CN105160669A - Method for detecting and locating insulator defects in power transmission line image via a drone - Google Patents

Abstract

The present invention discloses a method for detecting and locating insulator defects in power transmission line image via a drone, and belongs to the technical field of image extraction and recognition. The detection and location method for the insulator defects comprises the steps of: S1. converting an aerial image from an RGB color space into an HSI (Hue-Saturation-Lightness) space, and carrying out binarization on extracted H component image and S component image separately to obtain a binary image of a preliminary contour of an insulator string; S2. using an ant colony algorithm based on a particle swarm optimization parameter to extract a contour of an insulator monomer; and S3. using a least square method to perform ellipse fitting on the contour of the insulator monomer, and locating the insulator defects by detecting the space between the contours of insulator monomers. By using the technical scheme provided by the invention, the detection precision of the insulator defects can be significantly improved, background interferences are reduced, and the operation speed is relatively high.

Description

A kind of unmanned plane patrols and examines the detection and positioning method of defects of insulator in transmission line of electricity image

Technical field

The invention belongs to image zooming-out and recognition technology field, more particularly, relate to a kind of detection and positioning method that unmanned plane patrols and examines defects of insulator in transmission line of electricity image.

Background technology

Along with Chinese national economy sustained and rapid development and urban construction scale expanding day, the industry fast and dense development such as high-tech industry, finance, health care, the demand of electric power energy is strengthened day by day, and economic development not only makes urban and rural power grids load increase fast, also the reliability of powering and power supply quality are had higher requirement.Therefore, Utilities Electric Co. needs to carry out regular visit to power circuit equipment especially line of electric force and electric force pole tower, to ensure the stability and safety operation of electrical power transmission system and normally carrying out of social production life.The power circuit corridor of China, often need the geographical environment passing through various complexity, frequent through lake and reservoir and high and steep mountains etc., therefore, transmission line of electricity just has that coverage is large, distributed areas are wide, transmission range is long, geographical conditions are complicated and changeable and affects the feature such as significantly by amblent air temperature, thus brings great challenge to the day-to-day operation of circuit, maintenance and maintenance.

The tour of China's transmission line of electricity generally adopts manual patrol mode, though this method is simple, efficiency is lower, and the cycle is longer, and needs to be equipped with a large amount of optical device and high, the veteran track walker of quality, higher to the requirement of manpower, financial resources.And when shaft tower is higher, around geographical environment is more complicated, artificial line walking is just more difficult, easily omits fault, causes line walking not thorough, thus make artificial line walking mode be difficult to meet the operation maintenance requirement of high-voltage fence gradually.

From last century the nineties, more American-European developed countries just attempt unmanned plane to be applied to the work such as transmission line of electricity repairing, this technology relative maturity so far.Helicopter routing inspection transmission line of electricity technology, have safe and efficient, by region restriction little, the advantages such as fault can be found fast.China attempts helicopter walking operation from the later stage in last century, and in recent years, China also increased the Innovation Input of unmanned plane line walking technology, and Shandong Electric Power Group in 2012 takes the lead in the whole nation achieving includes unmanned plane line walking in line data-logging normalization application.2013, " power transmission line in high altitude region nobody patrol and examine the applied research of technology " born by overhauling company of national grid Qinghai Electric Power Corporation smoothly by the examination of State Grid Corporation of China, and by the qualification of Science and Technology Department of Qinghai Province.

Insulator is the one of isolation electrical equipment, plays support wire and prevent electric current from returning the effect on ground in transmission line of electricity.Owing to be exposed to for a long time in air and to be operated in the rugged surroundings such as highfield, snow and rain mist, chemicals attack, add the reasons such as materials own, manufacture craft level and artificial destruction, insulator is difficult to invariably electric fault to occur.The electrical resistance fault of insulator mainly contains flashover and self-destruction two kinds, flashover occurs in insulator surface, burn vestige can be seen, usually insulating property are not lost, and self-destruction causes due to insulator manufacturing process and surrounding Environmental variations mostly, insulator self-destruction disappearance can badly influence the safety and effectiveness that transmission line of electricity runs, and likely causes loss difficult to the appraisal.Therefore, how can accurately detect insulator and identify its electric fault in time from the Aerial Images of background complexity, especially identify that its self-destruction fault is just particularly important.Both at home and abroad about utilizing unmanned plane to carry out, research that transmission line of electricity robotization patrols and examines is more, but about accurately extracting insulator rapidly and to carry out the research of detection and positioning to insulator electric fault then less from transmission line of electricity image, and existing detection method is comprehensive not strong, due to the reason such as complicacy and weather effect of electric transmission line erection environment, the image range gathered is wide, background is complicated, object is more, comprise vegetation, greatly, the interfere informations such as steel tower, object in parts of images can be made overlapping with disturbing factor staggered, thus further increase the detection difficulty of defects of insulator, be difficult to the accuracy of detection ensureing defect, exist undetected or examine indeterminable deficiency.As document " the glass insulator defect diagonsis based on coloured image " proposes the glass insulator defect diagonsis method for removing based on Histogram Matching criterion, a large amount of normal glass insulation sub-pictures can be got rid of fast, improve fault detection efficiency to a certain extent, but the location of the concrete trouble spot of insulator is not studied in the document, therefore can not replace the insulator broken down in time.And for example document " SegmentationofInsulatorImagesBasedonHSIColorSpace " proposes and uses maximum variance between clusters to carry out the method for Iamge Segmentation in HSI space, but it only have studied the extraction of blue insulator, the defects detection of insulator can not be used for.

Summary of the invention

1. invent the technical matters that will solve

The object of the invention is to overcome in prior art utilize unmanned plane to transmission line of electricity carry out robotization patrol and examine time, relatively low to the accuracy of detection of insulator electric fault in electric force pole tower, there is phenomenon that is undetected or false retrieval, thus the affected deficiency of the security that transmission line of electricity is run, provide a kind of detection and positioning method that unmanned plane patrols and examines defects of insulator in transmission line of electricity image.Method in the application of the invention, can extract from the transmission line of electricity image of background complexity quickly and accurately by insulator, and higher to the accuracy of detection of insulator electric fault, ensure that the safety of transmission line of electricity, reliability service.

2. technical scheme

For achieving the above object, technical scheme provided by the invention is:

A kind of unmanned plane of the present invention patrols and examines the detection and positioning method of defects of insulator in transmission line of electricity image, it is characterized in that: the steps include:

Step one, the image obtained taking photo by plane are transformed into HSI colourity saturation degree brightness space by rgb color space, extract H component image and the S component image of HSI colourity saturation degree brightness space, the H component image of extraction and S component image are carried out binary conversion treatment respectively and obtain each self-corresponding bianry image, then by above two width bianry images after medium filtering with the preliminary profile bianry image namely obtaining insulator chain;

Step 2, the ant group algorithm based on particle group optimizing parameter is adopted to extract the profile of insulator monomer in the preliminary profile bianry image of insulator chain;

Step 3, employing least square method carry out ellipse fitting to insulator monomer profiles, and are positioned by the defect of spacing to insulator detected between insulator monomer profiles.

Further, when the image obtained taking photo by plane in step one is transformed into HSI colourity saturation degree brightness space by rgb color space, to arbitrary pixel, its H component and S component calculate respectively by formula (1), formula (2):

$\left\{\begin{array}{c}H=\left\{\begin{array}{cc}\mathrm{\θ}& B\≤G\\ 360-\mathrm{\θ}& B>G\end{array}\right.\\ \mathrm{\θ}={\cos }^{-1}\left\{\frac{\[(R-G)+(R-B)\]/2}{\sqrt{{(R-G)}^2+(R-B)(G-B)}}\right\}\end{array}\right.---\left(1\right),$

$S=1-\frac{3*\min (R,G,B)}{R+G+B}---\left(2\right),$

Wherein, H and S represents chrominance component and the color saturation component of HSI colourity saturation degree brightness space respectively, and R, G, B represent the red component of rgb color space, green component and blue component respectively.

Further, maximum variance between clusters is adopted to carry out binary conversion treatment respectively to the H component image extracted and S component image in step one, concrete steps are: travel through each pixel in H component image and S component image, take out the gray-scale value of each pixel, suppose that the tonal range of pixel in H component image and S component image is 0 ~ m-1, m-1 is the maximum gradation value of pixel in H component image and S component image herein, and wherein gray scale is the probability that the pixel of i occurs is p _i, H component image and the gray average of S component image in 0 ~ m-1 tonal range are μ, suppose that there is gray threshold T is separated into G by the insulator object and background in two images ₀={ 0 ~ T-1} and G ₁=between T ~ m-1} two gray areas, and G ₀the probability occurred is w ₀, G ₁the probability occurred is w ₁, then G ₀and G ₁mean flow rate μ in interval ₀, μ ₁and the inter-class variance δ that these two interval ²(T) be respectively:

$\left\{\begin{array}{c}{\mathrm{\μ}}_0=\underset{i=0}{\overset{T-1}{\Σ}}\frac{{\mathrm{ip}}_i}{w_0}=\frac{\mathrm{\μ}\left(T\right)}{w\left(T\right)},{\mathrm{\μ}}_1=\frac{\mathrm{\μ}-\mathrm{\μ}\left(T\right)}{1-w\left(T\right)}\\ {\mathrm{\δ}}^2\left(T\right)=w_0{({\mathrm{\μ}}_0-\mathrm{\μ})}^2+w_1{({\mathrm{\μ}}_1-\mathrm{\μ})}^2\end{array}\right.---\left(3\right),$

In formula (3) and w ₀+ w ₁=1, w ₀μ ₀+ w ₁μ ₁=μ;

Gray threshold T is progressively increased progressively in 0 ~ m-1 tonal range, makes Gray-scale value T get all numerical value within the scope of 0 ~ m-1, calculate the inter-class variance δ obtained that at every turn circulates ²(T), circulation obtains maximum between-cluster variance max δ after terminating ²(T), T value is now optimum gradation segmentation threshold, the gray-scale value that gray-scale value in H component image and S component image is greater than the pixel of this T value is set to 1, gray-scale value gray-scale value being less than the pixel of this T value is set to 0, thus obtains H component image and S component image bianry image separately.

Further, the ant group algorithm based on particle group optimizing parameter is adopted to the concrete steps that the profile of insulator monomer in the preliminary profile bianry image of insulator chain extracts to be in step 2:

Step 1, suppose that the size of former Aerial Images is M*N, algorithm initialization (M/2) * (N/2) ant is randomly distributed in the different pixels point in the preliminary profile bianry image of insulator chain;

Step 2, above-mentioned (M/2) * (N/2) ant all carry out selecting to movement in the preliminary profile bianry image of insulator chain according to the transition probability formula in formula (4), and the respective maximum probability direction that namely all ants all calculate in formula (4) is moved:

$P_{(m,n).(l,f)}^{\left(t\right)}=\frac{{\left({\mathrm{\τ}}_{(m,n)(l,f)}\right(t\left)\right)}^a.{\left({\mathrm{\η}}_{l,f}\right)}^{\mathrm{\β}}}{\underset{(l,f)\∈\mathrm{\Ω}(m,n)}{\Σ}{\left({\mathrm{\τ}}_{(m,n)(l,f)}\right(t\left)\right)}^a.{\left({\mathrm{\η}}_{l,f}\right)}^{\mathrm{\β}}}---\left(4\right),$

In formula (4), t is iterations, and (m, n) is ant current place pixel, the arbitrary pixel in the 3*3 neighborhood that (l, f) is point (m, n), for the probability that ant in the t time iterative loop is shifted to pixel (l, f) by pixel (m, n), Ω (m, n) is with the set of all pixels in the 3*3 neighborhood of point (m, n), η _l,ffor the heuristic function at point (l, f) place, through type (5) calculates:

η _l,f＝c*▽I(l,f)

$\▿I(l,f)=\sqrt{{\left(\frac{\∂gray}{\∂l}\right)}^2+{\left(\frac{\∂gray}{\∂f}\right)}^2}---\left(5\right),$

In formula (5), C is magnification constant, and its numerical value gets 1; The shade of gray value that ▽ I (l, f) is ant position (l, f) place, the gray-scale value that gray obtains for traveling through each pixel in image;

In formula (4), τ _{(m, n) (l, f)}t () is at the t time iteration time point (m, n) to point (l, f) size of pheromones intensity on path, its initial value is 0.001, every iteration once, every ant all can be moved once, and produce pheromones in new position, thus the pheromones intensity of all pixels is upgraded, after each ant group algorithm iteration being completed, the pheromones intensity of each pixel and location updating are stored in M*N pheromones intensity matrix image, and the formula that above-mentioned pheromones intensity carries out iteration renewal is as follows:

${\mathrm{\τ}}_{(m,n)(l,f)}\left(t\right)=(1-\mathrm{\ξ}(t\left)\right){\mathrm{\τ}}_{(m,n)(l,f)}(t-1)+\underset{k=1}{\overset{(M/2)*(N/2)}{\Σ}}{\mathrm{\Δ\τ}}_{(m,n)(l,f)}^k(t-1)+{\mathrm{\Δ}}^1{\mathrm{\τ}}_{(m,n)(l,f)}(t-1)-{\mathrm{\Δ}}^2{\mathrm{\τ}}_{(m,n)(l,f)}(t-1)---\left(6\right),$

${\mathrm{\Δ}}^1{\mathrm{\τ}}_{(m,n)(l,f)}(t-1)=\underset{k=1}{\overset{\mathrm{\φ}(t-1)}{\Σ}}{\mathrm{\Δ\τ}}_{(m,n)(l,f)}^{\left(k\right)}(t-1)/L_1---\left(7\right),$

${\mathrm{\Δ}}^2{\mathrm{\τ}}_{(m,n)(l,f)}(t-1)=\underset{k=1}{\overset{\mathrm{\φ}(t-1)}{\Σ}}{\mathrm{\Δ\τ}}_{(m,n)(l,f)}^{\left(k\right)}(t-1)/L_2---\left(8\right),$

In formula (6)-(8), τ _{(m, n) (l, f)}(t-1) be the size of pheromones intensity on the t-1 time iterative loop time point (m, n) to the path of point (l, f), for a kth ant stays the pheromones amount on (m, n) to (l, f) path when the t-1 time iterative loop, if the pheromones amount that all ant iteration once produce is a given fixing normal number, Δ ¹τ _{(m, n) (l, f)}and Δ (t-1) ²τ _{(m, n) (l, f)}(t-1) be respectively and stay point (m when the t-1 time iterative loop, n) to (l, f) the pheromones total amount on local optimum path and locally worst path, point (m herein, n) to (l, f) local optimum path refers to pixel (m, n) to (l, f) shortest path, the local worst path of (m, n) to (l, f) refers to pixel (m, n) to the longest path of (l, f); L ₁and L ₂be respectively above local optimum path and local worst path length, φ (t-1) and be respectively the above local optimum path L that to pass by when the t-1 time iterative loop ₁with local worst path L ₂the quantity of upper ant; The volatility that ξ (t) is pheromones, its initial value is 0.5, and along with the carrying out of iterative loop, the attenuation equation of volatility ξ (t) is as follows:

$\mathrm{\ξ}\left(t\right)=\frac{{\mathrm{\τ}}_{max}-{\mathrm{\τ}}_{\min }}{t-1}*\mathrm{\ξ}(t-1)+J$

$J=\underset{t=0}{\overset{cn}{\Σ}}\frac{\mathrm{\ξ}\left(t\right)-\mathrm{\ξ}(t-1)}{t}---\left(9\right),$

In formula (9), ξ (t) and ξ (t-1) be respectively the t time, the t-1 time iteration time pheromones volatility, cn is algorithm current iteration number of times, τ _maxand τ _minwhen being respectively the t-1 time iteration, all ants stay point (m, n) to the maximal value of the pheromones total amount on point (l, f) path and minimum value, and J is volatility modified value;

In formula (4), α and β is respectively the weight factor of pheromones intensity and heuristic function, the initial value of random given α and β respectively, and the initial value of α and β is positive number, along with the circulation of ant group algorithm, use particle cluster algorithm to carry out parameter training optimization to α and β, the concrete mistake optimized is called:

Step a, the individual RANDOM SOLUTION vector theta of initialization (M/2) * (N/2) _i=(α _i, β _i) as random particles, by each θ _i=(α _i, β _i) be considered as the position of point in two-dimensional space, wherein, possess random velocity vector to i-th random particles

Step b, each particle are by calling f _maxsecondary ant group algorithm is trained, and wherein, i-th random particles carries out iteration renewal according to formula (10) to the locus of oneself and speed when calling ant group algorithm:

v _i(f)＝wv _i(f-1)+c ₁s ₁(P _best(f-1)-θ _i(f-1))+c ₂s ₂(G _best(i-1)-θ _i(f-1))

(10)，

θ _i(f)＝θ _i(f-1)+v _i(f)

Wherein, f is the number of times calling ant group algorithm, f _maxnumerical value get 5; v _i(f) and θ _if () is respectively the speed of i-th particle at the end of the f time ant group algorithm and position, v _iand θ (f-1) _i(f-1) speed of i-th particle at the end of the f-1 time ant group algorithm and position is respectively, wherein, 1≤i≤(M/2) * (N/2); W is Inertia Weight, gets the random number being greater than 1; P _best(f-1) be that i-th random particles is calling the optimum particle position found in the f-1 time ant group algorithm, G _best(i-1) be optimum particle position that now whole population finds; Constant c ₁, c ₂determine a particle selection P respectively _bestand G _besttendency, and be random number separate between [0-2]; Constant s ₁, s ₂for random number separate between [0-1];

Step c, detect insulator contour and motion renewal reaches default ant group algorithm cycle index f when i-th particle calls ant group algorithm _maxor again circulate upgrade result consistent with the result that circulated last time time then stop circulating, more new variables G _best;

Steps d, change next particle and again repeat above-mentioned steps b, step c until all particles all complete iteration, the G finally obtained _bestbe optimal particle group position, according to optimal particle group position G now _bestα and β in formula (4) is upgraded;

Step 3, when the call number of ant group algorithm reach setting maximum cycle time, namely complete the operation of ant group algorithm, use maximum variance between clusters determination best information element intensity segmentation threshold τ for the pheromones intensity matrix image that now whole population finds ₀, after Threshold segmentation, namely obtain the profile of insulator monomer.

Further, the concrete steps positioned by the defect of spacing to insulator detected between insulator monomer profiles in step 3 are: the coordinate points at traversal insulator elliptic contour center, and the insulator centre coordinate on same straight line is stored in same array A [x] [y], calculate the width of all insulator contours on this straight line and the mean value of height respectively, adopt above width mean value D and height average H as single insulator contour width and reference value highly respectively; The x coordinate of insulator contour central point on above-mentioned same straight line is arranged successively from small to large according to bubble sort method, calculates the spacing R between adjacent insulator according to formula (11):

$R=\sqrt{{(x_1-x_0)}^2+{(y_1-y_0)}^2}\≈(\mathrm{\λ}+1)R_0---\left(11\right),$

In above formula, R ₀for the average headway at the profile center of adjacent insulator on above-mentioned same straight line, if λ is >1, then illustrate between adjacent insulator A [x1] [y1] and A [x0] [y0], to there is λ insulator disappearance, now ask for the center of λ pseudo-insulator therebetween according to adjacent insulator coordinate and in former figure, draw out the profile of disordered insulator by its wide reference value D, height reference value H.

3. beneficial effect

Adopt technical scheme provided by the invention, compared with prior art, there is following remarkable result:

(1) a kind of unmanned plane of the present invention patrols and examines the detection and positioning method of defects of insulator in transmission line of electricity image, in conjunction with the imaging characteristics of insulator in transmission line of electricity image, select suitable color space, simultaneously in conjunction with H component and S component to Image Segmentation Using process, thus the color information of image can be made full use of, the impact of intensity of illumination on segmentation effect can be reduced again, substantially increase the segmentation effect of insulator and background in image; Also by selecting the compartition strategy matched in the present invention, further ensure the segmentation effect of insulator and background, for the detection of follow-up insulator monomer and fault thereof provides advantage.Inventor is through long-term theoretical analysis and practice, the ant group algorithm based on particle group optimizing parameter is adopted to extract the profile of insulator monomer in the preliminary profile bianry image of insulator chain, by particle cluster algorithm, the parameter in ant group algorithm is optimized, the two is combined and is further improved, substantially increase the extraction accuracy to insulator monomer profiles, thus ensure that the accuracy of detection to insulator breakdown, effectively can prevent phenomenon that is undetected or false retrieval, ensure that the safety of transmission line of electricity, reliability service.

Accompanying drawing explanation

Fig. 1 is the process flow diagram that a kind of unmanned plane of the present invention patrols and examines the detection and positioning method of defects of insulator in transmission line of electricity image;

Fig. 2 (a) is the three actual Aerial Images of going here and there insulator side by side obtained in the embodiment of the present invention;

Fig. 2 (b) is the actual Aerial Images of single insulator string of the existing defects obtained in the embodiment of the present invention;

Fig. 3 (a) is the three preliminary profile bianry images of going here and there insulator side by side in Fig. 2 (a);

Fig. 3 (b) is the preliminary profile bianry image of single insulator string of existing defects in Fig. 2 (b);

Fig. 4 (a) is the profile diagram adopting the ant group algorithm based on particle group optimizing parameter to extract the insulator monomer obtained from Fig. 3 (a);

Fig. 4 (b) is the profile diagram adopting the ant group algorithm based on particle group optimizing parameter to extract the insulator monomer obtained from Fig. 3 (b);

Fig. 5 (a) is the profile diagram of the insulator monomer in the Fig. 4 (a) after ellipse fitting;

Fig. 5 (b) is the profile diagram of the insulator monomer in the Fig. 4 (b) after ellipse fitting;

Fig. 6 (a) and Fig. 6 (b) are the contrast images before the defects of insulator in the embodiment of the present invention marks and after mark.

Embodiment

For understanding content of the present invention further, existing the present invention is described in detail in conjunction with the accompanying drawings and embodiments.

Embodiment 1

The main manifestations of insulator in Aerial Images is following characteristics: (1) insulator monomer profile is the ellipse of unified length and width; (2) equally string-like arrangement in transmission line of electricity; (3) compared with background its colourity and intensity value higher; (4), in fixed voltage grade transmission line of electricity, the insulator number in insulator chain is certain; (6) insulator is generally light green color, translucent, and in Aerial Images, its color is similar to backgrounds such as surface vegetation, general green lake water.

A kind of unmanned plane of the present invention patrols and examines the detection and positioning method of defects of insulator in transmission line of electricity image, is to develop in conjunction with the above feature of insulator in Aerial Images, and as shown in Figure 1, its concrete steps are its flow process:

Step one, as Fig. 2 (a), Fig. 2 (b) is depicted as the actual Aerial Images of the present embodiment, light green color is generally in conjunction with insulator, translucent, in Aerial Images, its color and surface vegetation, the feature that the backgrounds such as general green lake water are similar, the image obtained taking photo by plane is transformed into HSI colourity saturation degree brightness space by rgb color space, extract H component image and the S component image of HSI colourity saturation degree brightness space, thus the color information of image can be made full use of, and reduce the impact of intensity of illumination on segmentation effect, substantially increase the segmentation effect of insulator and background in image, avoid due to season, the factors such as Changes in weather are to the image of image segmentation.Wherein, to arbitrary pixel, its H component and S component calculate respectively by formula (1), formula (2):

$\left\{\begin{array}{c}H=\left\{\begin{array}{cc}\mathrm{\θ}& B\≤G\\ 360-\mathrm{\θ}& B>G\end{array}\right.\\ \mathrm{\θ}={\cos }^{-1}\left\{\frac{\[(R-G)+(R-B)\]/2}{\sqrt{{(R-G)}^2+(R-B)(G-B)}}\right\}\end{array}\right.---\left(1\right),$

$S=1-\frac{3*\min (R,G,B)}{R+G+B}---\left(2\right),$

In above formula, H and S represents chrominance component and the color saturation component of HSI colourity saturation degree brightness space respectively, and R, G, B represent the red component of rgb color space, green component and blue component respectively.

The H component image of extraction and S component image are carried out binary conversion treatment respectively and obtains each self-corresponding bianry image, then by above two width bianry images after medium filtering with the preliminary profile bianry image namely obtaining insulator chain, the preliminary profile bianry image of the insulator chain corresponding to Fig. 2 (a), Fig. 2 (b) is respectively as shown in Fig. 3 (a), Fig. 3 (b).Adopt maximum variance between clusters to carry out binary conversion treatment respectively to the H component image extracted and S component image in the present embodiment, concrete steps are: travel through each pixel in H component image and S component image, take out the gray-scale value of each pixel, suppose that the tonal range of pixel in H component image and S component image is 0 ~ m-1, m-1 is the maximum gradation value of pixel in H component image and S component image herein, and wherein gray scale is the probability that the pixel of i occurs is p _i, H component image and the gray average of S component image in 0 ~ m-1 tonal range are μ, suppose that there is gray threshold T is separated into G by the insulator object and background in two images ₀={ 0 ~ T-1} and G ₁=between T ~ m-1} two gray areas, and G ₀the probability occurred is w ₀, G ₁the probability occurred is w ₁, then G ₀and G ₁mean flow rate μ in interval ₀, μ ₁and the inter-class variance δ that these two interval ²(T) be respectively:

$\left\{\begin{array}{c}{\mathrm{\μ}}_0=\underset{i=0}{\overset{T-1}{\Σ}}\frac{{\mathrm{ip}}_i}{w_0}=\frac{\mathrm{\μ}\left(T\right)}{w\left(T\right)},{\mathrm{\μ}}_1=\frac{\mathrm{\μ}-\mathrm{\μ}\left(T\right)}{1-w\left(T\right)}\\ {\mathrm{\δ}}^2\left(T\right)=w_0{({\mathrm{\μ}}_0-\mathrm{\μ})}^2+w_1{({\mathrm{\μ}}_1-\mathrm{\μ})}^2\end{array}\right.---\left(3\right),$

In formula (3) and w ₀+ w ₁=1, w ₀μ ₀+ w ₁μ ₁=μ;

Gray threshold T is progressively increased progressively in 0 ~ m-1 tonal range, makes Gray-scale value T get all numerical value within the scope of 0 ~ m-1, calculate the inter-class variance δ obtained that at every turn circulates ²(T), circulation obtains maximum between-cluster variance max δ after terminating ²(T), T value is now optimum gradation segmentation threshold, the gray-scale value that gray-scale value in H component image and S component image is greater than the pixel of this T value is set to 1, gray-scale value gray-scale value being less than the pixel of this T value is set to 0, thus obtains H component image and S component image bianry image separately.

Step 2, adopt the ant group algorithm based on particle group optimizing parameter to extract the profile of insulator monomer in the preliminary profile bianry image of insulator chain, its concrete steps are:

Step 1, suppose that the size of former Aerial Images is M*N, algorithm initialization (M/2) * (N/2) ant is randomly distributed in the different pixels point in the preliminary profile bianry image of insulator chain.

Step 2, above-mentioned (M/2) * (N/2) ant all carry out selecting to movement in the preliminary profile bianry image of insulator chain according to the transition probability formula in formula (4), and the respective maximum probability direction that namely all ants all calculate in formula (4) is moved:

$P_{(m,n).(l,f)}^{\left(t\right)}=\frac{{\left({\mathrm{\τ}}_{(m,n)(l,f)}\right(t\left)\right)}^a.{\left({\mathrm{\η}}_{l,f}\right)}^{\mathrm{\β}}}{\underset{(l,f)\∈\mathrm{\Ω}(m,n)}{\Σ}{\left({\mathrm{\τ}}_{(m,n)(l,f)}\right(t\left)\right)}^a.{\left({\mathrm{\η}}_{l,f}\right)}^{\mathrm{\β}}}---\left(4\right).$

In formula (4), t is iterations, and (m, n) is ant current place pixel, the arbitrary pixel in the 3*3 neighborhood that (l, f) is point (m, n), for the probability that ant in the t time iterative loop is shifted to pixel (l, f) by pixel (m, n), Ω (m, n) is with the set of all pixels in the 3*3 neighborhood of point (m, n).η _l,ffor the heuristic function at point (l, f) place, through type (5) calculates:

η _l,f＝c*▽I(l,f)

$\▿I(l,f)=\sqrt{{\left(\frac{\∂gray}{\∂l}\right)}^2+{\left(\frac{\∂gray}{\∂f}\right)}^2}---\left(5\right),$

In formula (5), C is magnification constant, and its numerical value gets 1; ▽ I (l, f) be ant position (l, f) the shade of gray value at place, the gray-scale value that gray obtains for traveling through each pixel in image, with ant position (l in the present embodiment, f) the shade of gray value ▽ I (l at place, f) as heuristic function, because insulator contour marginal portion grey scale change is comparatively violent, better the pixel of insulator elevator profile and other pixels can be made a distinction like this, improve the accuracy of detection of insulator monomer profiles and the normal startup of guarantee algorithm.

In formula (4), τ _{(m, n) (l, f)}t () is at the t time iteration time point (m, n) to point (l, f) size of pheromones intensity on path, its initial value is 0.001, every iteration once, every ant all can be moved once, and produce pheromones in new position, thus the pheromones intensity of all pixels is upgraded, after each ant group algorithm iteration being completed in the present embodiment, the pheromones intensity of each pixel and location updating are stored in M*N pheromones intensity matrix image, and the formula that above-mentioned pheromones intensity carries out iteration renewal is as follows:

${\mathrm{\τ}}_{(m,n)(l,f)}\left(t\right)=(1-\mathrm{\ξ}){\mathrm{\τ}}_{(m,n)(l,f)}(t-1)+\underset{k=1}{\overset{(M/2)*(N/2)}{\Σ}}{\mathrm{\Δ\τ}}_{(m,n)(l,f)}^k(t-1)+{\mathrm{\Δ}}^1{\mathrm{\τ}}_{(m,n)(l,f)}(t-1)-{\mathrm{\Δ}}^2{\mathrm{\τ}}_{(m,n)(l,f)}(t-1)---\left(6\right),$

${\mathrm{\Δ}}^1{\mathrm{\τ}}_{(m,n)(l,f)}(t-1)=\underset{k=1}{\overset{\mathrm{\φ}(t-1)}{\Σ}}{\mathrm{\Δ\τ}}_{(m,n)(l,f)}^{\left(k\right)}(t-1)/L_1---\left(7\right),$

${\mathrm{\Δ}}^2{\mathrm{\τ}}_{(m,n)(l,f)}(t-1)=\underset{k=1}{\overset{\mathrm{\φ}(t-1)}{\Σ}}{\mathrm{\Δ\τ}}_{(m,n)(l,f)}^{\left(k\right)}(t-1)/L_2---\left(8\right),$

In formula (6)-(8), τ _{(m, n) (l, f)}(t-1) be the size of pheromones intensity on the t-1 time iterative loop time point (m, n) to the path of point (l, f), for a kth ant stays the pheromones amount on (m, n) to (l, f) path when the t-1 time iterative loop, if the pheromones amount that all ant iteration once produce is a given fixing normal number, Δ ¹τ _{(m, n) (l, f)}and Δ (t-1) ²τ _{(m, n) (l, f)}(t-1) be respectively and stay point (m when the t-1 time iterative loop, n) to (l, f) the pheromones total amount on local optimum path and locally worst path, point (m herein, n) to (l, f) local optimum path refers to pixel (m, n) to (l, f) shortest path, the local worst path of (m, n) to (l, f) refers to pixel (m, n) to the longest path of (l, f); L ₁and L ₂be respectively above local optimum path and local worst path length, φ (t-1) and be respectively the above local optimum path L that to pass by when the t-1 time iterative loop ₁with local worst path L ₂the quantity of upper ant; Pheromones by adopting elitism strategy to process ant release on worst path in the present embodiment, amplify the pheromone concentration of the ant release on optimal path simultaneously, can prevent from like this searching for and be absorbed in the appearance that local optimum puppet separates this situation, make flase drop be that pheromones amount on profile drops to minimum, thus increase the accuracy rate of contour detecting.ξ is the volatility of pheromones, and its initial value is 0.5, and along with the carrying out of iterative loop, the attenuation equation of volatility ξ is as follows:

$\mathrm{\ξ}\left(t\right)=\frac{{\mathrm{\τ}}_{max}-{\mathrm{\τ}}_{\min }}{t-1}*\mathrm{\ξ}(t-1)+J$

$J=\underset{t=0}{\overset{cn}{\Σ}}\frac{\mathrm{\ξ}\left(t\right)-\mathrm{\ξ}(t-1)}{t}---\left(9\right),$

In formula (9), ξ (t) and ξ (t-1) be respectively the t time, the t-1 time iteration time pheromones volatility, cn is algorithm current iteration number of times, τ _maxand τ _minwhen being respectively the t-1 time iteration, all ants stay point (m, n) to the maximal value of the pheromones total amount on point (l, f) path and minimum value, and J is volatility modified value;

In formula (4), α and β is respectively the weight factor of pheromones intensity and heuristic function, the initial value of random given α and β respectively, and the initial value of α and β is positive number, along with the circulation of ant group algorithm, use particle cluster algorithm to carry out parameter training optimization to α and β, the concrete mistake optimized is called:

Step a, the individual RANDOM SOLUTION vector theta of initialization (M/2) * (N/2) _i=(α _i, β _i) as random particles, by each θ _i=(α _i, β _i) be considered as the position of point in two-dimensional space, wherein, possess random velocity vector to i-th random particles

Step b, each particle are by calling f _maxsecondary ant group algorithm is trained, and wherein, i-th random particles carries out iteration renewal according to formula (10) to the locus of oneself and speed when calling ant group algorithm:

v _i(f)＝wv _i(f-1)+c ₁s ₁(P _best(f-1)-θ _i(f-1))+c ₂s ₂(G _best(i-1)-θ _i(f-1))

(10)，

θ _i(f)＝θ _i(f-1)+v _i(f)

Wherein, f is the number of times calling ant group algorithm, by a large amount of experimental studies to f _maxnumerical value be optimized, through optimizing the f that determines _maxnumerical value get 4-8, get 5 in the present embodiment, thus can prevent local optimum puppet separate generation, improve the accuracy of detection of insulator monomer profiles.V _i(f) and θ _if () is respectively the speed of i-th particle at the end of the f time ant group algorithm and position, v _iand θ (f-1) _i(f-1) speed of i-th particle at the end of the f-1 time ant group algorithm and position is respectively, wherein, 1≤i≤(M/2) * (N/2); W is Inertia Weight, gets the random number being greater than 1; P _best(f-1) be the optimum particle position that i-th random particles finds when calling the f-1 time ant group algorithm, G _best(i-1) be the i-th-1 random particles optimum particle position that whole population finds at the end of calling ant group algorithm; Constant c ₁, c ₂determine a particle selection P respectively _bestand G _besttendency, and be random number separate between [0-2]; Constant s ₁, s ₂for random number separate between [0-1].

In the present embodiment, i-th random particles often calls an ant group algorithm just to the optimum particle position P that it finds _bestupgrade, if the optimum particle position that i-th random particles finds at the end of the f-1 time ant group algorithm is P _best(f-1), the position θ of this particle at the end of the f time ant group algorithm _i(f)=(α _i(f), β _i(f)), when this particle being called the f time ant group algorithm ant walk the length L in path _fwith searching P _best(f-1) in the ant group algorithm called ant walk the length L in path _best(f-1) compare, if L _f≤ L _best(f-1), then with θ _i(f)=(α _i(f), β _i(f)) the optimum particle position P that finds when the f time ant group algorithm as this particle _best(f); If L _f> L _best(f-1), then still with P _best(f-1) the optimum particle position P found when the f time ant group algorithm as this particle _best(f).With the initial spatial location θ of i-th random particles in the present embodiment _i=(α _i, β _i) be its optimum particle position P searched out _bestinitial value.

Step c, detect insulator contour and motion renewal reaches default ant group algorithm cycle index f when i-th particle calls ant group algorithm _maxor the result that upgrades of again circulating is consistent with the result that circulated last time, i.e. the P that finds of this circulation time _bestwith the P found that circulated last time _bestthen stop circulation time identical, and i-th particle is found its optimum particle position P _besttime call ant in ant group algorithm walk path length with find G _best(i-1) call ant in ant group algorithm walk path length compare, carry out more new variables G so that path that ant walks is shorter for criterion equally _best.

Steps d, change next particle and again repeat above-mentioned steps b, step c until all particles all complete iteration, the G finally obtained _bestbe optimal particle group position, according to optimal particle group position G now _bestα and β in formula (4) is upgraded.

Step 3, when the call number of ant group algorithm reach setting maximum cycle time, namely complete the operation of ant group algorithm, use maximum variance between clusters determination best information element intensity segmentation threshold τ for the pheromones intensity matrix image that now whole population finds ₀, adopt τ ₀to the pheromones intensity matrix Image Segmentation Using that now whole population finds, pheromones intensity is greater than τ ₀pixel be pixel on insulator monomer profiles, therefore can obtain the profile of insulator monomer after Threshold segmentation, as Fig. 4 (a) and Fig. 4 (b).

What deserves to be explained is, although have relevant report about research particle swarm optimization algorithm and ant group algorithm being carried out combining in prior art, but there is not any research about the detection being applied to defects of insulator in unmanned plane transmission line of electricity image, and due to the complicacy of insulator imaging characteristics and by surface vegetation, the interference of the Similar color backgrounds such as lake water is comparatively serious, the existing improvement ant group algorithm based on population parameter optimization can not be directly used in the detection of defects of insulator, its accuracy of detection is lower, easy generation false retrieval and undetected phenomenon, and working time is longer, and this also becomes the problem of puzzlement inventor's long period.Inventor is by the analysis to insulator imaging characteristics, and the difference of background imaging and insulator imaging in combining image, through long-term large quantifier elimination and experimental simulation, by being further improved the existing improvement ant group algorithm based on population parameter optimization, just obtain technical scheme of the present invention, the extraction of insulator monomer profiles in the preliminary profile bianry image of the insulator chain obtained after making the ant group algorithm after improvement be applicable to segmentation, and the imaging characteristics of insulator can be adapted to, thus the accuracy of detection substantially increased defects of insulator, reduce the interference of background, improve travelling speed.

Step 3, the center point coordinate traveling through each insulator contour and major and minor axis length, due to insulator monomer profiles in the picture profile be the ellipse of unified length and width and equally string-like arrangement in transmission line of electricity, least square method is adopted to carry out ellipse fitting to insulator monomer profiles, and variable is set in order to connected domain number in computed image, the insulator monomer profiles after matching is as Fig. 5 (a) and Fig. 5 (b).Positioned by the defect of spacing to insulator detected between insulator monomer profiles, to the concrete steps that the defect of insulator positions be: the coordinate points at traversal insulator elliptic contour center, and the insulator centre coordinate on same straight line is stored in same array A [x] [y], calculate the width of all insulator contours on this straight line and the mean value of height respectively, adopt above width mean value D and height average H as single insulator contour width and reference value highly respectively.The x coordinate of insulator contour central point on above-mentioned same straight line is arranged successively from small to large according to bubble sort method, calculates the spacing R between adjacent insulator according to formula (11):

$R=\sqrt{{(x_1-x_0)}^2+{(y_1-y_0)}^2}\≈(\mathrm{\λ}+1)R_0---\left(11\right),$

In above formula, R ₀for the average headway at the profile center of adjacent insulator on above-mentioned same straight line, if λ is >1, then illustrate between adjacent insulator A [x1] [y1] and A [x0] [y0], to there is λ insulator disappearance, now ask for the center of λ pseudo-insulator therebetween according to adjacent insulator coordinate and in former figure, draw out the profile of disordered insulator by its wide reference value D, height reference value H.

As shown in Fig. 6 (a) He Fig. 5 (b), be respectively the comparison diagram before defects of insulator marks in the present embodiment and after mark, the number of the number and insulator disappearance that count insulator in present image in figure is respectively 35 and 1.The platform of the present embodiment is based on Qt software development, and Riming time of algorithm is about 20ms, and portable such as embedded platform uses flexibly, also can be applied to video real time processing system after secondary development.

Claims (5)

1. unmanned plane patrols and examines a detection and positioning method for defects of insulator in transmission line of electricity image, it is characterized in that: the steps include:

Step one, the image obtained taking photo by plane are transformed into HSI colourity saturation degree brightness space by rgb color space, extract H component image and the S component image of HSI colourity saturation degree brightness space, the H component image of extraction and S component image are carried out binary conversion treatment respectively and obtain each self-corresponding bianry image, then by above two width bianry images after medium filtering with the preliminary profile bianry image namely obtaining insulator chain;

Step 2, the ant group algorithm based on particle group optimizing parameter is adopted to extract the profile of insulator monomer in the preliminary profile bianry image of insulator chain;

Step 3, employing least square method carry out ellipse fitting to insulator monomer profiles, and are positioned by the defect of spacing to insulator detected between insulator monomer profiles.

2. a kind of unmanned plane according to claim 1 patrols and examines the detection and positioning method of defects of insulator in transmission line of electricity image, it is characterized in that: when the image obtained taking photo by plane in step one is transformed into HSI colourity saturation degree brightness space by rgb color space, to arbitrary pixel, its H component and S component calculate respectively by formula (1), formula (2):

$\left\{\begin{array}{c}H=\left\{\begin{array}{cc}\mathrm{\θ}& B\≤G\\ 360-\mathrm{\θ}& B>G\end{array}\right.\\ \mathrm{\θ}={\cos }^{-1}\left\{\frac{\[(R-G)+(R-B)\]/2}{\sqrt{{(R-G)}^2+(R-B)(G-B)}}\right\}\end{array}\right.---\left(1\right),$

$S=1-\frac{3*\min (R,G,B)}{R+G+B}---\left(2\right),$

Wherein, H and S represents chrominance component and the color saturation component of HSI colourity saturation degree brightness space respectively, and R, G, B represent the red component of rgb color space, green component and blue component respectively.

3. a kind of unmanned plane according to claim 2 patrols and examines the detection and positioning method of defects of insulator in transmission line of electricity image, it is characterized in that: in step one, adopt maximum variance between clusters to carry out binary conversion treatment respectively to the H component image extracted and S component image, concrete steps are: travel through each pixel in H component image and S component image, take out the gray-scale value of each pixel, suppose that the tonal range of pixel in H component image and S component image is 0 ~ m-1, m-1 is the maximum gradation value of pixel in H component image and S component image herein, wherein gray scale is the probability of the pixel appearance of i is p _i, H component image and the gray average of S component image in 0 ~ m-1 tonal range are μ, suppose that there is gray threshold T is separated into G by the insulator object and background in two images ₀={ 0 ~ T-1} and G ₁=between T ~ m-1} two gray areas, and G ₀the probability occurred is w ₀, G ₁the probability occurred is w ₁, then G ₀and G ₁mean flow rate μ in interval ₀, μ ₁and the inter-class variance δ that these two interval ²(T) be respectively:

$\left\{\begin{array}{c}{\mathrm{\μ}}_0=\underset{i=0}{\overset{T-1}{\Σ}}\frac{{\mathrm{ip}}_i}{w_0}=\frac{\mathrm{\μ}\left(T\right)}{w\left(T\right)},{\mathrm{\μ}}_1=\frac{\mathrm{\μ}-\mathrm{\μ}\left(T\right)}{1-w\left(T\right)}\\ {\mathrm{\δ}}^2\left(T\right)=w_0{({\mathrm{\μ}}_0-\mathrm{\μ})}^2+w_1{({\mathrm{\μ}}_1-\mathrm{\μ})}^2\end{array}\right.---\left(3\right),$

In formula (3) and w ₀+ w ₁=1, w ₀μ ₀+ w ₁μ ₁=μ;

Gray threshold T is progressively increased progressively in 0 ~ m-1 tonal range, makes Gray-scale value T get all numerical value within the scope of 0 ~ m-1, calculate the inter-class variance δ obtained that at every turn circulates ²(T), circulation obtains maximum between-cluster variance max δ after terminating ²(T), T value is now optimum gradation segmentation threshold, the gray-scale value that gray-scale value in H component image and S component image is greater than the pixel of this T value is set to 1, gray-scale value gray-scale value being less than the pixel of this T value is set to 0, thus obtains H component image and S component image bianry image separately.

4. a kind of unmanned plane according to any one of claim 1-3 patrols and examines the detection and positioning method of defects of insulator in transmission line of electricity image, it is characterized in that: adopt the ant group algorithm based on particle group optimizing parameter to the concrete steps that the profile of insulator monomer in the preliminary profile bianry image of insulator chain extracts to be in step 2:

Step 1, suppose that the size of former Aerial Images is M*N, algorithm initialization (M/2) * (N/2) ant is randomly distributed in the different pixels point in the preliminary profile bianry image of insulator chain;

Step 2, above-mentioned (M/2) * (N/2) ant all carry out selecting to movement in the preliminary profile bianry image of insulator chain according to the transition probability formula in formula (4), and the respective maximum probability direction that namely all ants all calculate in formula (4) is moved:

$P_{(m,n).(l,f)}^{\left(t\right)}=\frac{{\left({\mathrm{\τ}}_{(m,n)(l,f)}\right(t\left)\right)}^a.{\left({\mathrm{\η}}_{l,f}\right)}^{\mathrm{\β}}}{\underset{(l,f)\∈\mathrm{\Ω}(m,n)}{\Σ}{\left({\mathrm{\τ}}_{(m,n)(l,f)}\right(t\left)\right)}^a.{\left({\mathrm{\η}}_{l,f}\right)}^{\mathrm{\β}}}---\left(4\right),$

In formula (4), t is iterations, and (m, n) is ant current place pixel, the arbitrary pixel in the 3*3 neighborhood that (l, f) is point (m, n), for the probability that ant in the t time iterative loop is shifted to pixel (l, f) by pixel (m, n), Ω (m, n) is with the set of all pixels in the 3*3 neighborhood of point (m, n), η _l,ffor the heuristic function at point (l, f) place, through type (5) calculates:

η _l,f＝c*▽I(l,f)

$\▿I(l,f)=\sqrt{{\left(\frac{\∂gray}{\∂l}\right)}^2+{\left(\frac{\∂gray}{\∂f}\right)}^2}---\left(5\right),$

In formula (5), C is magnification constant, and its numerical value gets 1; The shade of gray value that ▽ I (l, f) is ant position (l, f) place, the gray-scale value that gray obtains for traveling through each pixel in image;

In formula (4), τ _{(m, n) (l, f)}t () is at the t time iteration time point (m, n) to point (l, f) size of pheromones intensity on path, its initial value is 0.001, every iteration once, every ant all can be moved once, and produce pheromones in new position, thus the pheromones intensity of all pixels is upgraded, after each ant group algorithm iteration being completed, the pheromones intensity of each pixel and location updating are stored in M*N pheromones intensity matrix image, and the formula that above-mentioned pheromones intensity carries out iteration renewal is as follows:

${\mathrm{\τ}}_{(m,n)(l,f)}\left(t\right)=(1-\mathrm{\ξ}){\mathrm{\τ}}_{(m,n)(l,f)}(t-1)+\underset{k=1}{\overset{(M/2)*(N/2)}{\Σ}}{\mathrm{\Δ\τ}}_{(m,n)(l,f)}^k(t-1)+{\mathrm{\Δ}}^1{\mathrm{\τ}}_{(m,n)(l,f)}(t-1)-{\mathrm{\Δ}}^2{\mathrm{\τ}}_{(m,n)(l,f)}(t-1)---\left(6\right),$

${\mathrm{\Δ}}^1{\mathrm{\τ}}_{(m,n)(l,f)}(t-1)=\underset{k=1}{\overset{\mathrm{\φ}(t-1)}{\Σ}}{\mathrm{\Δ\τ}}_{(m,n)(l,f)}^{\left(k\right)}(t-1)/L_1---\left(7\right),$

${\mathrm{\Δ}}^2{\mathrm{\τ}}_{(m,n)(l,f)}(t-1)=\underset{k=1}{\overset{\mathrm{\φ}(t-1)}{\Σ}}{\mathrm{\Δ\τ}}_{(m,n)(l,f)}^{\left(k\right)}(t-1)/L_2---\left(8\right),$

In formula (6)-(8), τ _{(m, n) (l, f)}(t-1) be the size of pheromones intensity on the t-1 time iterative loop time point (m, n) to the path of point (l, f), for a kth ant stays the pheromones amount on (m, n) to (l, f) path when the t-1 time iterative loop, if the pheromones amount that all ant iteration once produce is a given fixing normal number, Δ ¹τ _{(m, n) (l, f)}and Δ (t-1) ²τ _{(m, n) (l, f)}(t-1) be respectively and stay point (m when the t-1 time iterative loop, n) to (l, f) the pheromones total amount on local optimum path and locally worst path, point (m herein, n) to (l, f) local optimum path refers to pixel (m, n) to (l, f) shortest path, the local worst path of (m, n) to (l, f) refers to pixel (m, n) to the longest path of (l, f); L ₁and L ₂be respectively above local optimum path and local worst path length, φ (t-1) and be respectively the above local optimum path L that to pass by when the t-1 time iterative loop ₁with local worst path L ₂the quantity of upper ant; ξ is the volatility of pheromones, and its initial value is 0.5, and along with the carrying out of iterative loop, the attenuation equation of volatility ξ is as follows:

$\mathrm{\ξ}\left(t\right)=\frac{{\mathrm{\τ}}_{max}-{\mathrm{\τ}}_{\min }}{t-1}*\mathrm{\ξ}(t-1)+J$

$J=\underset{t=0}{\overset{cn}{\Σ}}\frac{\mathrm{\ξ}\left(t\right)-\mathrm{\ξ}(t-1)}{t}---\left(9\right),$

In formula (9), ξ (t) and ξ (t-1) be respectively the t time, the t-1 time iteration time pheromones volatility, cn is algorithm current iteration number of times, τ _maxand τ _minwhen being respectively the t-1 time iteration, all ants stay point (m, n) to the maximal value of the pheromones total amount on point (l, f) path and minimum value, and J is volatility modified value;

In formula (4), α and β is respectively the weight factor of pheromones intensity and heuristic function, the initial value of random given α and β respectively, and the initial value of α and β is positive number, along with the circulation of ant group algorithm, use particle cluster algorithm to carry out parameter training optimization to α and β, the concrete mistake optimized is called:

Step a, the individual RANDOM SOLUTION vector theta of initialization (M/2) * (N/2) _i=(α _i, β _i) as random particles, by each θ _i=(α _i, β _i) be considered as the position of point in two-dimensional space, wherein, possess random velocity vector to i-th random particles

Step b, each particle are by calling f _maxsecondary ant group algorithm is trained, and wherein, i-th random particles carries out iteration renewal according to formula (10) to the locus of oneself and speed when calling ant group algorithm:

v _i(f)＝wv _i(f-1)+c ₁s ₁(P _best(f-1)-θ _i(f-1))+c ₂s ₂(G _best(i-1)-θ _i(f-1))(10)，

θ _i(f)＝θ _i(f-1)+v _i(f)

Wherein, f is the number of times calling ant group algorithm, f _maxnumerical value get 5; v _i(f) and θ _if () is respectively the speed of i-th particle at the end of the f time ant group algorithm and position, v _iand θ (f-1) _i(f-1) speed of i-th particle at the end of the f-1 time ant group algorithm and position is respectively, wherein, 1≤i≤(M/2) * (N/2); W is Inertia Weight, gets the random number being greater than 1; P _best(f-1) be that i-th random particles is calling the optimum particle position found in the f-1 time ant group algorithm, G _best(i-1) be optimum particle position that now whole population finds; Constant c ₁, c ₂determine a particle selection P respectively _bestand G _besttendency, and be random number separate between [0-2]; Constant s ₁, s ₂for random number separate between [0-1];

Step c, detect insulator contour and motion renewal reaches default ant group algorithm cycle index f when i-th particle calls ant group algorithm _maxor again circulate upgrade result consistent with the result that circulated last time time then stop circulating, more new variables G _best;

Steps d, change next particle and again repeat above-mentioned steps b, step c until all particles all complete iteration, the G finally obtained _bestbe optimal particle group position, according to optimal particle group position G now _bestα and β in formula (4) is upgraded;

Step 3, when the call number of ant group algorithm reach setting maximum cycle time, namely complete the operation of ant group algorithm, use maximum variance between clusters determination best information element intensity segmentation threshold τ for the pheromones intensity matrix image that now whole population finds ₀, after Threshold segmentation, namely obtain the profile of insulator monomer.

5. a kind of unmanned plane according to claim 4 patrols and examines the detection and positioning method of defects of insulator in transmission line of electricity image, it is characterized in that: the concrete steps positioned by the defect of spacing to insulator detected between insulator monomer profiles in step 3 are: the coordinate points at traversal insulator elliptic contour center, and the insulator centre coordinate on same straight line is stored in same array A [x] [y], calculate the width of all insulator contours on this straight line and the mean value of height respectively, adopt above width mean value D and height average H as single insulator contour width and reference value highly respectively, the x coordinate of insulator contour central point on above-mentioned same straight line is arranged successively from small to large according to bubble sort method, calculates the spacing R between adjacent insulator according to formula (11):

$R=\sqrt{{(x_1-x_0)}^2+{(y_1-y_0)}^2}\≈(\mathrm{\λ}+1)R_0---\left(11\right),$

In above formula, R ₀for the average headway at the profile center of adjacent insulator on above-mentioned same straight line, if λ is >1, then illustrate between adjacent insulator A [x1] [y1] and A [x0] [y0], to there is λ insulator disappearance, now ask for the center of λ pseudo-insulator therebetween according to adjacent insulator coordinate and in former figure, draw out the profile of disordered insulator by its wide reference value D, height reference value H.