CN106384347A - Hydrophobicity image bright spot detection algorithm - Google Patents

Abstract

The present invention provides a hydrophobicity image bright spot detection algorithm. A camera device is employed to obtain an insulator surface image with water drop, a nonlinear complex diffusion equation is employed to perform complete nonlinear scale space analysis for the hydrophobicity image, and an image sequence from fine to coarse is constructed through the control of the iterations and the iteration speed; and the real part component of the image after the nonlinear complex diffusion processing is close to the linear Gaussian scale space, the imaginary part component of the image is close to the convolution of the real part and the second order smooth Laplace operator, with the traction of the diffusion time, the internal features of the image play a decisive effect on the image evolution, the multi-scale imaginary part image in the nonlinear complex diffusion evolution can effectively reflect the water drop bright spot features of the hydrophobicity image, and finally, the FCM algorithm is employed to extract the water drop bright spot of the hydrophobicity image. The hydrophobicity image bright spot detection algorithm is suitable for the composite insulator hydrophobicity analysis.

Description

A kind of hydrophobicity image bright spot detection algorithm

Technical field

The present invention relates to a kind of hydrophobicity image bright spot detection algorithm, belong to technical field of high voltage.

Background technology

Hydrophobicity of Composite Insulator image comprises abundant information, if rational some spies being provided using image Levy, then utilize artificial intelligence to differentiate that insulator hydrophobicity will be a kind of feasible new approaches.In Hydrophobicity of Composite Insulator In image, due to the physical characteristics of the globule, obvious retroreflective regions occur in the irradiation lower surface of light, i.e. bright spot, and water mark Surface does not have or rare bright spot.If can accurately detecting and extracting globule bright spot, then according to spies such as the geometry of bright spot, statistics Levy, not only can Hydrophobicity of Composite Insulator grade be carried out estimating sentencing, and can be for the quantity of the globule, position and size Scope is determined, and the segmentation further for the globule provides guarantee.

Under the influence of natural environment and anthropic factor, traditional method can not detect well and obtain hydrophobicity image water Pearl spot zone, needs to introduce new mathematical method and improves.Found by many experiments, for no matter clean picture or have Spot picture, after developing through non-linear multiple diffusion model, imaginary part image can effectively reflect hydrophobicity image bright spot.Compared to linear Multiple diffusion model, non-linear multiple diffusion model retains inherently to while similar ramped shaped globule bright spot smooth noise because it has The feature of architectural feature and show more excellent Detection results, therefore this algorithm propose a kind of based on non-linear multiple diffusion model Hydrophobicity image globule bright spot detection algorithm.

Content of the invention

The purpose of the present invention is, for Hydrophobicity of Composite Insulator image traditional technique in measuring and acquisition hydrophobicity image The problem that globule spot zone exists, the present invention discloses a kind of hydrophobicity image bright spot detection algorithm.

Realization the technical scheme is that, is obtained by camera installation and is covered with globule insulator surface image, using non- Linearly answer diffusion equation and hydrophobicity image carried out with a complete Nonlinear Scale Space Theory analysis, by iterationses with repeatedly For the control of speed, constitute by being fine to rough image sequence；After non-linear multiple DIFFUSION TREATMENT image real component close to Linear Gaussian scale-space, and imaginary is similar to real part and smooths Laplace operator convolution with second order, with diffusion time Passage, image internal characteristicses play decisive role to pattern evolution, the multiple dimensioned imaginary part figure under non-linear multiple diffusion is developed As can effectively reflect hydrophobicity image globule bright spot feature, FCM algorithm is finally adopted to extract hydrophobicity image globule bright spot.

Described non-linear multiple diffusion equation is：

$\left\{\begin{array}{c}\frac{\∂I(x,y,t)}{\∂t}=div\left(c\right(\left|I_m\right|)\▿I))\\ I(x,y,0)=I_0\\ c\left(I_m\right)=\frac{e^{j\mathrm{\θ}}}{1+{\left(\frac{\mathrm{Im}\left(I\right)}{k\mathrm{\θ}}\right)}^2}\end{array}\right.$

Wherein：I₀For hydrophobicity original image；I_mFor hydrophobicity imaginary part image；c(I_m) for spreading variable；K joins for threshold value Number；θ is phase angle；X is abscissa；Y is vertical coordinate；T is the evolution time.

Imaginary part image following formula under the described non-linear multiple multiple yardstick of diffusion equation represents：

$V_i={\left\{\frac{\mathrm{Im}\left(I\right)}{t\mathrm{\θ}}\right\}}_i,i=1,2,...;$

Wherein, V_iFor imaginary part image under the i-th yardstick.

Max pixel value in imaginary part image sequence is：

V_MAX=max (V_i), i=1,2 ...；

Wherein, V_MAXFor the max pixel value image in imaginary part image sequence.

The invention has the beneficial effects as follows, the present invention is obtained by camera installation and is covered with globule insulator surface image, adopts Non-linear multiple diffusion equation carries out a complete Nonlinear Scale Space Theory analysis to hydrophobicity image, by iterationses with The control of iteration speed, is constituted by being fine to rough image sequence.After non-linear multiple DIFFUSION TREATMENT, image real component is close In linear Gaussian scale-space, and imaginary is similar to real part and smooths Laplace operator convolution with second order, with diffusion time Passage, image internal characteristicses play decisive role to pattern evolution, the multiple dimensioned imaginary part under non-linear multiple diffusion is developed Image can effectively reflect hydrophobicity image globule bright spot feature, finally adopts FCM algorithm to extract the hydrophobicity image globule bright Point.

The inventive method is applied to Hydrophobicity of Composite Insulator analysis.

Brief description

Fig. 1 hydrophobicity image bright spot detection algorithm flow chart；

Fig. 2 difference hydrophobicity rank hydrophobicity image；

Fig. 2 (a), Fig. 2 (b), Fig. 2 (c), Fig. 2 (d) and Fig. 2 (e) are respectively HC1, HC2, HC3, HC4 and HC5 level hydrophobicity Image；

Fig. 3 final imaginary part image after nonlinear diffusion equations evolution；

Fig. 3 (a), Fig. 3 (b), Fig. 3 (c), Fig. 3 (d) and Fig. 3 (e) respectively HC1, HC2, HC3, HC4 and HC5 level are finally empty Portion's image；

Imaginary part image after Fig. 4 application FCM algorithm classification；

After Fig. 4 (a), Fig. 4 (b), Fig. 4 (c), Fig. 4 (d) and Fig. 4 (e) respectively HC1, HC2, HC3, HC4 and HC5 level classification Imaginary part image；

Fig. 5 hydrophobicity image bright spot extracts result；

Fig. 5 (a), Fig. 5 (b), Fig. 5 (c), Fig. 5 (d) and Fig. 5 (e) are respectively HC1, HC2, HC3, HC4 and HC5 level binary map Picture.

Specific embodiment

The specific embodiment of hydrophobicity image bright spot detection algorithm is as follows：

(1) compound inslation subsample is sprayed water, using camera installation, the surface of composite insulator being covered with the globule is clapped According to.

(2) on single yardstick, imaginary part image can not detect all of bright spot effectively, is owned according to hydrophobicity image Globule highlight structure size provides a corresponding change yardstick, is hated using non-linear multiple diffusion model iteration under different scale Aqueouss image, non-linear multiple diffusion equation such as formula 1, form the imaginary part image under multiple yardsticks, such as formula 2.

$\left\{\begin{array}{c}\frac{\∂I(x,y,t)}{\∂t}=div\left(c\right(\left|I_m\right|)\▿I))\\ I(x,y,0)=I_0\\ c\left(I_m\right)=\frac{e^{j\mathrm{\θ}}}{1+{\left(\frac{\mathrm{Im}\left(I\right)}{k\mathrm{\θ}}\right)}^2}\end{array}\right.---\left(1\right)$

Wherein：I₀For hydrophobicity original image；

I_mFor hydrophobicity imaginary part image.

$V_i={\left\{\frac{\mathrm{Im}\left(I\right)}{t\mathrm{\θ}}\right\}}_i,i=1,2,...---\left(2\right)$

(3) calculate max pixel value in imaginary part image sequence, such as formula 3, obtain globule bright spot under different scale in void Clearly represent in portion's image, form final imaginary part image

V_MAX=max (V_i), i=1,2 ... (3)

(4) imaginary part image intensity value substantially divides three levels, spot zone, gray area, and dark areas, using fuzzy clustering Target is divided into three classes by FCM algorithm, and multiple dimensioned imaginary part image is split, and when algorithmic statement, obtains in all kinds of clusters Fuzzy clustering result, for all kinds of degrees of membership, is finally carried out de-fuzzy, fuzzy clustering is changed into by the heart and each sample Definitiveness is classified, and realizes final cluster segmentation, and then extracts globule bright spot.

Claims (4)

1. a kind of hydrophobicity image bright spot detection algorithm is covered with the globule it is characterised in that described algorithm is obtained by camera installation Insulator surface image, carries out a complete Nonlinear Scale Space Theory using non-linear multiple diffusion equation and divides to hydrophobicity image Analysis, by the control to iterationses and iteration speed, composition is by being fine to rough image sequence；Non-linear multiple DIFFUSION TREATMENT Image real component is close to linear Gaussian scale-space afterwards, and imaginary is similar to real part and smooths Laplace with second order calculate Sub- convolution, with the passage of diffusion time, image internal characteristicses play decisive role to pattern evolution, in non-linear multiple diffusion Multiple dimensioned imaginary part image under developing can effectively reflect hydrophobicity image globule bright spot feature, is finally carried using FCM algorithm Take hydrophobicity image globule bright spot.

2. according to claim 1 a kind of hydrophobicity image bright spot detection algorithm it is characterised in that described non-linear multiple spread Equation is：

$\left\{\begin{array}{c}\frac{\∂I(x,y,t)}{\∂t}=div\left(c\left(\left|I_m\right|\right)\▿I\right))\\ I(x,y,0)=I_0\\ c\left(I_m\right)=\frac{e^{j\mathrm{\θ}}}{1+{\left(\frac{\mathrm{Im}\left(I\right)}{k\mathrm{\θ}}\right)}^2}\end{array}\right.$

Wherein, I₀For hydrophobicity original image；I_mFor hydrophobicity imaginary part image；c(I_m) for spreading variable；K is threshold parameter；θ is Phase angle；X is abscissa；Y is vertical coordinate；T is the evolution time.

3. according to claim 2 a kind of hydrophobicity image bright spot detection algorithm it is characterised in that described non-linear multiple spread Imaginary part image following formula under the multiple yardstick of equation represents：

$V_i={\left\{\frac{Im\left(I\right)}{t\mathrm{\θ}}\right\}}_i,i=1,2,...;$

Wherein, V_iFor imaginary part image under the i-th yardstick.

4. according to claim 3 a kind of hydrophobicity image bright spot detection algorithm it is characterised in that described imaginary part image sequence In max pixel value be：

V_MAX=max (V_i), i=1,2 ...；

Wherein, V_MAXFor the max pixel value in imaginary part image sequence.