition random field, generate insulator notable figure；The method of the present invention significantly improves the accuracy of isolator detecting under complex environment.
Description
Technical field
The present invention relates to a kind of insulator conspicuousness detection methods based on the fusion of multiple dimensioned reconstructed error, belong to technology neck
Domain.
Background technique
In recent years, the reliability and operating condition for ensureing transmission line of electricity become the important content for building smart grid.Power transformation
The safe operation of equipment is to ensure the premise of power system stability and safety.Insulator is indispensable as electric power transmission line
Insulation component, its operation conditions directly affect the reliability and safety of power grid.Insulator plays in transmission line of electricity simultaneously
The effect of electric insulation and support；And the problems such as filth, crackle, breakage on its surface, seriously threatens the safety fortune of transmission line of electricity
Row.According to statistics, the highest accident of proportion is as caused by defects of insulator in electric power system fault at present.Therefore to exhausted
The situation of edge is monitored, and it is particularly important to complete fault diagnosis in time.
Currently, traditional manual inspection mode needs staff's climbing to high voltage iron tower in terms of power transmission line inspection
In equipment, judge whether equipment faulty by way of eye-observation, this routine inspection mode can not adapt at all by
Gradually increased transmission line of electricity and huge inspection work requirements.In recent years, it is automatic to become intelligent substation for automatic crusing robot
The major way of inspection.A large amount of insulator image information is obtained using the camera loaded on platform, if to these seas
Spirogram picture uses the interpretation of staff's naked eyes, not only heavy workload, is easy to happen and fails to judge and misjudgment phenomenon, and is difficult to accurately send out
Security risk existing for existing insulator.And realize that the important prerequisite that failure detects automatically is the insulation identified and positioned in image
Son.Therefore the automatic testing method of research insulator is very important.
Although there is also certain for these methods currently, there are some automatic identifying methods in relation to insulator
Disadvantage.Such as：Since substation's image background is complicated, there are other power equipments similar with insulator shape, such as electricity in image
Current transformer and arrester etc. are easy to produce the result of misrecognition.
Summary of the invention
It is an object of the invention to overcome deficiency in the prior art, provide it is a kind of based on multiple dimensioned reconstructed error fusion
Insulator conspicuousness detection method solves the inaccurate technical problem of isolator detecting in the prior art.
In order to solve the above technical problems, the technical scheme adopted by the invention is that：Based on the fusion of multiple dimensioned reconstructed error
Insulator conspicuousness detection method, includes the following steps：
Super-pixel segmentation is carried out to insulation subgraph, extracts the super-pixel image block of insulation subgraph surrounding as background mould
Plate；
Calculate sparse reconstructed error and dense reconstructed error corresponding to each super-pixel image block of insulation subgraph；
Using the reconstructed error mechanism of transmission based on image context information, make sparse reconstructed error and dense reconstructed error
It is propagated between the super-pixel image block in each class；
The reconstructed error obtained in the case of different scale super-pixel segmentation is weighted fusion；
Sparse reconstructed error and dense reconstructed error are merged using condition random field, generate insulator notable figure；
Further, super-pixel segmentation is carried out to insulation subgraph using simple linear iterative algorithm, specifically included as follows
Step：
Initialization cluster and measuring similarity are carried out to original image；
Iteration cluster centre is updated using K mean cluster method；
It assigns most like cluster centre label to pixel, forms super-pixel image block.
Further, sparse reconstructed error is sought using sparse representation method, the specific method is as follows：
Calculate the corresponding sparse coefficient of each super-pixel image block；
Calculate sparse reconstructed error of each super-pixel image block on background template；
Sparse reconstructed error is normalized.
Further, dense reconstructed error is sought using Principal Component Analysis, the specific method is as follows：
Calculate the covariance matrix of background template：
Eigenvalues Decomposition is carried out to covariance matrix, according to the size of characteristic value, selected characteristic vector；
Super-pixel image block is projected in feature vector, the dense expression coefficient of corresponding super-pixel image block is obtained；
It is calculated using dense expression coefficient and obtains dense reconstructed error corresponding to each super-pixel image block；
Dense reconstructed error is subjected to linear normalization.
Further, sparse reconstructed error and dense reconstructed error are propagated between the super-pixel image block in each class
The specific method is as follows：
Using K mean cluster algorithm, obtained super-pixel image block is divided into K class；
Reconstructed error is updated, is specifically comprised the following steps：
According to reconstructed error size, descending arranges reconstructed error；
Calculate separately the reconstructed error after propagating：Assuming that i-th of super-pixel image block of insulator to be measured belongs to k-th of class
Not, according to the reconstructed error of other super-pixel image blocks in this kind, which is passed by contextual information
Reconstructed error after broadcasting is defined as follows：
Wherein, [k1,k2,k3,…,kNc] Nc super-pixel image block in kth classification is represented, τ is above formula(1- τ) εiCoefficient of balance；Representative and i-th of super-pixel image block belong to of a sort
For other super-pixel image blocks by the weighted average of reconstructed error after propagating, what j was represented is j-th to surpass picture in k-th of classification
Plain image block；εiFor i-th of super-pixel image block reconstructed error obtained in previous step；For in same class other are super
The weight of pixel image block, with the characteristic similarity after being normalized between the super-pixel image block of insulator to be measured come table
Show, calculation formula is as follows：
Wherein,And, δ () is impulse function for variance of the matrix X under each characteristic dimension；XiRepresent i-th of square
Battle array,Represent kthjA matrix.
Further, the specific method is as follows for generation insulator notable figure：
The conditional probability of the image labeling A (x, y) of input picture I (x, y) is expressed as by set up the condition random field models：
Wherein, Z is partition function；E (A | I) is energy function, energy function be defined as sparse reconstructed error into it is dense
The linear combination of reconstructed error and the two, is expressed as：
Wherein,WithThe notable figure that respectively dense reconstructed error and sparse reconstructed error obtain,WithPoint
It Wei not dense reconstructed error and the corresponding feature weight of sparse reconstructed error；ωmn(Sm-Sn)2Represent color difference penalty term；Sm
And SnRespectively m-th of pixel and the corresponding color characteristic item of nth pixel point；ωmnFor m-th of pixel and n-th of picture
The weight of the corresponding color difference penalty term of vegetarian refreshments；
Wherein, dcol(pm,pn) the normalized L2 norm of color difference between pixel pair；pmAnd pnRespectively m-th of picture
Vegetarian refreshments and nth pixel point,For difference matrix variance and square.
Compared with prior art, the beneficial effects obtained by the present invention are as follows being：
In the case where substation's background is complicated, insulator inspection is carried out using the insulator notable figure that the method for the present invention generates
It surveys, detection performance is good, and up to 89.72%, accurate rate is significantly improved and insulated under complex environment up to 93.76% recall rate
The accuracy of son detection.
Detailed description of the invention
Fig. 1 is flow chart of the invention.
Specific embodiment
Insulator conspicuousness detection method disclosed by the invention based on the fusion of multiple dimensioned reconstructed error, including walk as follows
Suddenly：Firstly, carrying out super-pixel segmentation to insulation subgraph using simple linear iterative algorithm, the super of insulation subgraph surrounding is extracted
Pixel image block is as background dictionary；Then, the corresponding sparse reconstructed error of each super-pixel image block and thick is calculated separately
Close reconstructed error；Then, using the reconstructed error mechanism of transmission based on contextual information, make two reconstructed errors in each class
Image block between propagate, and the reconstructed error obtained in the case of different scale super-pixel segmentation is weighted fusion；Finally,
Sparse structure error and dense reconstructed error are merged using condition random field frame, preferably pressed down by the combination of the two
The noise of imaged middle background generates more accurate insulator notable figure.
In the case where substation's background is complicated, insulator inspection is carried out using the insulator notable figure that the method for the present invention generates
It surveys, detection performance is good, and up to 89.72%, accurate rate is significantly improved and insulated under complex environment up to 93.76% recall rate
The accuracy of son detection.
The invention will be further described below in conjunction with the accompanying drawings.Following embodiment is only used for clearly illustrating the present invention
Technical solution, and not intended to limit the protection scope of the present invention.
As shown in Figure 1, being flow chart of the invention, include the following steps：
Step 1：Super-pixel segmentation is carried out to insulation subgraph, extracts the super-pixel image block conduct of insulation subgraph surrounding
Background template, that is, background dictionary.It is specifically included the following steps：
Step 1.1：Initialization cluster
Assuming that original image contains N number of pixel, the super-pixel number to be divided is K, and the size in each super-pixel region is
N/K, with spacingAlternatively K initial seed CkPrimary condition, then
Ck=[lt,at,bt,xt,yt]T(t=1,2.....K)
Wherein：L is the brightness of insulator color of image；A is the red green color information of each pixel, the generation if a is greater than zero
Table is red, represents green less than zero Shi Ze；B is the champac colouring information of each pixel, represents yellow if b is greater than zero, is less than
Zero represents blue；The value range of a and b is between [- 100,100]；(x, y) represents the coordinate of pixel；T is initial t-th
Seed.
In order to improve computational efficiency when cluster, selection is on the xy space plane of insulation subgraph with each cluster centre
2S*2S region, cluster areas when as super-pixel segmentation, and be not to be found in entire image.Then successively count
The gradient magnitude of 9 pixels in cluster centre 3*3 neighborhood is calculated, the smallest pixel of obtained gradient is in new cluster
The heart.
Step 1.2：Measuring similarity
For each pixel in insulation subgraph, the phase of calculating and the pixel between the smallest cluster centre
Like degree, and the pixel will be assigned to the label of the immediate cluster centre of this pixel.Due to insulation subgraph in most
Significant information is position and colouring information, thus using between pixel and pixel color and positional distance measure similar journey
Degree.Measuring similarity calculation formula is as follows：
Wherein, dcFor the color difference between pixel g and pixel h, dxySpace between pixel h and pixel g away from
It is the balance parameters for balancing color value and spatial information from w, the value is bigger, clusters compacter；S is area size；D (g, k) is
Similarity between pixel g and k-th of cluster centre, the value is smaller, and the two is more similar；lhFor the luminance information of pixel h；
lgFor the luminance information of pixel g；ahFor the red green color information of pixel hj；ahFor the red green color information of pixel h；bhFor
The blue yellow color information of pixel h；bgFor the blue yellow color information of pixel g；xhFor the abscissa information of pixel h；xgFor picture
The abscissa information of vegetarian refreshments g
Step 1.3：K mean cluster method updates iteration cluster centre
Assuming that on an x-y plane, the associated pixel point of cluster centre is in its region 2S*2S.When all pixels in image
After point is associated with nearest cluster centre, then new cluster centre is changed to the flat of generic 5 dimensional vector of all pixels point
Mean value.Above procedure is repeated, stops iteration when convergence.
Step 1.4：Form super-pixel
Wherein pixel will be assigned by the most similar cluster centre label, and then form K super-pixel.
Step 1.5：Tectonic setting template
The super-pixel image block of surrounding in insulation subgraph is extracted as background template, and represents insulator background with b
The D dimensional feature of template, and then construct insulator background template collection B=[b1,b2,…,bu].Wherein, u is that background template concentrates packet
The quantity of the block containing background image.
Step 2：The reconstructed error of each super-pixel image block is calculated, and is utilized respectively rarefaction representation and principal component analysis side
Method calculates it, and super-pixel image block is mapped to background dictionary, obtains the corresponding sparse reconstructed error of entire image and dense
Reconstructed error.It specifically includes the following steps：
Step 2.1：Calculate the corresponding sparse coefficient of each image block
Use background template collection B=[b obtained above1,b2,…,bu] base vector as rarefaction representation, to each
Super-pixel image block is reconstructed, wherein the corresponding sparse coefficient of each super-pixel image block is as follows：
Wherein, yiIndicate i-th of super-pixel image block；αiIndicate xiRarefaction representation；BiIndicate i-th of dictionary matrix；Indicate square of two norms；λ is the regularization parameter for balancing plus sige the right and left error term and bound term, here
It is set as 0.01.
Step 2.2：Calculate sparse reconstructed error of each super-pixel image block on background template B：
Step 2.3：Further to sparse reconstructed errorIt is normalized, makes it between [0,1], It is bigger, show that the correlation of image block and background template is smaller, i.e. super-pixel image block belongs to
A possibility that insulator salient region, is bigger.
Step 2.4：Calculate the dense expression coefficient of each image block：
Firstly, calculating the covariance matrix of insulator background template B：
Wherein, E represents covariance matrix,；Cov(b1, b2) it is b1And b2Between variance；Cov(b1, bu) it is b1And buIt
Between variance；Var(b1) it is b1And b1Between variance；Var(bu) it is buAnd buBetween variance；The calculation formula of variance isbfFor f-th of background template；For the average value of background template；
Then, Eigenvalues Decomposition is carried out to covariance matrix, according to the size of characteristic value, λ1≥λ2≥...≥λkBefore selection
The corresponding feature vector of T characteristic valueAnd in this, as the substrate of Principal Component Analysis.By insulator super-pixel image block
It is projected inOn, obtain the dense expression coefficient of corresponding super-pixel image block：
Wherein：For the dense expression coefficient of i-th of super-pixel image block；For the corresponding feature of preceding T characteristic value
Vector；yiFor i-th of super-pixel image block；For the mean value of super-pixel image block；
Step 2.5：After obtaining the coefficient, so that it may it is right to obtain each insulator super-pixel image block institute using the coefficient
The dense reconstructed error answered：
Wherein,Indicate the data mean value of sample matrix y；For the corresponding feature vector of preceding T characteristic value
Step 2.6：The significance of each super-pixel image block for ease of description, by dense reconstructed errorIt carries out linear
Normalization obtainsMake it in the range of [0,1].
Step 3：Using the reconstructed error mechanism of transmission based on image context information, make insulation of the error in each class
It is propagated between subimage block.It specifically includes the following steps：
Step 3.1：Using K mean cluster algorithm, obtained insulator super-pixel image block is divided into K class.
Step 3.2：Reconstructed error is updated using the thought that error is propagated online is similar to, detailed process is：
The reconstructed error size obtained according to step 2, arranges in descending order, then calculates separately the reconstructed error after propagating：
Assuming that i-th of super-pixel image block of insulator to be measured belongs to k-th of classification, according to other super-pixel image blocks in this kind
Reconstructed error, which is defined as follows：
Wherein, [k1,k2,k3,…,kNc] Nc super-pixel image block in kth classification is represented, τ is above formula(1- τ) εiCoefficient of balance；Representative and i-th of super-pixel image block belong to of a sort
Weighted average of other super-pixel image blocks by reconstructed error after propagating, εiIt is i-th of super-pixel image block in previous step
Obtained reconstructed error；What j was represented is j-th of super-pixel image block in k-th of classification；For in same class other are super
The weight of pixel image block.The present invention utilizes mutual shadow of the insulator super-pixel image block between the image block in same class
It rings, so that the error calculated is more accurate.In addition, similar with the feature after being normalized between insulation subimage block to be measured
The weight to indicate other image blocks in same class is spent, calculation formula is as follows：
Wherein,And, δ () is impulse function for variance of the matrix X under each characteristic dimension.XiRepresent i-th of square
Battle array,Represent kthjA matrix.
Step 4：Multiple dimensioned reconstructed error fusion
The super-pixel figure under different scale segmentation is obtained according to above-mentioned steps and is obtained after context error propagation
Dense and sparse reconstructed error.Ask flat by the way that the reconstructed error of the super-pixel rank after propagating under different scale to be weighted
, the corresponding error of each pixel is obtained, and then obtains the significant result figure of insulator of a width full resolution.
Specific calculating process is as follows：
Wherein,Represent the notable figure after context error propagation, n(h)Indicate the pixel at different segmentation scale h
The affiliated super-pixel image block of point z；Ns indicates that the quantity of super-pixel image block, S value are [1, Ns]；ωzn(s)Indicate segmentation scale h
Shared weight, formula are as follows：
Wherein, fzIndicate the D dimensional feature vector of pixel z,Indicate the super-pixel image block n where pixel z(h)'s
Feature.
Step 5：Sparse and dense reconstructed error is merged using condition random field, makes the two that can complement each other,
Better ambient noise always, generates more accurate insulator notable figure.Specifically comprise the following steps：
Step 5.1：Set up the condition random field (CRF) model, by the item of the image labeling A (x, y) of input picture I (x, y)
Part probability is expressed as：
Wherein, Z is partition function, and E (A | I) it is energy function.
Step 5.2：Definition energy function E (A | I).
In order to enable being highlighted of more preparing of the insulator in image, then need using two above-mentioned reconstructed errors as
Feature is merged, therefore energy function E (A | I) is defined as the linear of sparse reconstructed error and dense reconstructed error and the two
Combination, can be expressed as：
Wherein,WithThe notable figure that respectively dense reconstructed error and sparse reconstructed error obtain,WithPoint
It Wei not dense reconstructed error and the corresponding feature weight of sparse reconstructed error；ωmn(Sm-Sn)2Represent color difference penalty term；Sm
And SnRespectively m-th of pixel and the corresponding color characteristic item of nth pixel point；ωmnFor m-th of pixel and n-th of picture
The weight of the corresponding color difference penalty term of vegetarian refreshments；
Wherein, dcol(pm,pn) the normalized L2 norm of color difference between pixel pair；pmAnd pnRespectively m-th of picture
Vegetarian refreshments and nth pixel point,For difference matrix variance and square.
Combined with specific embodiments below, isolator detecting is carried out using the method for the present invention, it according to testing result can be more
Intuitively know beneficial effects of the present invention：
It is tested using 300 images comprising insulator, also includes power transmission line in image in addition to comprising insulator
The complex backgrounds such as road, trees, it is therefore an objective to the insulator in image is detected in complex background.
It is required that computer environment：Operating system is 14.04 version of Linux, video card GTX980ti, software platform：
MatlabR2014a。
Test set is subjected to conspicuousness detection according to above-mentioned steps 1-5；Using accuracy rate P- recall rate R and F measure this two
A different evaluation index measures the detection effect of the method for the present invention.
Accuracy rate is used to part that is that measure algorithm detects and belonging to GT (really insulate subregion) and accounts for the institute detected
There is the ratio of part, recall rate is used to the ratio that part that is that measure algorithm detects and belonging to GT accounts for GT.F measurement is a use
Carry out the parameter of overall merit recall rate and accuracy rate.
In order to prominent in insulator conspicuousness, accuracy rate is more important than recall rate, by β2It is set as 0.3；Precision
For accuracy rate；Recall is recall rate.
It can be verified and be obtained according to recall rate and accuracy rate curve：The method of the present invention recall rate with higher and accurate
Rate illustrates that the method for the present invention has preferable robustness.
The above is only a preferred embodiment of the present invention, it is noted that for the ordinary skill people of the art
For member, without departing from the technical principles of the invention, several improvement and deformations can also be made, these improvement and deformations
Also it should be regarded as protection scope of the present invention.
Claims (6)
1. the insulator conspicuousness detection method based on the fusion of multiple dimensioned reconstructed error, which is characterized in that include the following steps：
Super-pixel segmentation is carried out to insulation subgraph, extracts the super-pixel image block of insulation subgraph surrounding as background template；
Calculate sparse reconstructed error and dense reconstructed error corresponding to each super-pixel image block of insulation subgraph；
Using the reconstructed error mechanism of transmission based on image context information, make sparse reconstructed error and dense reconstructed error every
It is propagated between super-pixel image block in a class；
The reconstructed error obtained in the case of different scale super-pixel segmentation is weighted fusion；
Sparse reconstructed error and dense reconstructed error are merged using condition random field, generate insulator notable figure.
2. the insulator conspicuousness detection method according to claim 1 based on the fusion of multiple dimensioned reconstructed error, feature
It is, super-pixel segmentation is carried out to insulation subgraph using simple linear iterative algorithm, is specifically comprised the following steps：
Initialization cluster and measuring similarity are carried out to original image；
Iteration cluster centre is updated using K mean cluster method；
It assigns most like cluster centre label to pixel, forms super-pixel image block.
3. the insulator conspicuousness detection method according to claim 1 based on the fusion of multiple dimensioned reconstructed error, feature
It is, seeks sparse reconstructed error using sparse representation method, the specific method is as follows：
Calculate the corresponding sparse coefficient of each super-pixel image block；
Calculate sparse reconstructed error of each super-pixel image block on background template；
Sparse reconstructed error is normalized.
4. the insulator conspicuousness detection method according to claim 1 based on the fusion of multiple dimensioned reconstructed error, feature
It is, seeks dense reconstructed error using Principal Component Analysis, the specific method is as follows：
Calculate the covariance matrix of background template：
Eigenvalues Decomposition is carried out to covariance matrix, according to the size of characteristic value, selected characteristic vector；
Super-pixel image block is projected in feature vector, the dense expression coefficient of corresponding super-pixel image block is obtained；
It is calculated using dense expression coefficient and obtains dense reconstructed error corresponding to each super-pixel image block；
Dense reconstructed error is subjected to linear normalization.
5. the insulator conspicuousness detection method according to claim 1 based on the fusion of multiple dimensioned reconstructed error, feature
It is, the specific method for propagating sparse reconstructed error and dense reconstructed error between the super-pixel image block in each class is such as
Under：
Using K mean cluster algorithm, obtained super-pixel image block is divided into K class；
Reconstructed error is updated, is specifically comprised the following steps：
According to reconstructed error size, descending arranges reconstructed error；
Calculate separately the reconstructed error after propagating：Assuming that i-th of super-pixel image block of insulator to be measured belongs to k-th of classification,
According to the reconstructed error of other super-pixel image blocks in this kind, by the super-pixel image block after contextual information is propagated
Reconstructed error be defined as follows：
Wherein, [k1,k2,k3,…,kNc] Nc super-pixel image block in kth classification is represented, τ is above formulaWith
(1-τ)εiCoefficient of balance；It represents and i-th of super-pixel image block belongs to other of a sort super-pixel images
For block by the weighted average of reconstructed error after propagating, what j was represented is j-th of super-pixel image block in k-th of classification；εiIt is
I super-pixel image block reconstructed error obtained in previous step；For the power of other super-pixel image blocks in same class
Weight indicates that calculation formula is such as with the characteristic similarity after being normalized between the super-pixel image block of insulator to be measured
Under：
Wherein,And, δ () is impulse function for variance of the matrix X under each characteristic dimension；XiI-th of matrix is represented,Represent kthjA matrix.
6. the insulator conspicuousness detection method according to claim 1 based on the fusion of multiple dimensioned reconstructed error, feature
It is, generating insulator notable figure, the specific method is as follows：
The conditional probability of the image labeling A (x, y) of input picture I (x, y) is expressed as by set up the condition random field models：
Wherein, Z is partition function；E (A | I) is energy function, energy function be defined as sparse reconstructed error into dense reconstruct
The linear combination of error and the two, is expressed as：
Wherein,WithThe notable figure that respectively dense reconstructed error and sparse reconstructed error obtain,With
Respectively dense reconstructed error and the corresponding feature weight of sparse reconstructed error；ωmn(Sm-Sn)2Represent color difference
Penalty term；SmAnd SnRespectively m-th of pixel and the corresponding color characteristic item of nth pixel point；ωmnFor m-th of pixel
The weight of color difference penalty term corresponding with nth pixel point；
Wherein, dcol(pm,pn) the normalized L2 norm of color difference between pixel pair；pmAnd pnRespectively m-th pixel and
Nth pixel point,For difference matrix variance and square.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810611259.8A CN108876788B (en) | 2018-06-14 | 2018-06-14 | Insulator significance detection method based on multi-scale reconstruction error fusion |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810611259.8A CN108876788B (en) | 2018-06-14 | 2018-06-14 | Insulator significance detection method based on multi-scale reconstruction error fusion |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108876788A true CN108876788A (en) | 2018-11-23 |
| CN108876788B CN108876788B (en) | 2022-02-15 |
Family
ID=64338437
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810611259.8A Active CN108876788B (en) | 2018-06-14 | 2018-06-14 | Insulator significance detection method based on multi-scale reconstruction error fusion |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108876788B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110633708A (en) * | 2019-06-28 | 2019-12-31 | 中国人民解放军军事科学院国防科技创新研究院 | Deep network significance detection method based on global model and local optimization |
| CN111583171A (en) * | 2020-02-19 | 2020-08-25 | 西安工程大学 | Insulator defect detection method integrating foreground compact characteristic and multi-environment information |
| CN112946525A (en) * | 2021-04-02 | 2021-06-11 | 国家电网有限公司 | Metal oxide arrester surface pollution online detection method |
| CN117635679A (en) * | 2023-12-05 | 2024-03-01 | 之江实验室 | Curved surface efficient reconstruction method and device based on pre-training diffusion probability model |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102938153A (en) * | 2012-11-21 | 2013-02-20 | 清华大学 | Video image splitting method based on restrain spectral clustering and markov random field |
| CN106228544A (en) * | 2016-07-14 | 2016-12-14 | 郑州航空工业管理学院 | A kind of significance detection method propagated based on rarefaction representation and label |
| US9760807B2 (en) * | 2016-01-08 | 2017-09-12 | Siemens Healthcare Gmbh | Deep image-to-image network learning for medical image analysis |
-
2018
- 2018-06-14 CN CN201810611259.8A patent/CN108876788B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102938153A (en) * | 2012-11-21 | 2013-02-20 | 清华大学 | Video image splitting method based on restrain spectral clustering and markov random field |
| US9760807B2 (en) * | 2016-01-08 | 2017-09-12 | Siemens Healthcare Gmbh | Deep image-to-image network learning for medical image analysis |
| CN106228544A (en) * | 2016-07-14 | 2016-12-14 | 郑州航空工业管理学院 | A kind of significance detection method propagated based on rarefaction representation and label |
Non-Patent Citations (2)
| Title |
|---|
| RADHAKRISHNA ACHANTA: "SLIC：Superpixels Compared to State-of-the-Art Superpixel Methods", 《IEEE TRANSACTION ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE》 * |
| 李晓辉: "基于多尺度重构误差融合的绝缘子显著性检测方法", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110633708A (en) * | 2019-06-28 | 2019-12-31 | 中国人民解放军军事科学院国防科技创新研究院 | Deep network significance detection method based on global model and local optimization |
| CN111583171A (en) * | 2020-02-19 | 2020-08-25 | 西安工程大学 | Insulator defect detection method integrating foreground compact characteristic and multi-environment information |
| CN111583171B (en) * | 2020-02-19 | 2023-04-07 | 西安工程大学 | Insulator defect detection method integrating foreground compact characteristic and multi-environment information |
| CN112946525A (en) * | 2021-04-02 | 2021-06-11 | 国家电网有限公司 | Metal oxide arrester surface pollution online detection method |
| CN112946525B (en) * | 2021-04-02 | 2022-08-09 | 国家电网有限公司 | Metal oxide arrester surface pollution online detection method |
| CN117635679A (en) * | 2023-12-05 | 2024-03-01 | 之江实验室 | Curved surface efficient reconstruction method and device based on pre-training diffusion probability model |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108876788B (en) | 2022-02-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108876788A (en) | Insulator conspicuousness detection method based on the fusion of multiple dimensioned reconstructed error | |
| CN107742093B (en) | Real-time detection method, server and system for infrared image power equipment components | |
| CN109446925A (en) | A kind of electric device maintenance algorithm based on convolutional neural networks | |
| CN112200178B (en) | Transformer substation insulator infrared image detection method based on artificial intelligence | |
| Zheng et al. | Infrared image detection of substation insulators using an improved fusion single shot multibox detector | |
| CN109063780A (en) | Partial discharge of transformer recognition methods based on particle group optimizing core neighbour's propagation algorithm | |
| CN107370150B (en) | The Power system state estimation Bad data processing method measured based on synchronized phasor | |
| CN115719338A (en) | PCB (printed circuit board) surface defect detection method based on improved YOLOv5 | |
| CN103093243B (en) | The panchromatic remote sensing image clouds of high-resolution sentences method | |
| CN111598942A (en) | Method and system for automatically positioning electric power facility instrument | |
| CN113159334B (en) | Electrical equipment infrared image real-time detection and diagnosis method based on light-weight deep learning | |
| CN107341470A (en) | A kind of transmission of electricity line detecting method based on Aerial Images | |
| CN113344852A (en) | Target detection method and device for power scene general-purpose article and storage medium | |
| CN109886937A (en) | Defects of insulator detection method based on super-pixel segmentation image recognition | |
| CN105354850A (en) | Method for detecting size of complex curved part in three-dimensional matching manner based on electric-field properties | |
| Zhou et al. | Intelligent reading recognition method of a pointer meter based on deep learning in a real environment | |
| Zheng et al. | Arbitrary-oriented detection of insulators in thermal imagery via rotation region network | |
| CN102565554B (en) | Method for generating three-dimensional lightning positioned place flash point distribution map | |
| CN110197200A (en) | A kind of electronic tag for clothing generation method based on machine vision | |
| CN110321528A (en) | A kind of Hyperspectral imaging heavy metal-polluted soil concentration evaluation method based on semi-supervised geographical space regression analysis | |
| Li et al. | Insulator infrared image segmentation algorithm based on dynamic mask and box annotation | |
| CN109753896A (en) | A kind of unsupervised heterologous method for detecting change of remote sensing image based on general character autocoder | |
| CN108632752A (en) | Indoor location fingerprint location Radio Map method for building up based on tensor filling | |
| CN114838923A (en) | Fault diagnosis model establishing method and fault diagnosis method for on-load tap-changer | |
| CN107133284A (en) | A kind of view method for searching three-dimension model based on prevalence study |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |