CN103530620A - Method for identifying bird nest on electric transmission line tower - Google Patents

Abstract

The invention discloses a method for identifying a bird nest on an electric transmission line tower in the technical field of digital image processing and online monitoring of the electric transmission line tower. The method comprises the steps as follows: acquiring an image of the electric transmission line tower, determining whether the acquired image of the electric transmission line tower is in one picture or more pictures, using a connected domain method to judge whether the bird nest exists on the electric transmission line tower if the acquired image of the electric transmission line tower is in one picture, and using a machine learning method to judge whether the bird nest exists on the electric transmission line tower if the acquired image of the electric transmission line tower is in more pictures. According to the method provided by the invention, the situation that whether the bird nest exists on the electric transmission line tower in the single picture is judged by adopting the connected domain method, so that the identifying and scanning scope is simplified and the calculated amount is reduced; in addition, the situation that whether the bird nest exists on the electric transmission line tower in the more pictures is judged by adopting the machine learning method, so that the calculated amount in the calculation process is reduced while the identifying effectiveness is taken into consideration.

Description

A kind of recognition methods of electric power line pole tower Bird's Nest

Technical field

The invention belongs to Digital Image Processing and electric power line pole tower on-line monitoring technique field, relate in particular to a kind of recognition methods of electric power line pole tower Bird's Nest.

Background technology

Electric power line pole tower plays very important effect in electric system, is directly connected to huge numbers of families' electrical problem, and major power outage will bring immeasurable loss to national economy.The nest building behavior of birds can affect the safe operation of electric power line pole tower, very easily causes line short or tripping operation.

At present electric power line pole tower Bird's Nest is monitored to adopted means, be divided into and draw bird and drive two kinds of solution thinkings of bird.Wherein draw bird and mainly comprise that people is that vertical rod is established nest, at overhead line structures safe location, installed man-made nest additional and install and draw three kinds of schemes such as Bird's Nest on the related facility of circuit corridor.Drive bird and be by artificial generation noise and various flush things be installed on shaft tower and realized scaring expulsion birds and make the object that it dare not be close.But in actual conditions, conventional drawing bird and drive bird measure all needs deeply operation on the spot of staff, with geographical constraints larger with time restriction, and labor intensive material resources are more, and monitoring coverage is also very limited.

Adopt image processing techniques to identify the birds nest building behavior on electric power line pole tower, be a kind of more simply, solution accurately and efficiently.Image processing techniques is applied to, in the Bird's Nest identification of electric power line pole tower, all belong to comparatively novel research field at home and abroad.The present invention is based on Digital image technology, for the digital picture intercepting near the video camera electric power line pole tower video flowing that take and that be sent to Surveillance center being arranged on shaft tower and be research object by the electric power line pole tower image that helicopter is patrolled and examined acquisition, the method that adopts video image to process is identified and classifies electric power line pole tower Bird's Nest, the method is identified the Bird's Nest on electric power line pole tower by video image processing technology, can be intelligent, position fast, detect, manpower and materials have greatly been saved, make whole monitoring system become efficient and practical, improve the utilization ratio of on-line monitoring system, for guaranteeing that safe operation of power system provides a kind of new intuitively method accurately, there is very important realistic meaning.

Summary of the invention

The object of the invention is to, a kind of recognition methods of electric power line pole tower Bird's Nest is provided, the problem existing when by video image identification electric power line pole tower Bird's Nest for solving prior art.

To achieve these goals, the technical scheme that the present invention proposes is that a kind of recognition methods of electric power line pole tower Bird's Nest, is characterized in that described method comprises:

Step 1: obtain electric power line pole tower vision signal by helicopter routing inspection or camera, gather corresponding electric power line pole tower original object image;

Step 2: determine that the electric power line pole tower image gathering is single width picture or several pictures;

Step 3: whether have Bird's Nest according to the electric power line pole tower spectral discrimination electric power line pole tower gathering.

Whether the described electric power line pole tower spectral discrimination electric power line pole tower according to collection exists Bird's Nest to comprise following sub-step:

Sub-step 11: single width picture is carried out to gray scale processing and obtain gray processing picture; Gray processing picture is carried out to two-value processing and obtain binaryzation picture;

Sub-step 12: binaryzation picture is divided into M sub-pictures, in each sub-pictures, add up respectively binary value and be 0 the shared ratio of pixel, on duty is that 0 pixel is communicated with and shared ratio is greater than the threshold value of setting, judges that electric power line pole tower exists Bird's Nest.

Whether the described electric power line pole tower spectral discrimination electric power line pole tower according to collection exists Bird's Nest to comprise following sub-step:

Sub-step 21: choose p and determined picture and p the definite picture that does not have Bird's Nest that has Bird's Nest, p is setting value;

Sub-step 22: the picture of choosing is carried out to gray processing processing;

Sub-step 23: the picture that utilizes bilinear interpolation method to process gray processing is normalized, generates the gray scale picture of setting pixel;

Sub-step 24: the image data matrix that the gray scale picture of all setting pixels is carried out obtaining after Gabor filtering combines;

Sub-step 25: adopt principal component analysis (PCA) to extract the characteristic of the filtered image data matrix of Gabor, generating training data collection and test data set;

Sub-step 26: utilize training dataset training ELM neural network, and utilize test data set checking ELM neural network model;

Sub-step 27: utilize several picture generating feature data sets;

Sub-step 28: described characteristic data set is input in trained ELM neural network and calculates output valve;

Sub-step 29: judge according to described output valve whether electric power line pole tower exists Bird's Nest.

The present invention adopts connected domain method to judge whether the electric power line pole tower in single width picture exists Bird's Nest, has simplified the scope of identification scanning, has reduced calculated amount; In addition, the present invention adopts machine learning method to judge whether the electric power line pole tower in several pictures exists Bird's Nest, when taking into account identification validity, has reduced the calculated amount in calculating process.

Accompanying drawing explanation

Fig. 1 is the recognition methods process flow diagram of electric power line pole tower Bird's Nest provided by the invention;

Fig. 2 is that the electric power line pole tower image that gathers judges whether electric power line pole tower exists the process flow diagram of Bird's Nest while being single width picture;

Fig. 3 is the picture after the single width picture binaryzation that provides of embodiment 1;

Fig. 4 is the sub-pictures that the picture after the single width picture binaryzation that provides of embodiment 1 extracts;

Fig. 5 is that the electric power line pole tower image that gathers judges whether electric power line pole tower exists the process flow diagram of Bird's Nest while being several pictures;

Fig. 6 is the filtered picture of several pictures Gabor that embodiment 2 provides;

Fig. 7 is ELM neural network diagram.

Embodiment

Below in conjunction with accompanying drawing, preferred embodiment is elaborated.Should be emphasized that, following explanation is only exemplary, rather than in order to limit the scope of the invention and to apply.

Fig. 1 is the recognition methods process flow diagram of electric power line pole tower Bird's Nest provided by the invention.As shown in Figure 1, the recognition methods of electric power line pole tower Bird's Nest provided by the invention comprises:

Step 1: obtain electric power line pole tower vision signal by helicopter routing inspection or camera, gather corresponding electric power line pole tower original object image.

In electric system, the mode of obtaining electric power line pole tower image has two kinds of modes conventionally: the one, and adopt helicopter to patrol and examine, obtain electric power line pole tower image; The 2nd, by camera device is installed on high-pressure tower, gather shaft tower image, the transmission line of electricity digital video signal collecting is sent back to Surveillance center by transmission channel in real time in the mode of video flowing, in Surveillance center, remote video monitoring is carried out in scene, and intercepting monitors target image from video flowing.

Step 2: determine that the electric power line pole tower image gathering is single width picture or several pictures.

Step 3: whether have Bird's Nest according to the electric power line pole tower spectral discrimination electric power line pole tower gathering.

In the image gathering, have the picture of varying number, the method for processing is also different.For the electric power line pole tower image gathering, be single width picture, the present invention adopts connected domain method to determine whether electric power line pole tower exists Bird's Nest.

Embodiment 1

Fig. 2 is that the electric power line pole tower image that gathers judges whether electric power line pole tower exists the process flow diagram of Bird's Nest while being single width picture.As shown in Figure 2, adopt connected domain method to determine whether electric power line pole tower exists the detailed process of Bird's Nest to be:

Sub-step 11: single width picture is carried out to gray scale processing and obtain gray processing picture; Gray processing picture is carried out to two-value processing and obtain binaryzation picture.

The method that picture is carried out to gray scale processing has multiple, such as maximum value process, mean value method and weighted average method.Maximum value process gets maximal value in three color components of each pixel (R, G, B) as the gray-scale value of this pixel.Mean value method is to get the mean value of three color components of each pixel (R, G, B) as the gray-scale value of this pixel.Weighted average method is first three color components of each pixel (R, G, B) to be multiplied by respectively to weights separately, then the gray-scale value using acquired results summation as this pixel.In the present embodiment, the method that gray scale is processed is restriction not, uses any in said method all can.

Single width picture is being carried out after gray scale processing, carry out two-value processing to gray scale picture, the present embodiment adopts global threshold method to carry out two-value processing to gray processing picture, comprising:

Steps A: using the mean value T0 of the gray scale maximal value of gray processing picture and minimum gray value as initial value.

Step B: make k=0.

Step C: the pixel in gray processing picture is divided into object pixel and background pixel, and the grey scale pixel value in gray processing picture is less than T _ktime, this pixel is object pixel; Grey scale pixel value in gray processing picture is more than or equal to T _ktime, this pixel is background pixel.

Step D: the average gray value Z that calculates respectively object pixel _aaverage gray value Z with background pixel _b.Wherein, $Z_A=\frac{\underset{z(i,j)<T_k}{\mathrm{\&Sigma;}}z(i,j)\&times;N(i,j)}{\underset{z(i,j)<T_k}{\mathrm{\&Sigma;}}N(i,j)},Z_B=\frac{\underset{z(i,j){\&GreaterEqual;T}_k}{\mathrm{\&Sigma;}}z(i,j)\&times;N(i,j)}{\underset{z(i,j){\&GreaterEqual;T}_k}{\mathrm{\&Sigma;}}N(i,j)},$ Z (i, j) is the gray-scale value of pixel (i, j), and N (i, j) is that in gray processing picture, gray-scale value is the quantity of Z (i, j).

Step e: order $T_{k+1}=\frac{Z_A+{\mathrm{<figure-callout\; id="2"\; label="Z\; B"\; filenames="HDA0000404710880000011.png,HDA0000404710880000032.png"\; state="\{\{state\}\}">Z</figure-callout>}}_B}{2}.$

Step F: judgement T _k=T _k+1whether set up, if T _k=T _k+1, perform step G; Otherwise, make k=k+1, return to step C.

Step G: the binary value of all object pixels is set to 0, the binary value of all background pixels is set to 1.

By above-mentioned steps A-G, gray processing picture is converted to binaryzation picture.Fig. 3 has shown electric power line pole tower picture that a width that the present embodiment provides the exists Bird's Nest result after gray processing and binary conversion treatment.

Sub-step 12: binaryzation picture is divided into M sub-pictures, in each sub-pictures, add up respectively binary value and be 0 the shared ratio of pixel, on duty is that 0 pixel is communicated with and shared ratio is greater than the threshold value of setting, judges that electric power line pole tower exists Bird's Nest.

Sub-pictures intermediate value is whether 0 pixel is communicated with and can adopts judgement in the following method:

First: utilize the gray-scale value of each pixel of sub-pictures to calculate the center of gravity of sub-pictures .Wherein, the horizontal ordinate of center of gravity $\stackrel{\&OverBar;}{x}=\frac{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}z(i,j)\&times;}{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}z(i,j)},$ The ordinate of center of gravity $\stackrel{\&OverBar;}{y}=\frac{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}z(i,j)\&times;i}{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}z(i,j)}.$ Z (i, j) is the gray-scale value of the pixel (i, j) of sub-pictures, the number of lines of pixels that n is sub-pictures, the pixel columns that m is sub-pictures.

Then: and if center of gravity

in adjacent pixel, the binary value that has 3 pixels at least is 0, and the pixel that in sub-pictures, binary value is 0 is communicated with.If with center of gravity

in adjacent pixel, the pixel that binary value is 0 is 0,1 or 2, and the pixel that in sub-pictures, binary value is 0 is disconnected.

Fig. 4 is the sub-pictures of the single width picture after binaryzation, and the pixel that this sub-pictures intermediate value is 0 is communicated with, and judges that electric power line pole tower exists Bird's Nest.

Embodiment 2

For the electric power line pole tower image gathering, be several pictures, the present invention adopts machine learning method to determine whether electric power line pole tower exists Bird's Nest.

Fig. 5 is that the electric power line pole tower image that gathers judges whether electric power line pole tower exists the process flow diagram of Bird's Nest while being several pictures.As shown in Figure 5, adopt machine learning method to determine whether electric power line pole tower exists Bird's Nest to comprise following sub-step:

Sub-step 21: choose p and determined picture and p the definite picture that does not have Bird's Nest that has Bird's Nest, p is setting value.

Be there is to the picture of Bird's Nest and definitely do not exist the picture of Bird's Nest to be divided into respectively two groups in determining of choosing.In the present embodiment, p=45.45 have been determined and existed the picture of Bird's Nest to be divided into two groups, one group of 30 picture, another organizes 15 pictures.Similarly, 45 have been determined and do not existed the picture of Bird's Nest to be divided into two groups, one group of 30 picture, another organizes 15 pictures.

By determining first group in the picture have Bird's Nest and having determined that first in the picture that does not have Bird's Nest combines, form the first picture group (having 60, picture), for generating training data collection.By determining second group in the picture that has Bird's Nest and having determined that second group in the picture that does not have Bird's Nest forms second picture group (30, total picture), for generating test data set.

Sub-step 22: each picture to 90 pictures carries out gray scale processing.

Gray scale is processed and is adopted weighted average method, and the grey scale pixel value computing formula of picture is: I=0.220R+0.587G+0.114B.Wherein, R, G, B are respectively the component of three colors of original image pixel, and 0.220,0.587 and 0.114 is respectively the weights of three components.

Sub-step 23: utilize bilinear interpolation method to be normalized the picture of processing through gray scale, generate the gray scale picture of setting pixel.

90 pictures after this step is processed step 22 gray scale are normalized, and each picture are converted to the gray scale picture of 256x256 pixel.

Sub-step 24: the image data matrix that the gray scale picture of all setting pixels is carried out obtaining after Gabor filtering combines.

Gray scale picture to each 256x256 pixel carries out Gabor filtering, obtains image data matrix.When carrying out filtering, Gabor filtering adopts formula: ${\tilde{I}}_j(x,y)=I_j(x,y)\&CircleTimes;G(x,y,\mathrm{\&theta;},u,\mathrm{\&sigma;}),$ Wherein ${\tilde{I}}_j(x,y)$ For the image data matrix that the gray scale picture of j 256 * 256 pixels is carried out obtaining after Gabor filtering, I _j(x, y) is the pixel grey scale value matrix of the gray scale picture of j 256 * 256 pixels, j=1, and 2 ..., 90,

for convolution algorithm.G (x, y, θ, u, σ) is Gabor wave filter, and

$G(x,y,\mathrm{\&theta;},u,\mathrm{\&sigma;})=\frac{1}{{2\mathrm{\&pi;\&sigma;}}^2}\exp \{-\frac{x^2+{\mathrm{<figure-callout\; id="2"\; label="y"\; filenames="HDA0000404710880000011.png,HDA0000404710880000032.png"\; state="\{\{state\}\}">y</figure-callout>}}^2}{{2\mathrm{\&sigma;}}^2}\}\exp \left\{2\mathrm{\&pi;i}(\mathrm{ux}\cos \mathrm{\&theta;}+\mathrm{uy}\sin \mathrm{\&theta;})\right\}$

X and y are respectively horizontal ordinate and the ordinate of the pixel of the gray scale picture of setting pixel, and i is that imaginary symbols is θ be Gabor filter function direction control coefrficient and

k=1,2 ..., 8.Centered by u frequency and

p=0,1,2,3,4.σ be Gaussian envelope standard deviation and

Because the pixel grey scale value matrix of the gray scale picture of 256 * 256 pixels is matrixes of 256 * 256, therefore, through Gabor filtering, can obtain 1 * 65536 image data matrix.

The image data matrix combination that picture for generating training data collection is obtained through gray scale processing, normalized and Gabor filtering, obtains target data matrix $X=\left[\begin{array}{c}{\tilde{I}}_1(x,y)\\ {\tilde{I}}_2(x,y)\\ \&CenterDot;\\ \&CenterDot;\\ \&CenterDot;\\ {\tilde{I}}_{60}(x,y)\end{array}\right].$ Image data matrix combination by obtaining through gray scale processing, normalized and Gabor filtering for generating the picture of test data set, obtains target data matrix $X^{\&prime;}=\left[\begin{array}{c}{\tilde{I}}_1(x,y)\\ {\tilde{I}}_2(x,y)\\ \&CenterDot;\\ \&CenterDot;\\ \&CenterDot;\\ {\tilde{I}}_{30}(x,y)\end{array}\right].$

Sub-step 25: adopt principal component analysis (PCA) to extract the characteristic of the filtered image data matrix of Gabor, generating training data collection and test data set.

Respectively target data matrix X and target data matrix X ' are carried out to principal component analysis (PCA), generating training data collection and test data set.The target data matrix X of take is example, and principal component analysis (PCA) adopts formula:

$\left\{\begin{array}{c}\max V=\max u_1^{T}X{X}^T{u}_1\\ s.t.{\left|\right|{\mathrm{<figure-callout\; id="1"\; label="u"\; filenames="HDA0000404710880000011.png,HDA0000404710880000032.png"\; state="\{\{state\}\}">u</figure-callout>}}_1\left|\right|}_2^2=1\end{array}\right.$

Wherein, u ₁for base vector to be asked,

for the transposition of base vector to be asked, X is target data matrix, and X is target data transpose of a matrix, || || ₂for the computing of 2-norm.When obtaining, make

maximum base vector u ₁after, according to

the data matrix obtaining is characteristic matrix, and each row of this characteristic matrix are the concentrated data of training data.The object of principal component analysis (PCA) is dimensionality reduction, and target data matrix X is through principal component analysis (PCA), and the characteristic matrix of extraction can drop to 60 * 50 by dimension, and each row of this characteristic matrix are the concentrated data of training data.

Sub-step 26: utilize training dataset training ELM neural network, and utilize test data set checking ELM neural network model.

Utilize training dataset training ELM neural network, actual is the process of determining ELM neural network parameter.

First: produce at random hidden layer node parameter a _land b _l,

for hidden layer neuron number.In the present embodiment

Secondly: according to formula calculate hidden layer output matrix H.Wherein, g (a _l, b _l, x _j) be activation function and

j=1,2 ..., N, the dimension that N is training dataset.

Last: according to formula β=H ⁺y calculates output weights β.Y be output valve and

in picture, exist Bird's Nest to be, output valve is 1; In picture, do not exist Bird's Nest to be, output valve is 0.

If ELM neural network can free from error prediction training data, the weights of hidden layer and output layer have solution so.Especially, when

time, certainly there is solution.But in practical problems,

much smaller than N, solve so the problem of weight vector without separating often, be i.e. between network output and actual value, have error, can define cost function and be: J=(H β-Y) ^t(H β-Y).For 2 kinds of situations, solve optimum weight vector, make loss function J minimum:

(1) if H is the matrix of row full rank, can find best weights by least square so, it is separated as β=H ⁺y, H ⁺=(H ^th ^-1) ^t.

(2) if when H is the matrix of non-row full rank, with the generalized inverse matrix of Singular-value Decomposition Solution H, calculate best weight value.

By above-mentioned training, can determine the output weights β of ELM nerve net, utilize test data set can verify the error of output weights β.

Sub-step 27: utilize several picture generating feature data sets.

When the electric power line pole tower original object image gathering is several pictures, first utilize several picture generating feature data sets.Its process is the same with the process that the first picture group generating training data collection and second picture group generate test data set.Also be first several pictures to be carried out to gray scale processing, then the picture of gray scale being processed is converted into the gray scale picture of 256 * 256 pixels, again the gray scale picture of 256 * 256 pixels is carried out to Gabor filtering (as shown in Figure 6), by after the image data matrix combination obtaining, adopt principal component analysis (PCA) dimensionality reduction, obtain final characteristic data set.

Sub-step 28: characteristic data set is input in trained ELM neural network and calculates output valve.

Because ELM neural network is through training, the parameters such as its hidden layer node parameter, hidden neuron number, output weights are known, as shown in Figure 7.Therefore characteristic data set is input in trained ELM neural network, can calculates definite output valve.

Sub-step 29: judge according to output valve whether electric power line pole tower exists Bird's Nest.

Owing to setting output valve in training process, be 1 to represent that electric power line pole tower exists Bird's Nest, output valve represents that electric power line pole tower does not exist Bird's Nest for-1, while so in the end judging, when output valve is 1, there is Bird's Nest in electric power line pole tower, when output valve is-1, there is not Bird's Nest in electric power line pole tower, thereby realized the identification whether several pictures exist Bird's Nest.

In the identification of several pictures, owing to having adopted known Bird's Nest classification results figure, therefore the picture result of output is the probability results having in statistical significance.For further determining the result of several pictures, the picture group that can repeatedly intercept same target is carried out arithmetic mean to obtain total result.

The above; be only the present invention's embodiment preferably, but protection scope of the present invention is not limited to this, is anyly familiar with in technical scope that those skilled in the art disclose in the present invention; the variation that can expect easily or replacement, within all should being encompassed in protection scope of the present invention.Therefore, protection scope of the present invention should be as the criterion with the protection domain of claim.

Claims (3)

1. a recognition methods for electric power line pole tower Bird's Nest, is characterized in that described method comprises:

Step 1: obtain electric power line pole tower vision signal by helicopter routing inspection or camera, gather corresponding electric power line pole tower original object image;

Step 2: determine that the electric power line pole tower image gathering is single width picture or several pictures;

Step 3: whether have Bird's Nest according to the electric power line pole tower spectral discrimination electric power line pole tower gathering.

2. recognition methods according to claim 1, is characterized in that when the electric power line pole tower image gathering is single width picture, and whether the described electric power line pole tower spectral discrimination electric power line pole tower according to collection exists Bird's Nest to comprise following sub-step:

Sub-step 11: single width picture is carried out to gray scale processing and obtain gray processing picture; Gray processing picture is carried out to two-value processing and obtain binaryzation picture;

Sub-step 12: binaryzation picture is divided into M sub-pictures, in each sub-pictures, add up respectively binary value and be 0 the shared ratio of pixel, on duty is that 0 pixel is communicated with and shared ratio is greater than the threshold value of setting, judges that electric power line pole tower exists Bird's Nest.

3. recognition methods according to claim 1, is characterized in that when the electric power line pole tower image gathering is several pictures, and whether the described electric power line pole tower spectral discrimination electric power line pole tower according to collection exists Bird's Nest to comprise following sub-step:

Sub-step 21: choose p and determined picture and p the definite picture that does not have Bird's Nest that has Bird's Nest, p is setting value;

Sub-step 22: the picture of choosing is carried out to gray processing processing;

Sub-step 23: the picture that utilizes bilinear interpolation method to process gray processing is normalized, generates the gray scale picture of setting pixel;

Sub-step 24: the image data matrix that the gray scale picture of all setting pixels is carried out obtaining after Gabor filtering combines;

Sub-step 25: adopt principal component analysis (PCA) to extract the characteristic of the filtered image data matrix of Gabor, generating training data collection and test data set;

Sub-step 26: utilize training dataset training ELM neural network, and utilize test data set checking ELM neural network model;

Sub-step 27: utilize several picture generating feature data sets;

Sub-step 28: described characteristic data set is input in trained ELM neural network and calculates output valve;

Sub-step 29: judge according to described output valve whether electric power line pole tower exists Bird's Nest.

Images