CN107644428A - A kind of transmission line of electricity floating object image partition method remapped based on multiple domain degree - Google Patents
A kind of transmission line of electricity floating object image partition method remapped based on multiple domain degree Download PDFInfo
- Publication number
- CN107644428A CN107644428A CN201710904154.7A CN201710904154A CN107644428A CN 107644428 A CN107644428 A CN 107644428A CN 201710904154 A CN201710904154 A CN 201710904154A CN 107644428 A CN107644428 A CN 107644428A
- Authority
- CN
- China
- Prior art keywords
- domain
- image
- edge
- intensity
- point
- 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
Abstract
Method combination frequency domain and intensity domain provided by the invention carry out rim detection, it is thus possible to overcome the not high technological deficiency of the Detection accuracy present in exclusive use frequency domain/intensity domain progress rim detection, enhance the robustness of detection.
Description
Technical field
The present invention relates to ultra-high-tension power transmission line area of maintenance, more particularly, to it is a kind of remapped based on multiple domain degree it is defeated
Electric line floating object image partition method.
Background technology
The quantity size of transmission line of electricity front end camera is growing, and currently available technology mainly uses and manually checks video
Mode transmission line of electricity is monitored.In the case of long-term monitoring, this mode can cause that monitoring personnel is tired, and efficiency is low
Under, it is impossible to transmission line of electricity and the abnormal conditions that periphery occurs are found in time.It is quick in image processing techniques and mode identification technology
In the case of development, it is inexorable trend to be substituted personnel using intelligent inspection and maked an inspection tour.
The important hidden danger for hanging that thing is influence transmission line safety operation of floaing occurred on transmission line of electricity, is wound on power transmission line
Float and hang thing and easily cause the failures such as transmission line of electricity phase fault, single-phase earthing, or even cause electric grid large area power cut accident.Cause
This, float the object for hanging that thing is power transmission line intelligent video analytic system key monitoring.
Due to floaing, extension thing shape is not fixed, and the identification in natural environment has very big difficulty, it is necessary to which certain algorithm will
The extension thing that floats is split.Prior art typically by the extraction of thick edge and build signal intensity quantile estimate rectify a deviation come
Complete to float and hang the segmentation of thing.
During edge extracting, there are two class edge extracting methods:Intensity domain and frequency domain.Intensity domain method can be fine
Edge of the description with gradient characteristics, but the edge of low signal areas can be lost.Frequency domain method due to unrelated with intensity,
The edge of low contrast regions can be handled well, but artifact is also easily produced and responded by mistake.
The content of the invention
The present invention can not take into account the technological deficiency of low signal areas and artifact for the dividing method that solution prior art provides,
Provide a kind of transmission line of electricity floating object image partition method remapped based on multiple domain degree.
To realize above goal of the invention, the technical scheme of use is:
A kind of transmission line of electricity floating object image partition method remapped based on multiple domain degree, is comprised the following steps:
S1. the area-of-interest in image is extracted based on center line normal direction;
S2. the edge in image intensity domain is extracted using Sobel algorithms, then using phase equalization method to figure
As the edge of frequency domain is extracted;
S3. intensity domain and frequency domain are obtained most by calculating maximum similarity of the image under intensity domain and frequency domain
Good weight coefficient, it is then based on optimal weight coefficient and the edge in the image intensity domain of extraction, the edge in picture frequency domain is carried out
Weighting merges, and the new edge strength figure after merging is merged with area-of-interest, in the area-of-interest after fusion
Carry out the extraction of thick edge point;
S4. for each thick edge point of extraction, the marginal point and the average distance of front and rear each n thick edge point are calculated
Then maximum judges whether the deviation measurement of thick edge point is more than set threshold value as measurement is deviateed, if then should
Thick edge point is deleted, and otherwise retains the thick edge point;
S5. the quantile estimate correction based on intensity is carried out to thick edge point, the thick edge point after correction is then based on and obtains
To segmentation result.
Compared with prior art, the beneficial effects of the invention are as follows:
Method combination frequency domain and intensity domain provided by the invention carry out rim detection, it is thus possible to which overcoming individually makes
With the not high technological deficiency of the Detection accuracy present in frequency domain/intensity domain carries out rim detection, the stalwartness of detection is enhanced
Property.
Brief description of the drawings
Fig. 1 is the schematic flow sheet of method.
Fig. 2 is the example schematic of segmentation.
Embodiment
Accompanying drawing being given for example only property explanation, it is impossible to be interpreted as the limitation to this patent;
Below in conjunction with drawings and examples, the present invention is further elaborated.
Embodiment 1
As shown in figure 1, a kind of transmission line of electricity floating object image partition method remapped based on multiple domain degree, including following step
Suddenly:
S1. the area-of-interest in image is extracted based on center line normal direction;
S2. the edge in image intensity domain is extracted using Sobel algorithms, then using phase equalization method to figure
As the edge of frequency domain is extracted;
S3. intensity domain and frequency domain are obtained most by calculating maximum similarity of the image under intensity domain and frequency domain
Good weight coefficient, it is then based on optimal weight coefficient and the edge in the image intensity domain of extraction, the edge in picture frequency domain is carried out
Weighting merges, and the new edge strength figure after merging is merged with area-of-interest, in the area-of-interest after fusion
Carry out the extraction of thick edge point;
S4. for each thick edge point of extraction, the marginal point and the average distance of front and rear each n thick edge point are calculated
Then maximum judges whether the deviation measurement of thick edge point is more than set threshold value as measurement is deviateed, if then should
Thick edge point is deleted, and otherwise retains the thick edge point;
S5. the quantile estimate correction based on intensity is carried out to thick edge point, the thick edge point after correction is then based on and obtains
To segmentation result.
Fig. 2 is the schematic diagram of obtained segmentation result.
Wherein, the detailed process principle extracted using Sobel algorithms to the edge in image intensity domain is as follows:
There can be the thinking of larger signal intensity difference based on adjacent edges, the operator makees intensity gradient in certain neighborhood
For estimating for edge strength.Specifically, it does convolution using 23 × 3 templates with original image, for calculated level and hangs down
Derivative in straight both direction.The calculating gradient of pixel interpolation can be avoided using 3 × 3 template size.If I is original graph
Picture, GxAnd GyIt is according to the image that both horizontally and vertically derivative calculations go out, is defined as follows:
Wherein * represents two-dimensional convolution calculating.
Again because Sobel operators can be decomposed into an average and a differential core, make it significantly more efficient calculating,
It is as follows:
Then GxAnd GxIt can be calculated by following formula convolution:
The intensity level of final each pixel is obtained by merging the intensity level of horizontal and vertical directions:
And the direction of the pixel can be calculated:
Above-mentioned processing is carried out to each pixel in image, obtains the intensity level of each pixel, is then judged every
Whether the intensity level of individual pixel is more than set threshold value, if so, the pixel to be then defined as to the edge in image intensity domain
Point;Thus the extraction at image intensity domain edge is completed.
But from definition as can be seen that for the unconspicuous region of strength difference, the Sobel operators based on gradient are not
Good judgement can be made.
Frequency domain part carries out edge extracting using phase equalization method.What is be based on due to this method is each pixel
Phase information on dot frequency domain, rather than gray-scale intensity is directly handled, so being a kind of and brightness and contrast
All unrelated edge extracting method.Its principle is that foundation phase on marginal position can have the uniformity of height, and this aobvious
Work property is just consistent with judgement of the human eye vision to edge.
It mainly to each composition decompose by Fourier transform first obtains phase, and decomposition formula is described as:
Wherein f (x) is Fourier space, AnIt is the amplitude of the n-th Fourier components, φn(x)=nwx+ φ are represented at x
Local phase.There is maximum phase uniformity, phase equalization in each component of Fourier according to principle marginal point interested
Formula be defined as follows:
In order that PC (x) obtains maximum, the θ in formula takes Local Phase when each component of Fourier has maximum at x
The weighted average of parallactic angle, i.e.,:Again because cos (x) is approximately equal to 1-x in Fourier space2/ 2, institute
To be exactly the position with minimum Local Phase potential difference with the consistent position of height phase.T is a noise threshold in formula, and ε is
One small real number, to avoid denominator from 0 situation occur.Formula (7) can be changed into:
But there is great difficulty by calculating of the formula (8) to phase equalization, with searching local energy peak value come generation
The method calculated for phase equalization, i.e. local energy be equal to phase equalization and Fourier component amplitude and product be:
It can be seen that local energy is directly proportional to phase equalization in formula (9), i.e., the peak value of local energy just etc. is all phase
The maximum of bit integrity.So as to which the calculating of phase equalization is converted into the calculating on local energy.Local energy letter herein
Several peak Es can be by doing convolution to calculate with orthogonal filter in the range of spatial domain:
Wherein F (x) is the part for removing DC component, and H (x) is results of the F (x) after Hilbert is converted, then PC
(x) it transform as:
In addition, the orthogonal filter introduced herein is using log Gabor as wavelet basis wave filter.The present invention usesWith
The odd, even orthogonal logarithm of the log Gabor wavelets in the case where yardstick is n is represented, then F (x) and H (x) can be expressed as signal I (x)
With the convolution between them and:
And frequency intensity in F (x) and it is:
Phase equalization is calculated by the flexible method, amount of calculation can greatly be reduced.
Because both intensity domain and frequency domain strength calculation methods are different, intensity dimension is also different because obtained from, so
Can not directly it be added.In order to obtain most suitable merging yardstick, by calculating the image maximum similarity under two kinds of information fields
To obtain optimum coefficient.Here entered using the improved form of Baddeley Delta Metric (BDM) measurements as distance
OK, in the form of adapting to gray scale:
Wherein w1And w2It is the weight coefficient between two domains,It is used to adjust point-to-point strong
Difference is spent, d is distance function of this between gray value and edge strength.
Finally, the new edge strength figure being remapped to the weighting merging of obtained weights is extracted interested with previous step
Region blends, and thick edge point position is that point for being defined as having in each subtemplate region maximum.
In previous step, because criterion is only the maximum in region of interest, so some of which is artifact
Edge and and it is non-real float hang thing edge, these point just simultaneously is defined as abnormity point.In addition, apart from correction in some utilizations
Strength relationship can not obtain the region of satisfactory result.In order to which metric point deviates situation, pass through the distance difference between neighboring edge point
Degree is differentiated.Distance herein is defined as the length between region template inward flange point and corresponding centerline points.And
Distance difference degree deviates metric function to weigh by defining an edge point position.In view of artifact or noise spot be difficult
The continuity kept within the scope of one, n continuity point is differentiated before and after the point by reference herein.The point is with before
The maximum of the average distance of n point and the average distance of n point after the point is measured as deviation:
Wherein DcRepresent current point, ipreAnd isucRepresent its forerunner's point and follow-up point.
During abnormity point is distinguished, a threshold value is set to divide them.Threshold value herein need not accurately be set,
But what should be set is as far as possible big because erroneous judgement correctly influence caused by point of the point than introducing mistake it is small more.
During abnormity point is rectified a deviation, the correction value of those abnormity points is set to the average of front and rear point:
WhereinAnd DpRepresent normal point set and the distance corresponding to the p of its midpoint.| | for of set interior element
Number.
It is now strong come the image for building central area image intensity with being put on edge by application multiple linear regression application
Linear relationship between degree, the correction in intensity is carried out to the edge after distance correction.
Existing many widely used homing methods at present, such as ridge regression, linear multiple regression etc., but its major part
It is all based on conditional mean model.As in past over half a century, using least square method and its derivation method to even
It is considered as the statistical tool that Modern Heavy is wanted that continuous property dependent variable and the relation of independent variable, which carry out regression modeling,.But in actual conditions
In, homogeneity of variance is assumed often to be breached so that the original hypothesis of model can not be met.Also, only consider average
Position can lose the useful information of dependent variable distribution.In a word, just because of conditional mean is deep by the interference of different value point, so it is to center
The measurement of position is inappropriate and with misleading.
However, quantile estimate can be by ignoring being carried significantly it is assumed that allowing it to return robustness for different value point homogeneity of variance
It is high.Provide any real-valued random variable X and its distribution function:
F (x)=Prob (X≤x) (18)
As any 0 < τ < 1:
Q-1(τ)=inf { x:F (x) >=τ }, 0 < τ < 1 (19)
This is referred to as X the τ points position, Q-1(0.5) point position in being.Quantile estimate is described referring to mean regression
Solution:
The basic thought of regression analysis is exactly to make sample value and the distance between match value most short, random for one group of Y
Sample { y1,y2,L yn, sample for reference mean regression is the method for making error sum of squares minimum, i.e.,:
Wherein given x y conditional mean ξ (x)=xTβ, wherein regression coefficient β can be estimated by following formula:
Similar, it is to make weighted error absolute value sum minimum that τ quantiles sample fractiles, which return, and is:
Above formula can be also expressed equivalently as:
Wherein ρ is an inspection function (check function), is defined as:
Its τ condition quantiles function representation is Qy(τ | x)=x ' β (τ),Referred to as q quantile estimates coefficient, and can
To be estimated by following formula:
Due to carrying absolute value in quantile estimate, thus be not available for differentiating, generally with linear programming technique come pairEstimated.
Using Technologies on Quantile Regression, the robustness of regression parameter estimation can be greatly improved.Because divide position
Number regression estimates value is not influenceed by different value point, and its robustness is obtained by minimizing the property of distance function in equation (24)
Enhancing.For example, if the dependent variable corresponding to some data point of modification on fitting quantile estimate line (under or)
Value, as long as this data point is still on the tropic (under or), this fitted regression line will keep constant.Change speech
It, if changing the value of dependent variable in the case of the sign symbol of residual error is not changed, this fit line will keep constant.This
In the case of, as single argument quantile, the influence of different value point is extremely limited.In addition, quantile estimate is false to distribution
If robustness, be because its estimator relies more heavily on distribution near specific quantile, rather than away from dividing position
Several distribution situation, so the inferential statistics value of quantile estimate is not influenceed by distribution situation.
In addition, being only fitted from the method for parameter estimation based on common least square, a curve institute is different, and quantile is returned
Cluster curve can be fitted by returning, can be more complete when distribution of the independent variable to the dependent variable of different piece produces Different Effects
The general characteristics for portraying condition distribution in face.
Obviously, the above embodiment of the present invention is only intended to clearly illustrate example of the present invention, and is not pair
The restriction of embodiments of the present invention.For those of ordinary skill in the field, may be used also on the basis of the above description
To make other changes in different forms.There is no necessity and possibility to exhaust all the enbodiments.It is all this
All any modification, equivalent and improvement made within the spirit and principle of invention etc., should be included in the claims in the present invention
Protection domain within.
Claims (2)
- A kind of 1. transmission line of electricity floating object image partition method remapped based on multiple domain degree, it is characterised in that:Including following step Suddenly:S1. the area-of-interest in image is extracted based on center line normal direction;S2. the edge in image intensity domain is extracted using Sobel algorithms, then using phase equalization method to image frequency Extracted at the edge in rate domain;S3. the best weights of intensity domain and frequency domain are obtained by calculating maximum similarity of the image under intensity domain and frequency domain Weight coefficient, is then based on optimal weight coefficient and the edge in the image intensity domain of extraction, the edge in picture frequency domain is weighted Merge, and the new edge strength figure after merging is merged with area-of-interest, carried out in the area-of-interest after fusion The extraction of thick edge point;S4. for each thick edge point of extraction, the maximum of the marginal point and the average distance of front and rear each n thick edge point is calculated Then value judges whether the deviation measurement of thick edge point is more than set threshold value as measurement is deviateed, if then by the thick side Edge point is deleted, and otherwise retains the thick edge point;S5. the quantile estimate correction based on intensity, the thick edge point minute being then based on after correction are carried out to thick edge point Cut result.
- 2. the transmission line of electricity floating object image partition method according to claim 1 remapped based on multiple domain degree, its feature It is:The detailed process that the step S2 is extracted using Sobel algorithms to the edge in image intensity domain is as follows:Make in image Pixel be expressed as I,ThenWherein Gx、GyThe intensity level of pixel horizontal and vertical directions is represented respectively, then the intensity level of pixel is:Above-mentioned processing is carried out to each pixel in image, the intensity level of each pixel is obtained, then judges each picture Whether the intensity level of vegetarian refreshments is more than set threshold value, if so, the pixel to be then defined as to the marginal point in image intensity domain;By This completes the extraction at image intensity domain edge.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710904154.7A CN107644428A (en) | 2017-09-29 | 2017-09-29 | A kind of transmission line of electricity floating object image partition method remapped based on multiple domain degree |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710904154.7A CN107644428A (en) | 2017-09-29 | 2017-09-29 | A kind of transmission line of electricity floating object image partition method remapped based on multiple domain degree |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107644428A true CN107644428A (en) | 2018-01-30 |
Family
ID=61122855
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710904154.7A Pending CN107644428A (en) | 2017-09-29 | 2017-09-29 | A kind of transmission line of electricity floating object image partition method remapped based on multiple domain degree |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107644428A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112801047A (en) * | 2021-03-19 | 2021-05-14 | 腾讯科技(深圳)有限公司 | Defect detection method and device, electronic equipment and readable storage medium |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104835152A (en) * | 2015-04-27 | 2015-08-12 | 国家电网公司 | Processing method and system of power transmission line inspection images |
CN105791862A (en) * | 2016-03-21 | 2016-07-20 | 杭州电子科技大学 | Three-dimensional video coding depth map internal mode selection method based on edge complexity |
US20170061246A1 (en) * | 2015-09-02 | 2017-03-02 | Fujitsu Limited | Training method and apparatus for neutral network for image recognition |
-
2017
- 2017-09-29 CN CN201710904154.7A patent/CN107644428A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104835152A (en) * | 2015-04-27 | 2015-08-12 | 国家电网公司 | Processing method and system of power transmission line inspection images |
US20170061246A1 (en) * | 2015-09-02 | 2017-03-02 | Fujitsu Limited | Training method and apparatus for neutral network for image recognition |
CN105791862A (en) * | 2016-03-21 | 2016-07-20 | 杭州电子科技大学 | Three-dimensional video coding depth map internal mode selection method based on edge complexity |
Non-Patent Citations (2)
Title |
---|
李致勋等: "Automatic coronary artery segmentation based on multi-domains remapping and quantile regression in angiographies", 《COMPUTERIZED MEDICAL IMAGING AND GRAPHICS》 * |
钱青 等: "一种改进的Sobel算子图像清晰度评价函数", 《计算机与数字工程》 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112801047A (en) * | 2021-03-19 | 2021-05-14 | 腾讯科技(深圳)有限公司 | Defect detection method and device, electronic equipment and readable storage medium |
CN112801047B (en) * | 2021-03-19 | 2021-08-17 | 腾讯科技(深圳)有限公司 | Defect detection method and device, electronic equipment and readable storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106778668B (en) | A kind of method for detecting lane lines of robust that combining RANSAC and CNN | |
Koster et al. | MIR: An approach to robust clustering-application to range image segmentation | |
CN104463795B (en) | A kind of dot matrix DM image in 2 D code processing method and processing device | |
CN103455991B (en) | A kind of multi-focus image fusing method | |
CN106407928B (en) | Transformer composite insulator casing monitoring method and system based on raindrop identification | |
CN105913415A (en) | Image sub-pixel edge extraction method having extensive adaptability | |
CN106530347A (en) | Stable high-performance circle feature detection method | |
Van de Weijer et al. | Curvature estimation in oriented patterns using curvilinear models applied to gradient vector fields | |
Hou et al. | Welding image edge detection and identification research based on canny operator | |
CN111126116A (en) | Unmanned ship river channel garbage identification method and system | |
CN109766838A (en) | A kind of gait cycle detecting method based on convolutional neural networks | |
CN103150725B (en) | Based on SUSAN edge detection method and the system of non-local mean | |
CN109685733A (en) | A kind of lead zinc floatation foam image space-time joint denoising method based on bubble motion stability analysis | |
CN109447036A (en) | A kind of segmentation of image digitization and recognition methods and system | |
CN104537381A (en) | Blurred image identification method based on blurred invariant feature | |
CN102005049B (en) | Unilateral generalized gaussian model-based threshold method for SAR (Source Address Register) image change detection | |
CN106254723B (en) | A kind of method of real-time monitoring video noise interference | |
CN107644428A (en) | A kind of transmission line of electricity floating object image partition method remapped based on multiple domain degree | |
CN105809085A (en) | Human eye positioning method and device | |
CN102201060A (en) | Method for tracking and evaluating nonparametric outline based on shape semanteme | |
CN111402256B (en) | Three-dimensional point cloud target detection and attitude estimation method based on template | |
CN105787432A (en) | Method for detecting human face shielding based on structure perception | |
Yutian et al. | A combined eye states identification method for detection of driver fatigue | |
CN108776968B (en) | SAR image change detection method based on deep forest | |
Wan et al. | A novel image segmentation algorithm via neighborhood principal component analysis and laplace operator |
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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180130 |