CN109102036A - A kind of image tagged method and device for transmission line malfunction identification - Google Patents
A kind of image tagged method and device for transmission line malfunction identification Download PDFInfo
- Publication number
- CN109102036A CN109102036A CN201811122966.7A CN201811122966A CN109102036A CN 109102036 A CN109102036 A CN 109102036A CN 201811122966 A CN201811122966 A CN 201811122966A CN 109102036 A CN109102036 A CN 109102036A
- Authority
- CN
- China
- Prior art keywords
- region
- similarity
- image
- size
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- 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
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
- G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The embodiment of the present application shows a kind of image tagged method and device for transmission line malfunction identification, and the technical solution shown in the embodiment of the present application carries out the image tagged of transmission line malfunction based on the selective search mode of region merging technique.Wherein the introducing of region merging technique thought improves the efficiency of image segmentation, can faster obtain region hypothesis.The use of improved SVM (Support Vector Machine) can carry out the classification of transmission line of electricity various faults simultaneously, improve the efficiency of image tagged.
Description
Technical field
The present invention relates to image identification technical field, in particular to a kind of image tagged for transmission line malfunction identification
Method and device.
Background technique
The rapid extension of grid power transmission route mileage and the increasingly complexity of corridor environment, in the limited reality of personnel
Under, route O&M service work faces enormous challenge.Currently, route transport inspection department doors at different levels mainly use manual inspection mode,
And be aided with the advanced technologies such as helicopter, unmanned plane and carry out work compound, to improve operating efficiency.The machine that power grid is carried out every year patrols
The inspection image/video amount generated in operation is huge, rely solely on manually to video image carry out interpretation, workload it is huge and
It easily fails to judge, causes to can be only done interpretation that is a small amount of and suspecting obvious defective data, mass data can be only placed at hard disk
In can not apply.
In recent years, with digital vedio recording and the fast development of computer vision technique and extensive use, using helicopter, nothing
The means such as man-machine equal carryings photograph (camera shooting) machine realize that efficient, quick polling transmission line also obtains rapid promotion and application.
By patrolling the depth digging utilization of visible image data (picture, video) to a large amount of machine, power transmission line can be effectively found
The transmission lines of electricity major defects such as road appearance, running environment, element exception, provide reference for equipment management and operation and maintenance.
Premise to unmanned plane inspection image, video progress depth excavation is to the transmission line malfunction in image, video
Carry out target label.The specific location for finding a failure in the picture, determines the specific classification of failure, and makes image mark
Note.Way before is mainly based upon exhaustive search, and a window is selected to scan whole image, changes window size, continues to sweep
Whole image is retouched, this way result is too complicated, and very time-consuming.When needing to spend a large amount of when finding suitable position
Between, cause the feature classified for object identification can not be too complicated, some simple features can only be used.Lead in failure modes
Frequently be bis- classification method of SVM, classification effectiveness is low.For a plurality of types of transmission line malfunctions can only be repeated as many times into
Row failure modes.
Therefore, it is necessary to carry out the search of fault target hypothesis on location using the selective search method based on region merging technique.
Since this step efficiency is higher in the hypothesis on location for obtaining object for selection search, Scale invariant features transform can be used
(SIFT) etc. operands are big, indicate the strong feature of ability.More classification method (the OVR of improved SVM can be used in assorting process
SVMs) classify, improve the classification effectiveness of polymorphic type failure.A plurality of types of transmission line malfunctions also can be carried out once
Classification marker.
Summary of the invention
Goal of the invention of the invention is to provide a kind of image tagged method and device for transmission line malfunction identification,
To solve the prior art.
The embodiment of the present application first aspect shows a kind of image tagged method for transmission line malfunction identification, the side
Method includes:
Input transmission line of electricity picture to be processed；
Vegetarian refreshments is imaged in the transmission line of electricity picture segmentation to be processed, the pixel is merged according to presetting rule, is generated
Combined region image；
The local feature for extracting combined region image constructs description of the local feature；
By the sub- input picture markup model of description, the image tagged of the transmission line of electricity picture to be processed is generated.
It is selectable, it is described that vegetarian refreshments is imaged in the transmission line of electricity picture segmentation to be processed, institute is merged according to presetting rule
The step of stating pixel, generating combined region image include:
Picture is abstracted as figure, the figure includes: pixel, and, branch；
Using pixel as the vertex of figure, weight: W (V is assigned according to colour-difference to the branch of figurei,Vj)；
As a region, the inside in region is poor on each vertex in figure are as follows: Int (C)；
Wherein, Int (C)=max W (e)；
Difference between region are as follows: Dif (C1,C2)；
Wherein, Dif (C1,C2)=min W ((V1,V2)), Dif takes infinity if two sides do not connect；
The judgment criteria of region segmentation are as follows:
Wherein, MInt (C1,C2)=min ((Int (C1))+τ(C1),(Int(C2))+τ(C2)), τ (C) is given threshold；
Region merging technique is carried out according to above-mentioned region segmentation standard；
As W (C1,C2) < MInt when two regions are merged；Finally obtain region segmentation set；
The similarity between adjacent area is calculated, similarity set is obtained, maximum two regions of similarity is closed
And until similarity collection be combined into empty set, obtain generate combined region image.
Selectable, the similarity calculated between adjacent area obtains similarity set, by similarity maximum two
A region merges, up to the step of similarity collection is combined into empty set, obtains generation combined region image includes:
The color similarity, texture similarity, size similarity, similarity of coincideing for calculating the region, obtain similarity
Set；
Maximum two regions of similarity are searched out from similarity set to merge；
Similarity relevant to original region is removed in similarity set, calculates the similar of new region region adjacent thereto
New calculated similarity is added in similarity set by degree, until similarity collection is combined into empty set, obtains generating combined region
Image.
Method that is selectable, calculating color similarity specifically:
It is normalized using L1 norm, obtains the histogram of each Color Channel in region, finally obtain a multi-C vector
Color similarity calculation is as follows between region:
Wherein Scolour(ri,rj) indicate region riAnd rjColor similarity between region；
Indicate the color channel histograms in region；
It needs to calculate histogram to new region again after region merges two-by-two, calculation is as follows:
Wherein size (ri), size (rj) indicate region ri, rjSize, CtIndicate the color histogram after region merging technique
Figure；
It is selectable, the calculation method of texture similarity specifically:
Texture similarity, which calculates, uses SIFT feature, carries out gaussian derivative meter to 8 different directions of each Color Channel
It calculates；
It is normalized using L1, each Color Channel takes histogram, each available multi-C vector in region；Texture similarity calculation between region is as follows:
Wherein Stexture(ri,rj) indicate ri, rjTexture similarity between region,Indicate the texture in region；
Need to re-start textural characteristics calculating after region merging technique, calculation formula is as follows:
Wherein texture (ri), texture (rj) indicate region ri, rjSize, TtIndicate that the texture for merging rear region is special
Sign；
Then the calculating of size similarity is carried out；In order to enable small region can preferentially merge, it is big that we carry out region
The calculating of small similarity, area size refer to include in region pixel how much, calculation formula is as follows:
Wherein size (ri), size (rj) indicate ri, rjThe size in region；
Size (im) indicates the size of picture, Ssize(ri,rj) indicate ri, rjArea size similarity.
It is selectable, then carry out the similarity calculation that coincide specifically:
Wherein Sfill(ri,rj) representation space degree of overlapping, size (BBij) indicate region merging technique after Bound Box it is big
Small, size (im) indicates the size of picture；
Weight is assigned to each similarity, obtains similarity total between two regions；
Calculation formula is as follows:
S(ri,rj)=α1Scolour(ri,rj)+α2Stexture(ri,rj)+α3Ssize(ri,rj)+α4Sfill(ri,rj)；
Wherein S (ri,rj) indicate region between total similarity, weight αi∈{0,1}。
Selectable, the local feature for extracting combined region image constructs description of the local feature:
Area image is handled using Gaussian Blur, the scale that area image is constructed in the form of gaussian pyramid is empty
Between, specifically, obtaining the different image of size by carrying out continuous depression of order sampling to area image；
From big to small, it is arranged successively from the bottom up, forms gaussian pyramid；
Then difference is done using gaussian pyramid upper layer and lower layer image, obtains difference of Gaussian image；
Each pixel on difference of Gaussian image on every image is compared with surrounding all pixels point, if
The pixel 26 pixels more adjacent than it are all big or small, then it is assumed that the point is extreme point；
Determine the position of the extreme point, direction, and, size, as description of local feature.
Selectable, the step of position of the determination extreme point, direction and size, includes:
Over-fitting three-dimensional quadratic function accurately to obtain the position of extreme point；
The reference direction for determining extreme point is sought using the method for image gradient；
The modulus value and direction calculating formula of gradient are as follows:
θ (x, y)=tan-1((L(x,y+1)-L(x,y-1))/(L(x+1,y)-L(x-1,y)))；
Wherein L is the scale space value where extreme point；X, y are the position of extreme point.
The embodiment of the present application second aspect shows a kind of image tagged device for transmission line malfunction identification, the dress
It sets and includes:
Input module, for inputting transmission line of electricity picture to be processed；
Cutting module merges institute according to presetting rule for vegetarian refreshments to be imaged in the transmission line of electricity picture segmentation to be processed
Pixel is stated, combined region image is generated；
Module is constructed, for extracting the local feature of combined region image, constructs description of the local feature；
Mark module, for generating the transmission line of electricity picture to be processed for the sub- input picture markup model of description
Image tagged.
From the above technical scheme, the embodiment of the present application shows a kind of image tagged for transmission line malfunction identification
Method and device, the technical solution shown in the embodiment of the present application carry out transmission line of electricity based on the selective search mode of region merging technique
The image tagged of failure.Wherein the introducing of region merging technique thought improves the efficiency of image segmentation, can faster obtain area
Assume in domain.The use of improved SVM (Support Vector Machine) can carry out transmission line of electricity various faults simultaneously
Classification, improves the efficiency of image tagged.
Detailed description of the invention
It in order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, below will be to institute in embodiment
Attached drawing to be used is needed to be briefly described, it should be apparent that, the accompanying drawings in the following description is only some implementations of the invention
Example, for those of ordinary skill in the art, without creative efforts, can also obtain according to these attached drawings
Obtain other attached drawings.
Fig. 1 is that a kind of the embodiment of the present application for exemplifying is preferably implemented according to one is a kind of for transmission line malfunction identification
The flow chart of image tagged method；
Fig. 2 is that the svm classifier schematic diagram exemplified is preferably implemented according to one；
Fig. 3 is that the graphical rule space schematic diagram exemplified is preferably implemented according to one；
Fig. 4 is that a kind of the embodiment of the present application for exemplifying is preferably implemented according to one is a kind of for transmission line malfunction identification
The structure chart of image tagged device.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Whole description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall within the protection scope of the present invention.
Referring to Fig. 1, the embodiment of the present application first aspect shows a kind of image tagged for transmission line malfunction identification
Method, which comprises
S101 inputs transmission line of electricity picture to be processed；
Vegetarian refreshments is imaged in the transmission line of electricity picture segmentation to be processed by S102, merges the pixel according to presetting rule,
Generate combined region image；
S103 extracts the local feature of combined region image, constructs description of the local feature；
The sub- input picture markup model of description is generated the image mark of the transmission line of electricity picture to be processed by S104
Note.
Specifically, in video processing and field of image recognition, the label accuracy pair of the positive sample for model training
The training of model plays a crucial role.At present when doing sample labeling, the method for mostly using handmarking greatly, this side
Method takes time and effort, inefficiency.The present invention carries out goal hypothesis position using the selective search method based on region merging technique and searches
Rope obtains all hypothesis on location that may be present.Polymorphic type point is carried out to fault target using improved svm classifier method again
Class, and then realize that the automation to fault target marks.
Above-mentioned technical problem of the invention mainly passes through what following two parts were realized:
First part needs to carry out region segmentation to image before carrying out transmission line malfunction classification and marking.It is based on
Region merging technique principle, early period first carry out original segmentation to image, obtain the original area of image, then use some consolidation strategies
Original area is merged, the regional structure of a stratification is obtained, and the event in these structures comprising that may need to mark
Hinder target.
Step 1.1, original point is obtained using the method for Efficient Graph-Based Image Segmentation
Cut region R={ r1,r2,…rn}。
Step 1.2, similarity set is initialized
Step 1.3, the similarity between every two adjacent area is calculated, the similarity being calculated is added to set S
In.
Step 1.4, it is found out from similarity set S, the maximum two region r of similarityiAnd rj, it is merged into being one
A region rt, removed from similarity set originally and riAnd rjThe similarity calculated between adjacent area calculates rtIt is adjacent thereto
Its result is added in similarity set S by the similarity in region.Simultaneously by new region rtIt is added in regional ensemble R.Circulation
The above process, circulation terminates when set S is empty.
Step 1.5, the Bound Boxes in each region is obtained, this result is exactly the possible outcome of object space.
Second part, by the region merging technique of first part, the hypothesis on location of available some transmission line malfunctions, the
The task of two parts is how to look for out of order actual location from these hypothesis on location and determines the type of failure, thus right
Transmission line malfunction is marked., can be big using operand in object identification feature extraction, indicate the strong scale of ability not
Become eigentransformation (SIFT) and carries out image characteristics extraction.The failure modes of transmission line of electricity are then carried out using improved SVM classifier
Various faults classification.Specific step is as follows:
Step 2.1, feature extraction.Scale invariant features transform algorithm (SIFT) is used when carrying out image characteristics extraction,
The Sampling characters that image is carried out under preset scale extract.By the available one-dimensional feature of feature extraction to
Amount.
Step 2.2, improved SVM classifier training.Classification method carries out transmission line malfunction multiclass using improved SVM
Type classification.The first step is the setting of positive and negative samples, i.e., using true transmission line malfunction target as positive sample, selection with just
Sample repetitive rate 20%-50% sample as negative sample, and exclude Duplication in the selection course of sample and be more than
70% negative sample.Second step is iteration, is trained to model.
The present invention carries out the image tagged of transmission line malfunction using the selective search mode based on region merging technique.Wherein
The introducing of region merging technique thought improves the efficiency of image segmentation, can faster obtain region hypothesis.Improved svm classifier
The use of device can carry out the classification of transmission line of electricity various faults simultaneously, improve the efficiency of image tagged.
Embodiment 1:
In conjunction with attached drawing, technical solution of the present invention is further described.
Step 1, the transmission line of electricity color image for needing to mark is inputted first.
Step 2, using selective target identification (Efficient Graph-Based Image Segmentation) side
Method carries out region segmentation.Image is divided into point one by one according to pixel first, is then closed by the region being previously set
And strategy merges, and finally obtains a reasonable image segmentation collection R={ r1,r2,…rn, wherein riIndicate an image
Segmentation.
It is first figure, vertex of the pixel as figure, to branch's foundation color of figure by image abstraction in the step 2
Difference assigns certain weight W (Vi,Vj), wherein Vi, VjIndicate the vertex of figure, then weight is arranged by ascending order.Secondly image is carried out
Original segmentation S, each vertex is as a region in figure.The inside difference in region is expressed as Int (C), wherein Int (C)=max
W (e), W (e) indicate the weight of the branch of figure, and the difference between region is expressed as Dif (C1,C2), wherein Dif (C1,C2)=min
W((V1,V2)), Dif takes infinity if two sides do not connect.The judgment criteria of region segmentation are as follows:
Wherein, MInt (C1,C2)=min ((Int (C1))+τ(C1),(Int(C2))+τ(C2)), τ (C) is given threshold, D
(C1,C2) indicate region between segmentation.Region merging technique is carried out according to above-mentioned region segmentation standard, as W (C1,C2) < MInt
When two regions are merged, on the contrary then two region of nonjoinder.Union operation is repeated, region segmentation set is finally obtained.
Step 3, similarity set initializes.We calculate the similarity between adjacent area in terms of four, respectively
It is color similarity, texture similarity, size similarity, similarity of coincideing.Similarity set is initialized as empty set
Referring to Fig. 2, specific assorting process is as shown in Figure 2.
Step 4, the similarity between every two adjacent area is calculated separately, is then added to calculated similarity similar
It spends in set S.
In the step 4, color similarity is calculated first, is normalized using L1 norm, obtains each Color Channel of image
25bins histogram, finally obtain 75 dimensional vectorsColor similarity calculation between region
It is as follows:
Wherein Scolour(ri,rj) indicate region riAnd rjColor similarity between region.Indicate the face in region
Chrominance channel histogram.
It needs to calculate histogram to new region again after region merges two-by-two, calculation is as follows:
Wherein size (ri), size (rj) indicate region ri, rjSize, CtIndicate the color histogram after region merging technique
Figure.
Followed by the calculating of texture similarity.Texture similarity, which calculates, uses SIFT feature, to 8 of each Color Channel
Different directions carry out gaussian derivative calculating, and wherein variance is set as σ=1.It is normalized using L1, each Color Channel takes 10bins
Histogram, each available 240 dimensional vectors in regionTexture similarity calculating side between region
Formula is as follows:
Wherein Stexture(ri,rj) indicate ri, rjTexture similarity between region,Indicate the texture in region.
Need to re-start textural characteristics calculating after region merging technique, calculation formula is as follows:
Wherein texture (ri), texture (rj) indicate region ri, rjSize, TtIndicate that the texture for merging rear region is special
Sign.
Then the calculating of size similarity is carried out.In order to enable small region can preferentially merge, it is big that we carry out region
The calculating of small similarity, area size refer to include in region pixel how much, calculation formula is as follows:
Wherein size (ri), size (rj) indicate ri, rjThe size in region, size (im) indicate the size of picture, Ssize
(ri,rj) indicate ri, rjArea size similarity.
Then the similarity calculation that coincide is carried out.In order to make can more to coincide after two region merging techniques, kissed
Close similarity calculation.It is framed after two region merging techniques with rectangle as small as possible, i.e. Bound Box, rectangle is smaller to be shown
The goodness of fit is better.Calculation formula is as follows:
Wherein Sfill(ri,rj) representation space degree of overlapping, size (BBij) indicate region merging technique after Bound Box it is big
Small, size (im) indicates the size of picture.
Certain weight is finally assigned to each similarity, obtains similarity total between two regions.Calculation formula is such as
Under:
S(ri,rj)=α1Scolour(ri,rj)+α2Stexture(ri,rj)+α3Ssize(ri,rj)+α4Sfill(ri,rj)；
Wherein S (ri,rj) indicate region between total similarity, weight αi∈{0,1}。
Step 5, the similarity between adjacent area can be calculated by step 4, is then found out from similarity set S
Maximum two regions of similarity merge, and remove phase relevant to original region after region merging technique in similarity set
Like degree, and the similarity in new region region adjacent thereto is calculated, new calculated similarity is added in similarity set.
The region newly merged is added in regional ensemble.
Step 6, circulation step 4, step 5, when similarity set becomes empty set, circulation terminates.Region merging technique at this time
Whole process terminate.
Step 7, each new combined region is framed with rectangle as small as possible, obtains image segmentation
Bounding Boxes.All obtained small rectangles are exactly the hypothesis on location for carrying out image segmentation.
Step 8, feature generates.It is carried out using to the high scale invariant feature transfer algorithm of the tolerances such as light, noise
The local shape factor of image.Extreme point is sought in the scale space of generation first, then obtains position, the scale of extreme point
And direction, finally construct description of the matrix as characteristics of image.
In the step 8, image is handled using Gaussian Blur first.Fuzzy Template is generated using Gaussian function,
Original image and template are subjected to convolutional calculation, realization obscures image.Its calculation formula is:
L (x, y, σ)=G (x, y, σ) * I (x, y)
Wherein, * indicates that the convolution algorithm of two functions, G (x, y, σ) are Gaussian function, and I (x, y) is original image.Gaussian function
Number calculation formula are as follows:
Wherein think, x, y indicates that the coordinate at image midpoint, σ indicate variance.
Then the scale space of image is constructed in the form of gaussian pyramid.It is adopted by carrying out continuous depression of order to original image
Sample obtains the different image of size and is arranged successively from the bottom up from big to small, forms gaussian pyramid.Then height is utilized
This pyramid upper layer and lower layer image does difference, obtains difference of Gaussian image.
Followed by the detection of spatial extrema point, the Preliminary detection of extreme point is carried out on difference of Gaussian image.It allows every
Each pixel opened on image is compared with surrounding all pixels point, if 26 pixels that the pixel is more adjacent than it
Point (please referring to Fig. 3) is all big or small, then it is assumed that the point is extreme point, is then accurately obtained by being fitted three-dimensional quadratic function
The position of extreme point.
Extreme point is obtained later it needs to be determined that the reference direction of extreme point, is sought using the method for image gradient.Gradient
Modulus value and direction calculating formula are as follows:
θ (x, y)=tan-1((L(x,y+1)-L(x,y-1))/(L(x+1,y)-L(x-1,y)))
Wherein, m (x, y) indicates the modulus value of image gradient, and θ (x, y) indicates the direction of image gradient, and L is extreme point place
Scale space value, x, y indicate extreme point abscissa, ordinate.
Using the direction of pixel in statistics with histogram extreme value vertex neighborhood, 360 degree of directions are divided into 36 ranges, Mei Gefan
Enclose 10 degree.The peak value of histogram is then defined as the principal direction of extreme point.
There are position, scale and three, direction information by each extreme point of above step, therefore can use a vector
Extreme point to be described.Centered on extreme point, extreme point direction is principal direction, and surrounding neighbors are divided into 4 × 4 windows
Mouthful, the gradient information in each 8 direction of window calculation, therefore description of 128 dimensional vectors as image can be obtained.
Step 9, a variety of transmission line malfunction classification are carried out.Failure modes are carried out using improved SVM multi-categorizer, defeated
A kind of fault type of electric line is as positive sample, remaining fault type and normal condition are as negative sample.According to fault type
Positive negative sample is successively replaced, different training sets is constructed.Then training is iterated to model respectively using training set, obtained more
Type fault classifier.
In the step 9, it is assumed that transmission line malfunction is divided into shaft tower by we Bird's Nest, and insulator is broken, and small fitting lacks
These four types of normal condition of becoming estranged, are denoted as A, B, C, D respectively.When setting training set, according to following rule:
(1) A kind fault type is as positive sample collection, and B, C, D seed type are as negative sample collection.
(2) B kind fault type is as positive sample collection, and A, C, D seed type are as negative sample collection.
(3) C kind fault type is as positive sample collection, and A, B, D seed type are as negative sample collection.
(4) D kind transmission line of electricity normal condition is as positive sample collection, and A, B, C kind fault type are as negative sample collection.
Model is trained respectively using above 4 training sets, obtains four trained files.Carrying out transmission line of electricity event
By four trained file in parallel operations when barrier label, four training results, four results of Automatic Model Selection can be obtained simultaneously
Intermediate value is maximum, and stamps respective type label.
Step 10, last determination manually is carried out to the failure that model marks automatically.If in the case where confidence degree is horizontal,
Automatically the failure marked meets the requirements then through desk checking, needs to carry out manual setting if notable difference occurs in label.
Referring to Fig. 4, the embodiment of the present application second aspect shows a kind of image tagged for transmission line malfunction identification
Device, described device include:
Input module 21, for inputting transmission line of electricity picture to be processed；
Cutting module 22 merges for vegetarian refreshments to be imaged in the transmission line of electricity picture segmentation to be processed according to presetting rule
The pixel generates combined region image；
Module 23 is constructed, for extracting the local feature of combined region image, constructs description of the local feature；
Mark module 24, for generating the transmission line of electricity figure to be processed for the sub- input picture markup model of description
The image tagged of piece.
Those skilled in the art after considering the specification and implementing the invention disclosed here, will readily occur to of the invention its
Its embodiment.This application is intended to cover any variations, uses, or adaptations of the invention, these modifications, purposes or
Person's adaptive change follows general principle of the invention and including the undocumented common knowledge in the art of the present invention
Or conventional techniques.The description and examples are only to be considered as illustrative, and true scope and spirit of the invention are by following
Claim is pointed out.
It should be understood that the present invention is not limited to the precise structure already described above and shown in the accompanying drawings, and
And various modifications and changes may be made without departing from the scope thereof.The scope of the present invention is limited only by the attached claims.
Claims (9)
1. a kind of image tagged method for transmission line malfunction identification, which is characterized in that the described method includes:
Input transmission line of electricity picture to be processed；
Vegetarian refreshments is imaged in the transmission line of electricity picture segmentation to be processed, the pixel is merged according to presetting rule, generates and merges
Area image；
The local feature for extracting combined region image constructs description of the local feature；
By the sub- input picture markup model of description, the image tagged of the transmission line of electricity picture to be processed is generated.
2. the method according to claim 1, wherein described the transmission line of electricity picture segmentation to be processed is imaged
Vegetarian refreshments, the step of merging the pixel according to presetting rule, generate combined region image include:
Picture is abstracted as figure, the figure includes: pixel, and, branch；
Using pixel as the vertex of figure, weight: W (V is assigned according to colour-difference to the branch of figurei,Vj)；
As a region, the inside in region is poor on each vertex in figure are as follows: Int (C)；
Wherein, Int (C)=max W (e)；
Difference between region are as follows: Dif (C1,C2)；
Wherein, Dif (C1,C2)=min W ((V1,V2)), Dif takes infinity if two sides do not connect；
The judgment criteria of region segmentation are as follows:
Wherein, MInt (C1,C2)=min ((Int (C1))+τ(C1),(Int(C2))+τ(C2)), τ (C) is given threshold；
Region merging technique is carried out according to above-mentioned region segmentation standard；
As W (C1,C2) < MInt when two regions are merged；Finally obtain region segmentation set；
The similarity between adjacent area is calculated, similarity set is obtained, maximum two regions of similarity is merged, directly
It is combined into empty set to similarity collection, obtains generating combined region image.
3. according to the method described in claim 2, it is characterized in that, the similarity calculated between adjacent area, obtains phase
Gather like degree, maximum two regions of similarity are merged, until similarity collection is combined into empty set, obtains generating combined region
The step of image includes:
The color similarity, texture similarity, size similarity, similarity of coincideing for calculating the region, obtain similarity set；
Maximum two regions of similarity are searched out from similarity set to merge；
Similarity relevant to original region is removed in similarity set, calculates the similarity in new region region adjacent thereto,
New calculated similarity is added in similarity set, until similarity collection is combined into empty set, obtains generating combined region figure
Picture.
4. according to the method described in claim 3, it is characterized in that, the method for calculating color similarity specifically:
It is normalized using L1 norm, obtains the histogram of each Color Channel in region, finally obtain a multi-C vector
Color similarity calculation is as follows between region:
Wherein Scolour(ri,rj) indicate region riAnd rjColor similarity between region；
Indicate the color channel histograms in region；
It needs to calculate histogram to new region again after region merges two-by-two, calculation is as follows:
Wherein size (ri), size (rj) indicate region ri, rjSize, CtIndicate the color histogram after region merging technique.
5. according to the method described in claim 3, it is characterized in that, the calculation method of texture similarity specifically:
Texture similarity, which calculates, uses SIFT feature, carries out gaussian derivative calculating to 8 different directions of each Color Channel；
It is normalized using L1, each Color Channel takes histogram, each available multi-C vector in region；Texture similarity calculation between region is as follows:
Wherein Stexture(ri,rj) indicate ri, rjTexture similarity between region,Indicate the texture in region；
Need to re-start textural characteristics calculating after region merging technique, calculation formula is as follows:
Wherein texture (ri), texture (rj) indicate region ri, rjSize, TtIndicate the textural characteristics for merging rear region；
Then the calculating of size similarity is carried out；In order to enable small region can preferentially merge, we carry out area size phase
Like the calculating of degree, area size refer to include in region pixel how much, calculation formula is as follows:
Wherein size (ri), size (rj) indicate ri, rjThe size in region；
Size (im) indicates the size of picture, Ssize(ri,rj) indicate ri, rjArea size similarity.
6. according to the method described in claim 3, it is characterized in that, then carrying out the similarity calculation that coincide specifically:
Wherein Sfill(ri,rj) representation space degree of overlapping, size (BBij) indicate region merging technique after Bound Box size,
The size of size (im) expression picture；
Weight is assigned to each similarity, obtains similarity total between two regions；
Calculation formula is as follows:
S(ri,rj)=α1Scolour(ri,rj)+α2Stexture(ri,rj)+α3Ssize(ri,rj)+α4Sfill(ri,rj)；
Wherein S (ri,rj) indicate region between total similarity, weight αi∈{0,1}。
7. method according to claim 1-6, which is characterized in that the part for extracting combined region image is special
Sign constructs description of the local feature:
Area image is handled using Gaussian Blur, the scale space of area image is constructed in the form of gaussian pyramid,
Specifically, obtaining the different image of size by carrying out continuous depression of order sampling to area image；
From big to small, it is arranged successively from the bottom up, forms gaussian pyramid；
Then difference is done using gaussian pyramid upper layer and lower layer image, obtains difference of Gaussian image；
Each pixel on difference of Gaussian image on every image is compared with surrounding all pixels point, if the picture
Vegetarian refreshments 26 pixels more adjacent than it are all big or small, then it is assumed that the point is extreme point；
Determine the position of the extreme point, direction, and, size, as description of local feature.
8. the method according to the description of claim 7 is characterized in that the position of the determination extreme point, direction and ruler
Very little step includes:
Over-fitting three-dimensional quadratic function accurately to obtain the position of extreme point；
The reference direction for determining extreme point is sought using the method for image gradient；
The modulus value and direction calculating formula of gradient are as follows:
θ (x, y)=tan-1((L(x,y+1)-L(x,y-1))/(L(x+1,y)-L(x-1,y)))；
Wherein L is the scale space value where extreme point；X, y are the position of extreme point.
9. a kind of image tagged device for transmission line malfunction identification, which is characterized in that described device includes:
Input module, for inputting transmission line of electricity picture to be processed；
Cutting module merges the picture according to presetting rule for vegetarian refreshments to be imaged in the transmission line of electricity picture segmentation to be processed
Vegetarian refreshments generates combined region image；
Module is constructed, for extracting the local feature of combined region image, constructs description of the local feature；
Mark module, for generating the figure of the transmission line of electricity picture to be processed for the sub- input picture markup model of description
As label.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811122966.7A CN109102036A (en) | 2018-09-26 | 2018-09-26 | A kind of image tagged method and device for transmission line malfunction identification |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811122966.7A CN109102036A (en) | 2018-09-26 | 2018-09-26 | A kind of image tagged method and device for transmission line malfunction identification |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109102036A true CN109102036A (en) | 2018-12-28 |
Family
ID=64867840
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811122966.7A Pending CN109102036A (en) | 2018-09-26 | 2018-09-26 | A kind of image tagged method and device for transmission line malfunction identification |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109102036A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109992682A (en) * | 2019-03-29 | 2019-07-09 | 联想(北京)有限公司 | A kind of image-recognizing method, device and electronic equipment |
| CN113077481A (en) * | 2021-03-29 | 2021-07-06 | 上海闻泰信息技术有限公司 | Image processing method and device, computer equipment and storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103260016A (en) * | 2013-06-04 | 2013-08-21 | 成都思晗科技有限公司 | Remote and intelligent line-tracking method of electric transmission line |
| CN103413150A (en) * | 2013-06-28 | 2013-11-27 | 广东电网公司电力科学研究院 | Power line defect diagnosis method based on visible light image |
| CN106778633A (en) * | 2016-12-19 | 2017-05-31 | 江苏慧眼数据科技股份有限公司 | A kind of pedestrian recognition method based on region segmentation |
| CN107392215A (en) * | 2017-08-02 | 2017-11-24 | 焦点科技股份有限公司 | A kind of multigraph detection method based on SIFT algorithms |
| CN108307146A (en) * | 2017-12-12 | 2018-07-20 | 张宝泽 | A kind of ultra-high-tension power transmission line Security Vulnerability Detecting System and method |
-
2018
- 2018-09-26 CN CN201811122966.7A patent/CN109102036A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103260016A (en) * | 2013-06-04 | 2013-08-21 | 成都思晗科技有限公司 | Remote and intelligent line-tracking method of electric transmission line |
| CN103413150A (en) * | 2013-06-28 | 2013-11-27 | 广东电网公司电力科学研究院 | Power line defect diagnosis method based on visible light image |
| CN106778633A (en) * | 2016-12-19 | 2017-05-31 | 江苏慧眼数据科技股份有限公司 | A kind of pedestrian recognition method based on region segmentation |
| CN107392215A (en) * | 2017-08-02 | 2017-11-24 | 焦点科技股份有限公司 | A kind of multigraph detection method based on SIFT algorithms |
| CN108307146A (en) * | 2017-12-12 | 2018-07-20 | 张宝泽 | A kind of ultra-high-tension power transmission line Security Vulnerability Detecting System and method |
Non-Patent Citations (4)
| Title |
|---|
| J.R.R. UIJLINGS: "Selective Search for Object Recognition", 《INTERNATIONAL JOURNAL OF COMPUTER VISION》 * |
| P. F. FELZENSZWALB，ET.AL: "Efficient Graph-Based Image Segmentation", 《INTERNATIONAL JOURNAL OF COMPUTER VISION》 * |
| 孙苗苗: "基于图像处理的输电线故障巡检技术研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 柳杨: "《数字图像物体识别理论详解与实战》", 31 March 2018 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109992682A (en) * | 2019-03-29 | 2019-07-09 | 联想(北京)有限公司 | A kind of image-recognizing method, device and electronic equipment |
| CN113077481A (en) * | 2021-03-29 | 2021-07-06 | 上海闻泰信息技术有限公司 | Image processing method and device, computer equipment and storage medium |
| CN113077481B (en) * | 2021-03-29 | 2022-12-09 | 上海闻泰信息技术有限公司 | Image processing method and device, computer equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Yu et al. | A new spatial-oriented object detection framework for remote sensing images | |
| Zhao et al. | Cloud shape classification system based on multi-channel cnn and improved fdm | |
| CN109118479B (en) | Capsule network-based insulator defect identification and positioning device and method | |
| Bommes et al. | Computer vision tool for detection, mapping, and fault classification of photovoltaics modules in aerial IR videos | |
| Li et al. | An edge embedded marker-based watershed algorithm for high spatial resolution remote sensing image segmentation | |
| Peng et al. | Drone-based vacant parking space detection | |
| CN107833213A (en) | A kind of Weakly supervised object detecting method based on pseudo- true value adaptive method | |
| CN106610969A (en) | Multimodal information-based video content auditing system and method | |
| CN112560675B (en) | Bird visual target detection method combining YOLO and rotation-fusion strategy | |
| CN109033944B (en) | Method and system for classifying all-sky aurora images and positioning key local structure | |
| CN111402224B (en) | Target identification method for power equipment | |
| CN109034035A (en) | Pedestrian's recognition methods again based on conspicuousness detection and Fusion Features | |
| CN113408584B (en) | RGB-D multi-modal feature fusion 3D target detection method | |
| US10902264B2 (en) | Automatic generation of secondary class annotations | |
| Prusty et al. | Indiscapes: Instance segmentation networks for layout parsing of historical indic manuscripts | |
| CN113158891B (en) | Cross-camera pedestrian re-identification method based on global feature matching | |
| Shu et al. | Center-point-guided proposal generation for detection of small and dense buildings in aerial imagery | |
| CN110059539A (en) | A kind of natural scene text position detection method based on image segmentation | |
| CN106203539A (en) | The method and apparatus identifying container number | |
| Yan et al. | Insulator detection and recognition of explosion based on convolutional neural networks | |
| CN111488911A (en) | Image entity extraction method based on Mask R-CNN and GAN | |
| KR20230098745A (en) | System and method of recogning airport baggage vision sorter | |
| CN109102036A (en) | A kind of image tagged method and device for transmission line malfunction identification | |
| Manninen et al. | Multi-stage deep learning networks for automated assessment of electricity transmission infrastructure using fly-by images | |
| CN111597939B (en) | High-speed rail line nest defect detection method based on deep learning |
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: 20181228 |
|
| RJ01 | Rejection of invention patent application after publication |