CN106340006A - Icing degree assessment method based on insulator image umbrella stretch-out - Google Patents

Abstract

The invention discloses an icing degree assessment method based on insulator image umbrella stretch-out. The method mainly comprises the following steps that S1 image acquisition is performed: insulator images before and after icing are acquired by using a fixed camera; S2 the images before and after icing are segmented by using a GrabCut graph theoretic approach; S3 a convex hull and a convex defect between every two adjacent insulators in the contour of S2 are solved; S4 an umbrella stretch-out equivalent calculation method is performed; and S5 the insulator icing degree is estimated by the change of insulator umbrella stretch-out before and after icing. The icing and bridging situation of each part of the insulator strings can be detected by the change percentage of umbrella stretch-out before and after icing through the images, and the degree of severe icing can be estimated so that the visual, economical and effective insulator icing degree assessment method is provided.

Description

A kind of icing degree assessment method being stretched out based on insulation subgraph umbrella

Technical field

The invention belongs to the on-line monitoring field of image procossing and insulator, more particularly, to it is based on insulation subgraph umbrella and stretches out Icing degree assessment method.

Background technology

Powerline ice-covering can cause insulator rupture, shaft tower to collapse, line tripping etc., and the serious safety threatening electrical network is steady Fixed operation.In ice damage in 2008, the tripping operation of 500kv circuit 58% belongs to insulator rupture.Ice dodges main cause to be had: air and absolutely Electrolyte in edge surface filth increases frozen water electrical conductivity；The blocked up ice bridge that formed of insulator chain icing makes icing flashover voltage reduce.

Pollution severity of insulators depends on many factors, including the temperature of Distribution Area, humidity, wind speed, misty rain, pollution sources Property and quantity and insulator configuration etc., it can be by equivalent salt deposit density (esdd), Leakage Current, surface pollution layer electricity The characteristic quantity such as conductance (splc) and flashover field intensity characterizes.But between sub-pieces the bridge joint degree of icicle preferably characterize and Measuring method.

-- icing insulator bridge joint degree there is no effective detection method, particularly image detecting method at present.Set forth herein On the basis of grabcut method image segmentation, by identifying the convex defect of -- icing insulator string profile, calculate umbrella and stretch out, with umbrella The change stretched out characterizes icicle bridge joint degree axially and radially between sub-pieces.

Content of the invention

It is an object of the invention to overcoming the drawbacks described above that prior art exists, proposing a kind of insulation subgraph umbrella that is based on and stretching The icing degree assessment method going out, concrete technical scheme is as follows.

A kind of icing degree assessment method being stretched out based on insulation subgraph umbrella, it comprises the steps:

S1, Image Acquisition, specifically include:

S1.1, erection camera are monitored to easy icing region insulator, shoot non--- icing insulator photo, are designated as i；

S1.2, camera position and shooting angle constant it is assumed that after icing insulator photo be j；

S2, to before and after icing insulation subgraph i and j split with grabcut graph theoretic approach, obtain insulator before and after icing Profile c_iAnd c_j；

S3 solves insulator contour c_iAnd c_jConvex closure and convex defect；

S4 passes through convex defect and the convex closure solving in s3, and before and after calculating icing, the umbrella of insulator stretches out:

S4.1 excludes the convex defect interference that depth is less than setting value t；

S4.2 profile is by convex defect maximum to remaining left side and two, right side, and its depth is exactly that umbrella stretches out；

The change that s5 is stretched out by insulator umbrella before and after icing, estimates covering ice for insulator degree.

Further, step s2 specifically includes:

S2.1, with rectangle frame, each pixel is referred to gmm first_s, and gmm_t, rectangle inframe is gmm_s, outer rectangular frame is gmm_t, gmm_sThe mixed Gauss model of expression prospect, gmm_tRepresent the mixed Gauss model of background, gauss hybrid models such as formula (1) shown in-(2),

${\mathrm{gmm}}_a=\underset{i=1}{\overset{k}{\σ}}{\mathrm{\π}}_{a,i}{g}_{a,i}(x;{\mathrm{\μ}}_{a,i},{\σ}_{a,i})---\left(1\right)$

$\begin{array}{c}g(z_n;{\mathrm{\μ}}_{a,i},{\σ}_{a,i})\\ =\frac{1}{\sqrt{{\left(2\mathrm{\π}\right)}^d\left|{\σ}_{a,i}\right|}}\exp \[-\frac{1}{2}{(z_n-{\mathrm{\μ}}_{a,i})}^t{{\σ}_{a,i}}^{-1}(z_n-{\mathrm{\μ}}_{a,i})\]\end{array}---\left(2\right)$

Wherein,And 0≤π_i≤ 1, k are gauss hybrid models element number, herein for 3；A is s or t, and a is s When gmm_sThe gauss hybrid models of expression prospect, a is gmm during t_tRepresent the gauss hybrid models of background；π_a,iFor i-th Gaussian mode Type g (z_n；μ_a,i,∑_a,i) in gauss hybrid models gmm_aIn shared weight；X represents pixel to be split；μ_a,i、∑_a,iTable respectively Show average and the variance of i-th Gauss model；

S2.2 clustering algorithm (as k-means) is by gmm_sAnd gmm_tIn pixel basis color gray value be divided into k class；

S2.3 calculates gauss hybrid models parameter μ according to the pixel color gray value in each classification_a,i、∑_a,i；

One capaciated flow network of s2.4 describes image, calculates the capability value capacity v on two class sides_m,n、u_n,sAnd u_n,t；

S2.5 determines the minimal cut of image using maximum-flow algorithm；

If the minimal cut convergence of s2.6 image, segmentation completes, and is prospect inside minimal cut, and outside is background, otherwise Return to step s2.2 to continue executing with until minimal cut convergence.

Further, step s2.4 specifically includes:

The s2.4.1 wherein capacity v of first kind side v_m,nCalculating such as formula (3) shown in:

$v_{m,n}=\mathrm{\γ}\underset{(m,n)\&element;c}{\σ}\exp -\mathrm{\β}\left|\right|z_n-z_m|{|}^2---\left(3\right)$

z_mAnd z_nRepresent the color gray value of pixel m and n respectively, γ represents the preferential journey in Equations of The Second Kind relatively for the first kind Degree, γ is bigger, and the degree of priority on first kind side is higher；C represents a pair of neighborhood territory pixel；Exponential term β is for adapting to image contrast Degree, when the difference of relatively low i.e. a pair of the neighborhood territory pixel of picture contrast is less, then needs larger β to amplify difference, when image pair When higher than degree, then less β is needed to reduce difference, shown in the computing formula such as formula (4) of β:

$\mathrm{\&beta;}={\&lsqb;\frac{2}{n}<\underset{(m,n)\&element;c}{\&sigma;}{(z_m-z_n)}^2>\&rsqb;}^{-1}---\left(4\right)$

Wherein,<>represents that having traveled through all of neighborhood union in image sues for peace, and n represents m, the logarithm of n；

The capacity u on s2.4.2 Equations of The Second Kind side_n,sAnd u_n,tBe calculated as follows:

When n is defined as prospect, u_n,sFor l (desirable l=9 γ, γ define identical with formula (3)), u_n,tFor 0；When n is defined as carrying on the back Scape, u_n,sFor 0, u_n,tFor l；

When n is not determined as background or prospect, u_n,sAnd u_n,tSize with set up gauss hybrid models relevant, u_n,s And u_n,tCalculating see formula (5) and formula (6) respectively:

$\begin{array}{c}{u}_{n,s}=-\log \left({\mathrm{gmm}}_s\right)\\ =-{\mathrm{log\&pi;}}_{s,i}+\frac{1}{2}\log \det {\&sigma;}_{s,i}+\frac{1}{2}{\&lsqb;z_n-{\mathrm{\&mu;}}_{s,i}\&rsqb;}^t{{\&sigma;}_{s,i}}^{-1}\&lsqb;z_n-{\mathrm{\&mu;}}_{s,i}\&rsqb;\end{array}---\left(5\right)$

$\begin{array}{c}{u}_{n,t}=-\log \left({\mathrm{gmm}}_t\right)\\ =-{\mathrm{log\&pi;}}_{t,i}+\frac{1}{2}\log \det {\&sigma;}_{t,i}+\frac{1}{2}{\&lsqb;z_n-{\mathrm{\&mu;}}_{t,i}\&rsqb;}^t{{\&sigma;}_{t,i}}^{-1}\&lsqb;z_n-{\mathrm{\&mu;}}_{t,i}\&rsqb;\end{array}---\left(6\right).$

Further, step s3 specifically includes:

Mono- polygonal convex closure of s3.1 refers to comprise the minimal convex polygon of this polygon, and its convex defect refers to convex Part in addition to former polygon in bag；Three important parameters of convex defect are starting point, terminal, depth；The starting point of convex defect or The vertical range of the deepest point to convex defect for the terminal is exactly the depth of convex defect；

S3.2 assumes a height of h of minimum enclosed rectangle of the insulator contour after s2 segmentation it is assumed that insulator chain piece number is n, Detection height existsExtremelyI.e. often the profile convex closure of adjacent two panels insulator and convex defect, wherein i=0,1 ... n.

Further, step s5 specifically includes:

S5.1 umbrella spacing and umbrella stretch out and are represented with d and p respectively, before and after icing umbrella stretch out p ', p change percentage be

Umbrella before and after s5.2 solves every icing between adjacent two panels insulator stretches out change percentage, and the umbrella asking change maximum is stretched Go out, be designated as △ p_m(%)；

S5.3n represents the piece number of insulator；d_ilAnd d_irRepresent respectively before icing i-th with the left side of i+1 piece insulator and Right side umbrella spacing；d_il＇, d_ir＇ after representing icing respectively corresponding umbrella spacing and umbrella stretch out, bunch of insulator icing forward backward averaging Umbrella stretches out shown in change percentage computational methods such as formula (7),

${\mathrm{\&delta;p}}_a\left(\%\right)=\frac{{\&sigma;}_{i=1}^{n-1}\&lsqb;({p_{i1}}^{\&prime;}+{p_{ir}}^{\&prime;})-(p_i+p_{ir})\&rsqb;}{{\&sigma;}_{i=1}^{n-1}(p_{i1}+p_{ir})}---\left(7\right)$

S5.4 passes through △ p_m(%) it is known that sub-pieces between most serious degree: | △ p_m(%) | bigger, bridge joint is tighter Weight, when | △ p_m(%) | when=100%, bridge completely；

S5.5 passes through △ p_a(%) can know that the average icing order of severity of bunch of insulator: | △ p_a(%) | bigger, bridge joint More serious, when | △ p_a(%) | when=100%, bridge completely.

Compared with prior art, the invention has the advantages that and beneficial effect:

By the change percentage that umbrella before and after image detection icing stretches out, the present invention can detect that each position of insulator chain is covered Ice bridge connects situation, estimates the icing order of severity, is a kind of covering ice for insulator degree assessment method directly perceived, economic, effective.

Brief description

Fig. 1 is the icing degree assessment method being stretched out based on insulation subgraph umbrella in example.

Fig. 2 is grabcut graph theory split plot design flow chart.

Fig. 3 is max-flow little minimal cut schematic diagram.

Fig. 4 is insulator contour convex closure and convex defect schematic diagram.

Specific embodiment

Below in conjunction with example, the present invention is embodied as being described further, but embodiments of the present invention and protection do not limit In this.

As Fig. 1, a kind of icing degree assessment method being stretched out based on insulation subgraph umbrella, comprise the steps:

S1 Image Acquisition, specific as follows:

S1.1 sets up camera and easy icing region insulator is monitored, and shoots non--- icing insulator photo, is designated as i；

S1.2 camera position and shooting angle constant it is assumed that after icing insulator photo be j；

S2 is split with grabcut graph theoretic approach to insulation subgraph i and j before and after icing, obtains insulator before and after icing Profile c_iAnd c_j:

S2.1grabcut graph theory split plot design flow chart is as shown in Fig. 2 each pixel is referred to by circulation first with rectangle frame gmm_s(representing the mixed Gauss model of prospect) and gmm_t(representing the mixed Gauss model of background), rectangle inframe is gmm_s, rectangle Outer frame is gmm_t, shown in gauss hybrid models such as formula (1)-(2),

${\mathrm{gmm}}_a=\underset{i=1}{\overset{k}{\&sigma;}}{\mathrm{\&pi;}}_{a,i}{g}_{a,i}(x;{\mathrm{\&mu;}}_{a,i},{\&sigma;}_{a,i})---\left(1\right)$

$\begin{array}{c}g(z_n;{\mathrm{\&mu;}}_{a,i},{\&sigma;}_{a,i})\\ =\frac{1}{\sqrt{{\left(2\mathrm{\&pi;}\right)}^d\left|{\&sigma;}_{a,i}\right|}}\exp \&lsqb;-\frac{1}{2}{(z_n-{\mathrm{\&mu;}}_{a,i})}^t{{\&sigma;}_{a,i}}^{-1}(z_n-{\mathrm{\&mu;}}_{a,i})\&rsqb;\end{array}---\left(2\right)$

Wherein,And 0≤π_i≤ 1, k are gauss hybrid models element number, herein for 3；A is s or t, and a is s When gmm_sThe gauss hybrid models of expression prospect, a is gmm during t_tRepresent the gauss hybrid models of background；π_a,iFor i-th Gaussian mode Type g (z_n；μ_a,i,∑_a,i) in gauss hybrid models gmm_aIn shared weight；X represents pixel to be split；μ_a,i、∑_a,iTable respectively Show average and the variance of i-th Gauss model.

S2.2 clustering algorithm (as k-means) is by gmm_sAnd gmm_tIn pixel basis color gray value be divided into k class；

S2.3 calculates gauss hybrid models parameter μ according to the pixel color gray value in each classification_a,i、∑_a,i；

S2.4 describes image with a capaciated flow network, as shown in figure 3, calculating the capability value capacity v on two class sides_m,n、u_n,sWith u_n,t:

The s2.4.1 wherein capacity v of first kind side v_m,nCalculating such as formula (3) shown in:

$v_{m,n}=\mathrm{\&gamma;}\underset{(m,n)\&element;c}{\&sigma;}\exp -\mathrm{\&beta;}\left|\right|z_n-z_m|{|}^2---\left(3\right)$

z_mAnd z_nRepresent the color gray value of pixel m and n respectively, γ represents the preferential journey in Equations of The Second Kind relatively for the first kind Degree, γ is bigger, and the degree of priority on first kind side is higher.C represents a pair of neighborhood territory pixel.Exponential term β is for adapting to image contrast Degree, when picture contrast is relatively low, the difference of a pair of neighborhood territory pixel is less, then need larger β to amplify difference, when image pair When higher than degree, then less β is needed to reduce difference, shown in the computing formula such as formula (4) of β:

$\mathrm{\&beta;}={\&lsqb;\frac{2}{n}<\underset{(m,n)\&element;c}{\&sigma;}{(z_m-z_n)}^2>\&rsqb;}^{-1}---\left(4\right)$

Wherein,<>represents that having traveled through all of neighborhood union in image sues for peace, and n represents m, the logarithm of n.

The capacity u on Equations of The Second Kind side in s2.4.2 Fig. 2_n,sAnd u_n,tBe calculated as follows:

When n is defined as prospect, u_n,sFor l (desirable l=9 γ, γ define identical with formula (3)), u_n,tFor 0；When n is defined as carrying on the back Scape, u_n,sFor 0, u_n,tFor l.

When n is not determined as background or prospect, u_n,sAnd u_n,tSize with set up gauss hybrid models relevant, u_n,s And u_n,tCalculating see formula (5) and formula (6) respectively:

$\begin{array}{c}{u}_{n,s}=-\log \left({\mathrm{gmm}}_s\right)\\ =-{\mathrm{log\&pi;}}_{s,i}+\frac{1}{2}\log \det {\&sigma;}_{s,i}+\frac{1}{2}{\&lsqb;z_n-{\mathrm{\&mu;}}_{s,i}\&rsqb;}^t{{\&sigma;}_{s,i}}^{-1}\&lsqb;z_n-{\mathrm{\&mu;}}_{s,i}\&rsqb;\end{array}---\left(5\right)$

$\begin{array}{c}{u}_{n,t}=-\log \left({\mathrm{gmm}}_t\right)\\ =-{\mathrm{log\&pi;}}_{t,i}+\frac{1}{2}\log \det {\&sigma;}_{t,i}+\frac{1}{2}{\&lsqb;z_n-{\mathrm{\&mu;}}_{t,i}\&rsqb;}^t{{\&sigma;}_{t,i}}^{-1}\&lsqb;z_n-{\mathrm{\&mu;}}_{t,i}\&rsqb;\end{array}---\left(6\right)$

S2.5 determines the minimal cut of image using maximum-flow algorithm；

If the minimal cut convergence of s2.6 image, segmentation completes, and is prospect inside minimal cut, and outside is background, otherwise Return to s2.2 to continue executing with until minimal cut convergence.

S3 computer solving insulator contour c_iAnd c_jConvex closure and convex defect:

Mono- polygonal convex closure of s3.1 refers to comprise the minimal convex polygon of this polygon, and its convex defect refers to convex Part in addition to former polygon in bag.Insulator in Fig. 4 (a), its profile, convex closure and convex defect are respectively as Fig. 4 (b) institute Show, black line is insulator contour, green line is insulator contour convex closure, violet region is convex defect.Convex defect have three important Parameter: starting point, terminal, depth.As shown in Fig. 4 (b), the beginning or end for convex defect of Bluepoint mark, red point mark be The deepest point of convex defect, the vertical range of red point to Bluepoint line is exactly the depth of convex defect；

S3.2 assumes a height of h of minimum enclosed rectangle of the insulator contour after s2 segmentation it is assumed that insulator chain piece number is n, Detection height existsExtremelyProfile (often adjacent two panels insulator) convex closure and convex defect, wherein i=0,1 ... n.

S4 passes through convex defect and the convex closure solving in s3, and before and after calculating icing, the umbrella of insulator stretches out:

S4.1 excludes the convex defect interference that depth is less than t；

S4.2 profile is by convex defect maximum to remaining left side and two, right side, and its depth is exactly that umbrella stretches out；

The change that s5 is stretched out by insulator umbrella before and after icing, estimation covering ice for insulator degree:

S5.1 such as umbrella spacing and umbrella are stretched out and are represented with d and p respectively, before and after icing umbrella stretch out change percentage be

Umbrella before and after s5.2 solves every icing between adjacent two panels insulator stretches out change percentage, and the umbrella asking change maximum is stretched Go out, be designated as △ p_m(%)；

S5.3 assumes that n represents the piece number of insulator；d_ilAnd d_irRepresent respectively before icing i-th with the left side of i+1 piece insulator Side and right side umbrella spacing；d_il＇, d_ir＇ after representing icing respectively corresponding umbrella spacing and umbrella stretch out, before and after bunch of insulator icing Average umbrella stretches out shown in change percentage such as formula (7).

${\mathrm{\&delta;p}}_a\left(\%\right)=\frac{{\&sigma;}_{i=1}^{n-1}\&lsqb;({p_{i1}}^{\&prime;}+{p_{ir}}^{\&prime;})-(p_{i1}+p_{ir})\&rsqb;}{{\&sigma;}_{i=1}^{n-1}(p_{i1}+p_{ir})}---\left(7\right)$

S5.4 passes through △ p_m(%) it is known that sub-pieces between most serious degree: | △ p_m(%) | bigger, bridge joint is tighter Weight, when | △ p_m(%) | when=100%, bridge completely；

S5.5 passes through △ p_a(%) it is known that bunch of insulator the average icing order of severity: | △ p_a(%) | bigger, bridge Connect more serious, when | △ p_a(%) | when=100%, bridge completely.

Claims (6)

1. a kind of icing degree assessment method being stretched out based on insulation subgraph umbrella is it is characterised in that comprise the steps:

S1, Image Acquisition, specifically include:

S1.1, erection camera are monitored to easy icing region insulator, shoot non--- icing insulator photo, are designated as i；

S1.2, camera position and shooting angle constant it is assumed that after icing insulator photo be j；

S2, to before and after icing insulation subgraph i and j split with grabcut graph theoretic approach, obtain insulator contour before and after icing c_iAnd c_j；

S3 solves insulator contour c_iAnd c_jConvex closure and convex defect；

S4 passes through convex defect and the convex closure solving in s3, and before and after calculating icing, the umbrella of insulator stretches out:

S4.1 excludes the convex defect interference that depth is less than setting value t；

S4.2 profile is by convex defect maximum to remaining left side and two, right side, and its depth is exactly that umbrella stretches out；

The change that s5 is stretched out by insulator umbrella before and after icing, estimates covering ice for insulator degree.

2. the icing degree assessment method being stretched out based on insulation subgraph umbrella according to claim 1 is it is characterised in that walk Rapid s2 specifically includes:

S2.1, with rectangle frame, each pixel is referred to gmm first_s, and gmm_t, rectangle inframe is gmm_s, outer rectangular frame is gmm_t, gmm_sThe mixed Gauss model of expression prospect, gmm_tRepresent the mixed Gauss model of background, gauss hybrid models such as formula (1)-(2) It is shown,

${\mathrm{gmm}}_a=\underset{i=1}{\overset{k}{\&sigma;}}{\mathrm{\&pi;}}_{a,i}{g}_{a,i}(x;{\mathrm{\&mu;}}_{a,i},{\mathrm{\&sigma;}}_{a,i})---\left(1\right)$

$\begin{array}{c}g(z_n;{\mathrm{\&mu;}}_{a,i},{\mathrm{\&sigma;}}_{a,i})\\ =\frac{1}{\sqrt{{\left(2\mathrm{\&pi;}\right)}^d\left|{\mathrm{\&sigma;}}_{a,i}\right|}}\exp \&lsqb;-\frac{1}{2}{(z_n-{\mathrm{\&mu;}}_{a,i})}^t{{\mathrm{\&sigma;}}_{a,i}}^{-1}(z_n-{\mathrm{\&mu;}}_{a,i})\&rsqb;\end{array}---\left(2\right)$

Wherein,And 0≤π_i≤ 1, k are gauss hybrid models element number, herein for 3；A is s or t, and a is gmm during s_s The gauss hybrid models of expression prospect, a is gmm during t_tRepresent the gauss hybrid models of background；π_a,iFor i-th Gauss model g (z_n；μ_a,i,∑_a,i) in gauss hybrid models gmm_aIn shared weight；X represents pixel to be split；μ_a,i、∑_a,iRepresent the respectively The average of i Gauss model and variance；

S2.2 clustering algorithm is by gmm_sAnd gmm_tIn pixel basis color gray value be divided into k class；

S2.3 calculates gauss hybrid models parameter μ according to the pixel color gray value in each classification_a,i、∑_a,i；

One capaciated flow network of s2.4 describes image, calculates the capability value capacity v on two class sides_m,n、u_n,sAnd u_n,t；

S2.5 determines the minimal cut of image using maximum-flow algorithm；

If the minimal cut convergence of s2.6 image, segmentation completes, and is prospect inside minimal cut, and outside is background, otherwise returns to Step s2.2 continues executing with until minimal cut convergence.

3. the icing degree assessment method being stretched out based on insulation subgraph umbrella according to claim 2 is it is characterised in that walk Rapid s2.4 specifically includes:

The s2.4.1 wherein capacity v of first kind side v_m,nCalculating such as formula (3) shown in:

$v_{m,n}=\mathrm{\&gamma;}\underset{(m,n)\&element;c}{\mathrm{\&sigma;}}\exp -\mathrm{\&beta;}\left|\right|z_n-z_m|{|}^2---\left(3\right)$

z_mAnd z_nRepresent the color gray value of pixel m and n respectively, γ represents the degree of priority in Equations of The Second Kind relatively for the first kind, γ Bigger, the degree of priority on first kind side is higher；C represents a pair of neighborhood territory pixel；Exponential term β is for adapting to image contrast, when The difference of relatively low i.e. a pair of the neighborhood territory pixel of picture contrast is less, then need larger β to amplify difference, when picture contrast relatively Gao Shi, then need less β to reduce difference, shown in the computing formula such as formula (4) of β:

$\mathrm{\&beta;}={\&lsqb;\frac{2}{n}<\underset{(m,n)\&element;c}{\mathrm{\&sigma;}}{\left(z_m-z_n\right)}^2>\&rsqb;}^{-1}---\left(4\right)$

Wherein,<>represents that having traveled through all of neighborhood union in image sues for peace, and n represents m, the logarithm of n；

The capacity u on s2.4.2 Equations of The Second Kind side_n,sAnd u_n,tBe calculated as follows:

When n is defined as prospect, u_n,sFor l, u_n,tFor 0；When n is defined as background, u_n,sFor 0, u_n,tFor l；

When n is not determined as background or prospect, u_n,sAnd u_n,tSize with set up gauss hybrid models relevant, u_n,sAnd u_n,t Calculating see formula (5) and formula (6) respectively:

$\begin{array}{c}{u}_{n,s}=-log\left({\mathrm{gmm}}_s\right)\\ =-{\mathrm{log\&pi;}}_{s,i}+\frac{1}{2}\log {\mathrm{det\&sigma;}}_{s,i}+\frac{1}{2}{\&lsqb;z_n-{\mathrm{\&mu;}}_{s,i}\&rsqb;}^t{{\mathrm{\&sigma;}}_{s,i}}^{-1}\&lsqb;z_n-{\mathrm{\&mu;}}_{s,i}\&rsqb;\end{array}---\left(5\right)$

$\begin{array}{c}{u}_{n,t}=-log\left({\mathrm{gmm}}_t\right)\\ =-{\mathrm{log\&pi;}}_{t,i}+\frac{1}{2}\log {\mathrm{det\&sigma;}}_{t,i}+\frac{1}{2}{\&lsqb;z_n-{\mathrm{\&mu;}}_{t,i}\&rsqb;}^t{{\mathrm{\&sigma;}}_{t,i}}^{-1}\&lsqb;z_n-{\mathrm{\&mu;}}_{t,i}\&rsqb;\end{array}---\left(6\right).$

4. the icing degree assessment method being stretched out based on insulation subgraph umbrella according to claim 2 is it is characterised in that walk Rapid s3 specifically includes:

Mono- polygonal convex closure of s3.1 refers to comprise the minimal convex polygon of this polygon, and its convex defect refers in convex closure Part in addition to former polygon；Three important parameters of convex defect are starting point, terminal, depth；The beginning or end of convex defect To the deepest point of convex defect vertical range be exactly convex defect depth；

S3.2 assumes a height of h of minimum enclosed rectangle of the insulator contour after s2 segmentation it is assumed that insulator chain piece number is n, detection Height existsExtremelyI.e. often the profile convex closure of adjacent two panels insulator and convex defect, wherein i=0,1 ... n.

5. the icing degree assessment method being stretched out based on insulation subgraph umbrella according to claim 2 is it is characterised in that walk Rapid s5 specifically includes:

S5.1 umbrella spacing and umbrella stretch out and are represented with d and p respectively, before and after icing umbrella stretch out p ', p change percentage be

Umbrella before and after s5.2 solves every icing between adjacent two panels insulator stretches out change percentage, and the umbrella asking change maximum stretches out, It is designated as △ p_m(%)；

S5.3n represents the piece number of insulator；d_ilAnd d_irRepresent respectively before icing i-th with the left side of i+1 piece insulator and right side Umbrella spacing；d_il＇, d_ir＇ after representing icing respectively corresponding umbrella spacing and umbrella stretch out, bunch of insulator icing forward backward averaging umbrella is stretched Go out to change shown in percentage computational methods such as formula (7),

${\mathrm{\&delta;p}}_a\left(\%\right)=\frac{{\mathrm{\&sigma;}}_{i=1}^{n-1}\&lsqb;({p_{i1}}^{\&prime;}+{p_{ir}}^{\&prime;})-(p_{i1}+p_{ir})\&rsqb;}{{\mathrm{\&sigma;}}_{i=1}^{n-1}(p_{i1}+p_{ir})}---\left(7\right)$

S5.4 passes through △ p_m(%) it is known that sub-pieces between most serious degree: | △ p_m(%) | bigger, bridge joint is more serious, When | △ p_m(%) | when=100%, bridge completely；

S5.5 passes through △ p_a(%) can know that the average icing order of severity of bunch of insulator: | △ p_a(%) | bigger, bridge joint is tighter Weight, when | △ p_a(%) | when=100%, bridge completely.

6. according to claim 3 based on the insulation icing degree assessment method that stretches out of subgraph umbrella it is characterised in that l =9 γ.