CN105572541A - High-voltage line patrol fault detection method and system based on visual attention mechanism - Google Patents

Abstract

The invention discloses a high-voltage line patrol fault detection system based on a visual attention mechanism. The system comprises an image sensor, an ADC analog-to-digital conversion module, an FPGA module, a DDR storage module, a DSP image processing module, an automatic control module and a communication module. The automatic control module is mainly used for receiving an instruction signal of the communication module and controlling operation of the other modules, and when receiving a fault signal from the DSP, notifying the communication module to send the fault information to a control station; the image sensor is used for carrying out video image acquisition, and after ADC analog-to-digital conversion, video images are subjected to video sampling and image preprocessing by the FPGA module; and the DSP image processing module is used for carrying out fault detection and identification on video data in an RAM based on the visual attention mechanism. The invention also discloses a high-voltage line patrol fault detection method based on the visual attention mechanism. Power transmission line characteristics are extracted through an image identification method, and then, fault detection is carried out, and thus the problems of high intensity, high cost and low efficiency and the like of a conventional patrol mode can be solved.

Description

A kind of high-voltage maintenance fault detection method of view-based access control model attention mechanism and system

Technical field

The present invention relates to electric power monitoring field, be specifically related to a kind of high-voltage maintenance fault detection method and system of view-based access control model attention mechanism.

Background technology

Power transmission line, as the carrier of electric power transfer, is electric power system important component part.At present, most of power transmission line and miscellaneous part are all exposed to open air, are corroded and destroy, if find defect reparation can cause serious security incident not in time.Manual inspection is the most frequently used routine inspection mode.High voltage transmission line territorial scope is wide, distance, and the geographical environment of power transmission line is severe, and manual inspection task difficulty is increasing, and efficiency is low, is easy to cause undetected flase drop.From the nineties in 20th century, helicopter uses more and more extensive in the works about electric power such as power-line patrolling.Helicopter carries various checkout equipment, still by the safety case of eye-observation power transmission line, observe the requirement of power transmission line situation to technician increase from shooting with video-corder fast image, and power transmission line corridor environment is severe, the security of flying can not be ensured completely.

In recent years along with development such as automatic control technology, GPS navigation technology, wireless communication technologys, the advantages such as unmanned plane is lightweight with it, volume is little, dirigibility is large are extended to civil area from military field gradually, apply more and more extensive.The line walking mode of existing employing UAV flight camera, major part is that view data is sent to control station, manually checks one by one, and its efficiency is also low.

Summary of the invention

For the efficiency for above high-voltage maintenance Fault Identification is low, the invention provides a kind of high-voltage maintenance fault detection method and system of efficient view-based access control model attention mechanism.This system is based on computer vision technique, first power transmission line state capture is carried out by imaging device, the image-recognizing method of view-based access control model attention mechanism is utilized to extract power transmission line feature, and then carry out failure exception detection, the problem such as the intensity that can solve traditional routine inspection mode is large, cost is high, efficiency is low.

Technical scheme of the present invention is:

A high-voltage maintenance fault detection system for view-based access control model attention mechanism, comprising the imageing sensor for taking transmission line, carrying out analog-to-digital ADC module, FPGA module, the DDR memory module of carrying out video backup, DSP image processing module, automatic control module and communication module.Imageing sensor obtains image, is undertaken sampling and Image semantic classification after ADC digital-to-analog conversion by FPGA, then by image buffer storage in RAM.Finally by DSP module, fault detect is carried out to the image information in RAM, if the fault of finding that there is, via automatic control module notice communication module, fault picture data message can be sent to control station to realize reporting to the police.On the other hand.Communication module receives the command signal of control console, carries out optimum configurations work accordingly via automatic control module.

Described imageing sensor is high-resolution image sensors, and the driving of imageing sensor is exported by FPGA, and driving coupling is carried out through amplifier in centre.

Described FPGA module, controls sampling and the pre-service of video image, by image buffer storage in RAM, for subsequent treatment.The bottom preprocessed data amount of real-time video image processing is large, but algorithm is relatively simple, is adapted at realizing in FPGA.FPGA also supports parallel and flowing structure and embedded DSP algorithm in addition, such FPGA can by the concurrent working of multiple processing unit, there is the algorithm of a large amount of multiplying by hardware implementing, the efficiency of fault detect can be improved, taken into account speed and dirigibility well.

Described communication module, can be GSM or the CDMA2000 pattern of 2G network, or is CDMA2000, WCDMA, TD-SCDMA, WiMAX pattern of 3G network, or is LTE, LTE-Advanced, WiMax, WirelessMAN pattern of 4G network.

Described automatic control module adopts single-chip microcomputer to be controller, control DSP, FPGA and the work of communication module.Automatic control module performs corresponding optimum configurations by the console instruction received from communication module.When receiving the failure message from DSP, then can control communication module and failure message is sent to control station, failure message can comprise the information such as trouble spot image, fault category and geographic position.

Described DSP image processing module is mainly used in image procossing, when fault having been detected, fault can be classified, more sorted failure message is issued automatic control module.

Present invention also offers a kind of high-voltage maintenance fault detection method of view-based access control model attention mechanism, comprise following step:

(1) Image semantic classification

Shooting due to video image is that carry completes at the imageing sensor of many rotor wing unmanned aerial vehicles, when image acquisition, during digitizing, random noise introduced by electronic component, over-exposed or not enough generation grain noise, the image motion that imaging system and the relative motion being taken scenery cause is fuzzy, therefore in order to reduce the interference of extracting target components such as power transmission lines, first Image semantic classification to be carried out.Image semantic classification mainly comprises image gray processing, optical correction, histogram equalization, image denoising based on wavelet transformation.

(2) extraction of power transmission line

First use Ratio operator pretreated image to be detected to the edge of image, then carry out Morphological scale-space, remove the planar Noise and Interference image in image.Adopt the random Hough transformation method based on Gradient direction information, then the feature of foundation line of electric force: power transmission line is long, generally more than one; Due to shooting at close range, power transmission line is similar to straight line in the picture, often runs through entire image, and pars intermedia branch is separated by shaft tower, can remove the line that angle differs greatly, and determines which line is line of electric force.The position that line of electric force is residing in captured picture can be determined like this.

When power transmission line region divided out after, by region growing method by around field and the straight line pixel with similar features include, thus by extend undetected for two ends, or the several line segments on same straight line are coupled together.

(3) the remarkable focus of the image of view-based access control model attention mechanism is extracted

Image is carried out whether foreign matter attachment detects employing based on Itti visual attention model.Comprise the following steps:

Step 1, to input image carry out multiple dimensioned division.Discrete Linear Gaussian filter is used to carry out the level and smooth of horizontal and vertical direction and down-sampled process to image.

Step 2, color, brightness and three, direction Visual Feature Retrieval Process; Specifically be divided into:

A), the extraction of color characteristic.Suppose r, g, b are the component of the redness of original input picture, green, blue three passages, then the formula of four colors of red-green passage and blue-yellow passage is as follows:

R＝r-(g+b)/2

G＝g-(r+b)/2

B＝b-(r+g)/2

Y＝(r+g)/2-|r-g|/2-b

RGBY wherein represents red, green, blue and yellow four kinds of colors of broad sense.Poor taking absolute value is done to two colors opposed in red-green passage and blue-yellow passage and just can obtain 2 Color Channels:

RG＝|R-G|

BY＝|B-Y|

B), the extraction of brightness.Monochrome information I represents, then I is:

$I=\frac{r+g+b}{3}$

Need with the image zooming-out brightness of gaussian pyramid to every one deck.

C), directional characteristic extraction.Use Gabor filter to image zooming-out direction character.The mathematical formulae of two-dimensional Gabor filter is:

$h(x,y)=g(x_{{\mathrm{\θ}}_i},y_{{\mathrm{\θ}}_i})\left(cos\right(2{\mathrm{\πfx}}_{{\mathrm{\θ}}_i})+jsin(2{\mathrm{\πfx}}_{{\mathrm{\θ}}_i}\left)\right)$

$g(x,y)=\exp (-\frac{x^2}{2{\mathrm{\σ}}_x^2}-\frac{x^2}{2{\mathrm{\σ}}_y^2})$

$x_{{\mathrm{\θ}}_i}=x{\mathrm{cos\θ}}_i+y{\mathrm{sin\θ}}_i$

$y_{{\mathrm{\θ}}_i}=-x{\mathrm{sin\θ}}_i+y{\mathrm{cos\θ}}_i$

In formula: x, y denotation coordination variable, σ _xand σ _yrepresent the variance of Gauss on x, y direction respectively, θ _ido not get 0 °, 45 °, 90 °, 135 °, 4 direction character figure that the Garbor filtering carrying out 4 directions to luminance graph I obtains.

The merging of step 3, characteristic pattern.Before feature extraction, multiple dimensioned division has been carried out to image, make the type of characteristic pattern and resolution inconsistent, therefore need by following 2 steps complete feature merge.

A), identical type but the merging of the different characteristic pattern of yardstick

Due to the center-periphery feature that Itti visual attention model is receptive field in simulation human eye retina, therefore center-periphery operation is carried out to the multi-scale image that it is obtained by pyramid model.

The computing formula of the characteristic remarkable picture of the brightness image I obtained thus is:

I(c,s)＝|I(c)ΘI(s)|

Wherein, I (c, s) significantly schemes for brightness, and I (c) and I (s) represents central brightness and surrounding brightness characteristic remarkable picture respectively;

The computing formula of the characteristic remarkable picture of color characteristic image is:

RG(c,s)＝|(R(c)-G(c))Θ(R(s)-G(s))|

BY(c,s)＝|(B(c)-Y(c))Θ(B(s)-Y(s))|

The computing formula of the remarkable figure in direction is

O(c,s,θ)＝|O(c,θ)ΘO(s,θ)|

In formula: RG (c, s) red green agonist character figure is represented, R (c) represents red central feature figure, G (c) represents green central feature figure, R (s) represents red surrounding features figure, G (s) represents green surrounding features figure, BY (c, s) blue yellow agonist character figure is represented, B (c) represents blue central feature figure, Y (c) represents yellow center characteristic pattern, B (s) represents blue ambient characteristic pattern, Y (s) represents yellow surrounding features figure, O (c, s, θ) represent orientative feature figure, O (c, θ) represent center hold characteristic pattern, O (s, θ) represent orientative feature figure around,

42 width Itti visual attention model characteristic remarkable pictures can be obtained: 6 width luminance graphs, 12 width colors are significantly schemed, 24 width directions are significantly schemed by above formulae discovery.

Itti visual attention model needs this 42 width characteristic remarkable image to merge, and uses a kind of normalization operator N ().By normalization operator N (), the marking area in characteristic pattern is all exaggerated, obviously highlights outstanding target signature to be noted, the interference of the feature that inhibit contribution little.The characteristic pattern of different scale is normalized the remarkable figure that just can form this feature passage secondary, formula is as follows:

$\stackrel{\‾}{I}=\underset{c=2}{\overset{4}{\⊕}}\underset{s=c+3}{\overset{c+4}{\⊕}}N\left(I\right(c,.s\left)\right)$

$\stackrel{\‾}{C}=\underset{c=2}{\overset{4}{\⊕}}\underset{s=c+3}{\overset{c+4}{\⊕}}\[N\left(RG\right(c,s\left)\right)+N\left(BY\right(c,s\left)\right)\]$

B), the merging of dissimilar characteristic pattern

Can obtain 3 dissimilar characteristic patterns by step 1, their resolution is the same, and the remarkable figure of these three passages is merged into S by the method be added in proportion by Itti model, and formula is as follows:

$S=\frac{1}{3}\left(N\right(\stackrel{\‾}{I})+N(\stackrel{\‾}{C})+N(\stackrel{\‾}{O}\left)\right)$

C), the extraction of vision attention focus and transfer

The competition mechanism that the victor is a king is adopted to extract the most significant target and note it, and by forbidding that return mechanisms finds next focus-of-attention.Wherein, if the power transmission line position detected in focal position distance (2) is greater than certain threshold value, then cast out.

(4) transmission line fault detects

By Itti attention mechanism in step (3), determine multiple focus-of-attention.Calculate the invariant moment features of object in each focus.

A. this object is on power transmission line, but is greater than certain threshold value with the not bending moment of the device (as stockbridge damper, insulator, shaft tower etc.) on known power transmission line is different, can preliminary judgement be foreign matter attachment.

B. the not bending moment of this object is similar to insulator, analyzes its shape facility, wavelet character further, can determine whether as insulator.If so, the string that falls then carrying out Shape-based interpolation number detects, based on fault detects such as the crackle of the method for region description and surface filths.

(5) detect based on the power transmission line of the chain code stock that breaks

Power transmission line edge in scan image, to its chain code value be 0 and non-zero pixel carry out record respectively, using wherein chain code value be 0 and the longest continuously one section as datum line; If the chain code value that the pixel in this sequence is corresponding is non-vanishing, the then distance of calculation level and sequence reference line, to this distance setting threshold value T1, and adds up the sum of all pixels that vertical range is greater than threshold value, if such pixel number has exceeded certain limit T2, be then suspicious disconnected stock by spectral discrimination.

Compared with prior art, the invention has the beneficial effects as follows:

(1) the present invention utilizes computer vision to carry out real-time online process to the image collected, and in time fault is fed back to control desk.Existing fault detection system needs transmission of video images to return control desk, and carry out fault detect by human eye, human cost is higher.The real-time intelligent process of this technology, improves line walking efficiency.In addition existing by by line walking video storage in internal memory, after patrolling and examining end, hold software to carry out unifying process by PC, have very large hysteresis quality, the maintenance for electrical network has harmful effect, real-time detects necessary, and the present invention well solves this drawback.

(2) the present invention adopts the transmission line fault detection method based on attention mechanism, finds the remarkable focal position in image fast, carries out signature analysis and fault detect to it.

(3) the present invention is by using DSP and FPGA collaborative work, carries out transmission line Failure detection and identification with special hardware configuration, and compare PC and hold software speed faster, efficiency is higher.

(4) the present invention is not when contacting high voltage transmission line, carries out fault detect, reduces the manpower and materials of hi-line being installed to extra means, well ensure that personal safety.

Accompanying drawing explanation

Fig. 1 is structural drawing of the present invention.

Embodiment

Below in conjunction with accompanying drawing and specific embodiments, the present invention is described in further detail.

As shown in Figure 1, the high-voltage maintenance fault detection system of view-based access control model attention mechanism in this enforcement, comprises imageing sensor, ADC analog-to-digital conversion module, FPGA module, the memory module of carrying out video backup, DSP image processing module, automatic control module and communication module.

The course of work of above-mentioned high-voltage maintenance fault detection system is as follows:

(1) system electrification, is in wait mode of operation after each module initialization;

(2) communication module receives instruction and sends to control module, and control module carries out instructions parse to control other modules; If start fault detect instruction, each module enters corresponding mode of operation, goes to step (3); If stop fault detect instruction, go to step (7);

(3) imageing sensor carries out video image acquisition, and via transferring to FPGA to complete sampling and Image semantic classification after ADC analog to digital conversion, pretreated view data is cached in RAM;

(4) executing arithmetic in DSP, carries out Failure detection and identification to the video data of RAM.If there is fault, first grade classification slightly can be carried out to fault, then via control module control communication module, failure message is sent to control station, go to step (5); If the fault of not detecting, go to step (6).

(5) when control station receives failure message, the yellow in control station, red led correspondingly can glimmer according to fault level.

(6) if do not receive new instruction, repeat step (3) ~ (5), if receive new instruction, go to step (2);

(7) system stops fault detect, and each module enters wait mode of operation.

In the present embodiment, high-voltage maintenance is carried out to the method for fault detect, concrete steps comprise:

Step 1: pre-service is carried out to image.

Pre-service comprises image gray processing, optical correction, histogram equalization, image denoising based on wavelet transformation.

Step 2: extract power transmission line in image after the pre-treatment, detailed process is:

2.1) Ratio operator is used pretreated image to be detected to the edge of image, and through Morphological scale-space to remove the planar Noise and Interference image in image;

2.2) the random Hough transformation method based on Gradient direction information is adopted, and the feature of foundation line of electric force, determine the position that line of electric force is residing in the picture;

2.3) split the power transmission line region in image, and by region growing method by around field and the straight line pixel with similar features include, thus to couple together by extend undetected for two ends or by the several line segments on same straight line.

Step 3: the remarkable focus of image of view-based access control model attention mechanism is extracted;

Image is carried out whether foreign matter attachment detects employing based on Itti visual attention model, specifically comprises:

3.1) adopt Discrete Linear Gaussian filter to carry out the level and smooth of horizontal and vertical direction and down-sampled process to image, complete the multiple dimensioned division of image;

3.2) color, brightness and three, direction visual signature is extracted in the image after division;

The extraction of color characteristic adopts two colors work differences to opposing in red-green passage and blue-yellow passage to take absolute value to obtain 2 Color Channels;

The extraction of brightness adopts gaussian pyramid to the image zooming-out brightness of every one deck;

Use Gabor filter to image zooming-out direction character.

3.3) extracted characteristic remarkable image is merged, and extract and transition diagram in vision attention focus; Specifically comprise:

3.3.1) to identical type but the characteristic pattern of different scale is normalized the remarkable figure merging into this feature passage secondary;

3.3.2) three that obtain dissimilar characteristic patterns are merged by the method be added in proportion;

3.3.3) extract the most significant target in image after merging and find focus-of-attention.

Step 4: the focus-of-attention obtained step 3, calculates the invariant moment features of object in each focus one by one with detection failure; Specifically comprise:

4.1) if the not bending moment of object and the not bending moment of known device is different is greater than certain threshold value in focus, be judged to be that foreign matter adheres to;

4.2) if the not bending moment of object is similar to insulator in focus, analyze its shape facility, wavelet character further, determine whether as insulator.

Step 5: the power transmission line edge in scan image, according to chain code value carry out power transmission line break stock detect;

Power transmission line edge in scan image, to its chain code value be 0 and non-zero pixel carry out record respectively, using wherein chain code value be 0 and the longest continuously one section as datum line; If the chain code value that the pixel in this sequence is corresponding is non-vanishing, the then distance of calculation level and datum line, to this distance setting threshold value T1, and adds up the sum of all pixels that vertical range is greater than threshold value, if this pixel number exceedes setting range T2, be then suspicious disconnected stock by spectral discrimination.

Claims (9)

1. a high-voltage maintenance fault detection system for view-based access control model attention mechanism, is characterized in that, comprises imageing sensor, ADC module, FPGA module, DDR memory module, DSP image processing module, automatic control module and communication module;

Described imageing sensor is for obtaining the status image of power transmission line;

Described ADC module is used for obtaining image to imageing sensor and carries out digital-to-analog conversion;

The picture signal that described FPGA module is used for ADC module exports is sampled and Image semantic classification, then by image buffer storage in RAM;

Described DSP image processing module is used for image procossing, when fault having been detected, fault can be classified, more sorted failure message is issued automatic control module.

2. a high-voltage maintenance fault detection method for view-based access control model attention mechanism, is characterized in that, comprise step:

1) pre-service is carried out to image;

2) power transmission line is extracted in image after the pre-treatment;

3) the remarkable focus of the image of view-based access control model attention mechanism is extracted;

4) to step 3) focus-of-attention that obtains, calculate the invariant moment features of object in each focus one by one with detection failure;

5) the power transmission line edge in scan image, according to chain code value carry out power transmission line break stock detect.

3. high-voltage maintenance fault detection method as claimed in claim 2, is characterized in that, described pre-service comprises image gray processing, optical correction, histogram equalization, image denoising based on wavelet transformation.

4. high-voltage maintenance fault detection method as claimed in claim 2, is characterized in that, described step 2) be specifically divided into:

2.1) Ratio operator is used pretreated image to be detected to the edge of image, and through Morphological scale-space to remove the planar Noise and Interference image in image;

2.2) the random Hough transformation method based on Gradient direction information is adopted, and the feature of foundation line of electric force, determine the position that line of electric force is residing in the picture;

2.3) split the power transmission line region in image, and by region growing method by around field and the straight line pixel with similar features include, thus to couple together by extend undetected for two ends or by the several line segments on same straight line.

5. high-voltage maintenance fault detection method as claimed in claim 2, is characterized in that, the step 3 described) in, image is carried out whether foreign matter attachment detects employing based on Itti visual attention model, specifically comprises the following steps:

3.1) adopt Discrete Linear Gaussian filter to carry out the level and smooth of horizontal and vertical direction and down-sampled process to image, complete the multiple dimensioned division of image;

3.2) color, brightness and three, direction visual signature is extracted in the image after division;

3.3) extracted characteristic remarkable image is merged, and extract and transition diagram in vision attention focus.

6. high-voltage maintenance fault detection method as claimed in claim 5, is characterized in that, described step 3.2) in:

The extraction of color characteristic adopts two colors work differences to opposing in red-green passage and blue-yellow passage to take absolute value to obtain 2 Color Channels;

The extraction of brightness adopts gaussian pyramid to the image zooming-out brightness of every one deck;

Use Gabor filter to image zooming-out direction character.

7. high-voltage maintenance fault detection method as claimed in claim 5, is characterized in that, described step 3.3) specifically comprise:

3.3.1) to identical type but the characteristic pattern of different scale is normalized the remarkable figure merging into this feature passage secondary;

3.3.2) three that obtain dissimilar characteristic patterns are merged by the method be added in proportion;

3.3.3) extract the most significant target in image after merging and find focus-of-attention.

8. high-voltage maintenance fault detection method as claimed in claim 2, is characterized in that, described step 4) in fault detect comprise:

4.1) if the not bending moment of object and the not bending moment of known device is different is greater than certain threshold value in focus, be judged to be that foreign matter adheres to;

4.2) if the not bending moment of object is similar to insulator in focus, analyze its shape facility, wavelet character further, determine whether as insulator.

9. high-voltage maintenance fault detection method as claimed in claim 2, is characterized in that, described step 5) be specially:

Power transmission line edge in scan image, to its chain code value be 0 and non-zero pixel carry out record respectively, using wherein chain code value be 0 and the longest continuously one section as datum line; If the chain code value that the pixel in this sequence is corresponding is non-vanishing, the then distance of calculation level and datum line, to this distance setting threshold value T1, and adds up the sum of all pixels that vertical range is greater than threshold value, if this pixel number exceedes setting range T2, be then suspicious disconnected stock by spectral discrimination.