CN110136097A - One kind being based on the pyramidal insulator breakdown recognition methods of feature and device - Google Patents
One kind being based on the pyramidal insulator breakdown recognition methods of feature and device Download PDFInfo
- Publication number
- CN110136097A CN110136097A CN201910283167.6A CN201910283167A CN110136097A CN 110136097 A CN110136097 A CN 110136097A CN 201910283167 A CN201910283167 A CN 201910283167A CN 110136097 A CN110136097 A CN 110136097A
- Authority
- CN
- China
- Prior art keywords
- insulator
- background image
- feature
- pyramidal
- insulator breakdown
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/06—Electricity, gas or water supply
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20016—Hierarchical, coarse-to-fine, multiscale or multiresolution image processing; Pyramid transform
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- Health & Medical Sciences (AREA)
- Economics (AREA)
- General Health & Medical Sciences (AREA)
- Primary Health Care (AREA)
- Artificial Intelligence (AREA)
- Water Supply & Treatment (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Human Resources & Organizations (AREA)
- Marketing (AREA)
- Quality & Reliability (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Life Sciences & Earth Sciences (AREA)
- Public Health (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses one kind to be based on the pyramidal insulator breakdown recognition methods of feature and device, this method comprises: obtaining the background image containing insulator；The background image is inputted into preset insulator breakdown identification model, insulator breakdown identification is carried out to the background image, identifies the faulty insulator in the background image.This method can fast and effeciently identify the failure of insulator, improve fault identification efficiency, it is ensured that the stable operation of electric system by constructing preset insulator breakdown identification model.
Description
Technical field
The present invention relates to image identification technical fields, more particularly to a kind of pyramidal insulator breakdown of feature that is based on to identify
Method and apparatus.
Background technique
Insulator is as one of the main component in power equipment, during the work time often due to electric field, harsh weather
The factors such as (dust, ice and snow, thunder and lightning) and self gravity break down, if perform corresponding processing not in time may threaten it is whole
The safety and stablization of a electric system are run.Therefore timely and accurately detection is carried out to insulation subgraph to be of great significance.Absolutely
Edge subgraph detects avoidable insulator and breaks down in the process of running, reduces the unnecessary loss generated due to burst accident
And maintenance shut-downs maintenance bring loss, and insulator image detection has the insulator breakdown diagnosis and maintenance in later period
Important value can help related personnel fast and accurately positioning failure and to repair, it is ensured that the stable operation of electric system.
In recent years, the method for distribution insulator breakdown detection is mainly by the way of unmanned plane inspection, but unmanned plane is shot
Photo background it is more complicated, by the way of traditional manual extraction feature to insulator breakdown identify when, vulnerable to background
Influence and cause gained feature cannot state insulator state well so that the recognition efficiency of insulator breakdown compares
It is low.
Summary of the invention
The object of the present invention is to provide one kind to be based on the pyramidal insulator breakdown recognition methods of feature and device, can be fast
Speed efficiently identifies the failure of insulator, improves fault identification efficiency, it is ensured that the stable operation of electric system.
The embodiment of the invention provides one kind to be based on the pyramidal insulator breakdown recognition methods of feature, comprising:
Obtain the background image containing insulator；
The background image is inputted into preset insulator breakdown identification model, insulator event is carried out to the background image
Barrier identification, identifies the faulty insulator in the background image.
Preferably, the insulator breakdown identification model, is constructed by following steps:
It acquires the background image that multiple contain insulator to be pre-processed, obtains image data set；
Convolutional neural networks model is constructed, and spy is carried out to described image data set by the convolutional neural networks model
Information extraction is levied, the insulator characteristic pattern of multiple different scales is obtained；
According to the insulator characteristic pattern construction feature pyramid network, and to each in the feature pyramid network
A pyramidal layer grade carries out candidate region division, obtains several candidate frames；
According to the candidate frame and the insulator characteristic pattern, position sensing score chart is constructed, the position sensing is obtained
The corresponding category score probability of score chart, and the corresponding class label of candidate frame according to the category score determine the probability；
Pass through the probability value of each class label in background image described in preset classifier calculated；
Using the candidate frame, the insulator characteristic pattern and the position sensing score chart as training sample, and according to
Batch gradient descent method is trained the preset classifier, obtains the insulator breakdown identification model for completing training.
Preferably, described multiple background images for containing insulator of acquisition are pre-processed, and obtain image data set, specifically
Include:
Amplification processing is carried out to multiple described background images for containing insulator；Wherein the mode of the amplification processing includes
Image is rotated, is cut out, saturation degree is adjusted, tone is adjusted and the time for exposure is adjusted；
Image category mark is carried out to amplification treated background image, and the background image after mark is normalized
Processing, obtains image data set.
Preferably, the convolutional neural networks model includes 5 convolution group modules and 1 average pond layer.
Preferably, described according to the insulator characteristic pattern construction feature pyramid network, and to the feature pyramid
Each of network pyramidal layer grade carries out candidate region division, obtains several candidate frames, specifically includes:
Using top-down path, obtained according to 4 convolution group modules rear in the convolutional neural networks model exhausted
Edge subcharacter figure construction feature pyramid；
Candidate regions are carried out to each of feature pyramid pyramidal layer grade according to preset candidate region network
Domain divides.
Preferably, described according to the candidate frame and the insulator characteristic pattern, position sensing score chart is constructed, institute is obtained
The corresponding category score probability of position sensing score chart is stated, and the candidate frame according to the category score determine the probability is corresponding
Class label specifically includes:
Increase default convolutional layer in the convolutional neural networks model, generates several position sensing score charts；
Obtain the category score probability of any position sensitivity score chart；
Using the maximum category score probability of numerical value as the corresponding class label of the candidate frame.
Preferably, the probability value by each class label in background image described in preset classifier calculated,
It specifically includes:
According to formulaPreset classifier is constructed, and passes through described point
Class device calculates the probability value of each class label in background image；
Wherein, rc(i, j/ θ)=∑(x, y) ∈ bin (i, j)zI, j, c(x+x0, y+y0/ θ)/n be c-th classification (i, j) son
The pondization in region responds；C is foreground object classification number in background image；zI, j, cFor in several described position sensing score charts
One position sensing score chart；N indicates the number of pixels in each sub-regions；6 be to learn to arrive in convolutional neural networks model
Parameter；(x0, y0) be the candidate frame is summed up it is average after to carry out the position that the c+1 that pondization ballot obtains is tieed up quick
Feel the centre coordinate in the score chart upper left corner.
Preferably, using the candidate frame, the insulator characteristic pattern and the position sensing score chart as training sample,
And the preset classifier is trained according to batch gradient descent method, obtain the insulator breakdown identification mould for completing training
Type specifically includes:
According to formula Loss=Lcls(Sc(θ), Sc*(θ))+λ[c*> 0] Lreg(t, t*), obtain the loss of the classifier
Function, and the classifier is trained according to the loss function；
Wherein,To intersect loss function, Sc
(θ) is the prediction label of candidate frame,For true tag, and its label value is only 0 or 1, and works asIt is 0
When, indicate that the candidate frame is background class label；For side
Boundary returns loss function,T=(tx, ty, tw, th) it is the candidate frame predicted
Location coordinate information, t*=(t* x, t* y, t* w, t* h) be true frame corresponding with candidate frame location coordinate information, x, y, w, h
Respectively indicate frame centre coordinate, width and height；λ is to return weight coefficient, and it is initialized as 10.
Preferably, described according to batch gradient descent method, the insulator breakdown identification model for completing training is carried out
Optimization, the insulator breakdown identification model optimized specifically include:
According to formula
Obtain completing the loss function of the insulator breakdown identification model of training；Wherein, NclsFor batch gradient decline
Per batch of quantity in method, NregIndicate the quantity of candidate frame；
The output valve in the insulator breakdown identification model for completing training is constantly iterated according to the loss function
It updates, until the output valve meets preset condition, deconditioning, the insulator breakdown identification model optimized.
Preferably, the preset condition is less than or equal to 0.0001.
The embodiment of the invention also provides one kind to be based on the pyramidal insulator breakdown identification device of feature, comprising:
Image collection module, for obtaining the background image containing insulator；
Insulator breakdown identification module, it is right for the background image to be inputted preset insulator breakdown identification model
The background image carries out insulator breakdown identification, identifies the faulty insulator in the background image.
Compared with the existing technology, provided in an embodiment of the present invention a kind of based on the pyramidal insulator breakdown identification side of feature
The beneficial effect of method is: described to be based on the pyramidal insulator breakdown recognition methods of feature, comprising: acquisition contains insulator
Background image；The background image is inputted into preset insulator breakdown identification model, insulator is carried out to the background image
Fault identification identifies the faulty insulator in the background image.This method identifies mould by constructing preset insulator breakdown
Type can fast and effeciently identify the failure of insulator, improve fault identification efficiency, it is ensured that the stable operation of electric system.
Detailed description of the invention
Fig. 1 is that a kind of process based on the pyramidal insulator breakdown recognition methods of feature provided in an embodiment of the present invention is shown
It is intended to；
Fig. 2 is a kind of background image containing insulator provided in an embodiment of the present invention；
Fig. 3 is provided in an embodiment of the present invention a kind of after based on the pyramidal insulator breakdown recognition methods of feature
Faulty insulator recognition result figure；
Fig. 4 is that a kind of structure based on the pyramidal insulator breakdown identification device of feature provided in an embodiment of the present invention is shown
It is intended to.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall within the protection scope of the present invention.
Referring to Fig. 1, it is provided in an embodiment of the present invention a kind of based on the pyramidal insulator breakdown identification side of feature
The flow diagram of method, it is described to be based on the pyramidal insulator breakdown recognition methods of feature, comprising:
S100: the background image containing insulator is obtained；
S200: inputting preset insulator breakdown identification model for the background image, carries out to the background image exhausted
Edge fault identification identifies the faulty insulator in the background image.
The new collected background image containing insulator is inputted in trained insulator breakdown identification model,
Export background image in insulator prediction classification, fast and effeciently realize insulator fault identification, facilitate staff and
When maintenance replacement is carried out to insulator assembly of problems, it is ensured that the stable operation of electric system.
In an alternative embodiment, the insulator breakdown identification model, is constructed by following steps:
It acquires the background image that multiple contain insulator to be pre-processed, obtains image data set；
Convolutional neural networks model is constructed, and spy is carried out to described image data set by the convolutional neural networks model
Information extraction is levied, the insulator characteristic pattern of multiple different scales is obtained；
According to the insulator characteristic pattern construction feature pyramid network, and to each in the feature pyramid network
A pyramidal layer grade carries out candidate region division, obtains several candidate frames；
According to the candidate frame and the insulator characteristic pattern, position sensing score chart is constructed, the position sensing is obtained
The corresponding category score probability of score chart, and the corresponding class label of candidate frame according to the category score determine the probability；
Pass through the probability value of each class label in background image described in preset classifier calculated；
Using the candidate frame, the insulator characteristic pattern and the position sensing score chart as training sample, and according to
Batch gradient descent method is trained the preset classifier, obtains the insulator breakdown identification model for completing training.
In the present embodiment, the calculation amount of whole process is efficiently reduced by constructing insulator breakdown identification model, fastly
Speed identification insulator breakdown.
In an alternative embodiment, described multiple background images for containing insulator of acquisition are pre-processed, and are obtained
Image data set specifically includes:
Amplification processing is carried out to multiple described background images for containing insulator；Wherein the mode of the amplification processing includes
Image is rotated, is cut out, saturation degree is adjusted, tone is adjusted and the time for exposure is adjusted；
Image category mark is carried out to amplification treated background image, and the background image after mark is normalized
Processing, obtains image data set.
In the present embodiment, due to the presence of take photo by plane personnel's technology, angle of taking photo by plane and weather environment factor, pass through unmanned plane
The available background image containing insulator taken photo by plane compares shortcoming, in order to avoid showing for over-fitting occurs in later period training pattern
As by carrying out amplification processing to background image, wherein rotation processing mainly rotates background image for interval with 10 degree
Operation, cutting out can be passed through with the upper left corner of random cropping background image, the upper right corner, the lower left corner, the lower right corner or middle section etc.
Amplification processing increases the data of training sample；
Wherein, image category mark is carried out mainly to institute present in background image to amplification treated background image
There is classification to be divided and marked, mainly divided with a rectangle frame, so that can be obtained by one includes image class
Distinguishing label (C), bezel locations coordinate information (tx, ty, tw, th) one 5 dimension data set；Then the position of data set is sat
Mark information is normalized.
In an alternative embodiment, the convolutional neural networks model includes 5 convolution group modules and 1 average pond
Change layer.
In the present embodiment, the convolutional neural networks model can be said by taking residual error network 52 (ResNet52) as an example
Bright, residual error network is made of 5 convolution group modules and average pond (average pool) layer, and passes through each convolution
The characteristic pattern that group module obtains is attached in the way of short connection with the convolutional layer of previous group, is formed the identical of feature and is reflected
Penetrate so that feature using low level large scale feature while so that parameter amount is held essentially constant.Wherein, first group
Convolution module (conv_1) only includes one layer of convolution operation, the feature of input picture can be extracted by convolution operation, then
Carry out corresponding Feature Dimension Reduction by one layer of maximum pond layer (max pooling), then pass through one layer of active coating (Relu) into
The Nonlinear Mapping of row feature；Second group of convolution (conv_2) module includes 3 groups of convolution kernels, and each group of convolution kernel includes 1*1,
The convolutional layer of 3*3 and 1*1 and one layer of active coating are reaching same convolution feature using the convolution operation of multiple 1*1 and 3*3
While effect, the effect for reducing parameter amount can achieve,.Third group convolution module (conv_3) includes 4 groups of convolution kernels, each
The convolution kernel of group includes the convolutional layer and one layer of active coating of 1*1,3*3 and 1*1；4th group of convolution module (conv_4) includes 6 groups
Convolution kernel, each group of convolution kernel include the convolutional layer and one layer of active coating of 1*1,3*3 and 1*1；5th group of convolution module
It (conv_5) include 3 groups of convolution kernels, each group of convolution kernel includes the convolutional layer and one layer of active coating of 1*1,3*3 and 1*1.By
Residual error network can form the insulator characteristic pattern of different scale.
In an alternative embodiment, described according to the insulator characteristic pattern construction feature pyramid network and right
Each of feature pyramid network pyramidal layer grade carries out candidate region division, obtains several candidate frames, specifically
Include:
Using top-down path, obtained according to 4 convolution group modules rear in the convolutional neural networks model exhausted
Edge subcharacter figure construction feature pyramid；
Candidate regions are carried out to each of feature pyramid pyramidal layer grade according to preset candidate region network
Domain divides.
In the present embodiment, according to three kinds of ratios, respectively 1: 2,1: 1 and 2: 1 carry out candidate frame division, i.e. each gold
Word tower level can mark off the candidate frame of three kinds of ratios, and residual error network is from second volume set modules to the 5th volume unit
Module one shares 4 volume set modules, therefore can mark off 12 kinds of candidate frames.
In an alternative embodiment, described according to the candidate frame and the insulator characteristic pattern, it is quick to construct position
Feel score chart, obtains the corresponding category score probability of the position sensing score chart, and according to the category score determine the probability
The corresponding class label of the candidate frame, specifically includes:
Increase default convolutional layer in the convolutional neural networks model, generates several position sensing score charts；
Obtain the category score probability of any position sensitivity score chart；
Using the maximum category score probability of numerical value as the corresponding class label of the candidate frame.
In the present embodiment, the convolution kernel of the default convolutional layer can be 1*1, by increasing the convolutional layer of a 1*1,
For generating the position sensing score chart in a channel k*k* (C+1)；Wherein every one kind has k*k position sensing score chart, the position
Sensitive score chart is used to describe the space lattice of corresponding position.Each position sensing score chart has C+1 channel, and C is preceding scenery
Body classification number, an additional background；Such as the candidate frame for w*h, it is intended to for candidate frame to be divided into the subregion of k*k, then it is each
The size of sub-regions is w/k*h/k.；General tension position sensitivity score chart can have several classification scoring probability, therefore
In order to guarantee the accuracy of data, need using the maximum category score probability of numerical value as the corresponding classification mark of the candidate frame
Label.
In an alternative embodiment, described to pass through each classification in background image described in preset classifier calculated
The probability value of label, specifically includes:
According to formulaPreset classifier is constructed, and passes through described point
Class device calculates the probability value of each class label in background image；
Wherein, rc(i, j/ θ)=∑(x, y) ∈ bin (i, j)zI, j, c(x+x0, y+y0/ θ)/n be c-th classification (i, j) son
The pondization in region responds；C is foreground object classification number in background image；zI, j, cFor in several described position sensing score charts
One position sensing score chart；N indicates the number of pixels in each sub-regions；6 be to learn to arrive in convolutional neural networks model
Parameter；(x0, y0) be the candidate frame is summed up it is average after to carry out the position that the c+1 that pondization ballot obtains is tieed up quick
Feel the centre coordinate in the score chart upper left corner.
It in the present embodiment, can be more according to obtained penalty values by calculating the penalty values of each class label
Accurately judge the discrimination of the corresponding class label of the candidate frame, provides data for the identification optimization of class label.
In an alternative embodiment, described by the candidate frame, the insulator characteristic pattern and the position sensing
Score chart is trained the preset classifier as training sample, and according to batch gradient descent method, obtains completing instruction
Experienced insulator breakdown identification model, specifically includes:
According to formulaObtain described point
The loss function of class device；
Wherein,To intersect loss function, Sc
(θ) is the prediction label of candidate frame,For true tag, and its label value is only 0 or 1, and works asIt is 0
When, indicate that the candidate frame is background class label；For side
Boundary returns loss function,T=(tx, ty, tw, th) it is the candidate frame predicted
Location coordinate information, t*=(t* x, t* y, t* w, t* h) be true frame corresponding with candidate frame location coordinate information, x, y, w, h
Respectively indicate frame centre coordinate, width and height；λ is to return weight coefficient, and it is initialized as 10；
According to formulaTo institute
Loss function is stated to optimize；Wherein, NclsFor batch of quantity every in the batch gradient descent method, NregIndicate candidate frame
Quantity；
When loss function output valve after optimization meets preset condition, stopping is trained the classifier, obtains
Complete the insulator breakdown identification model of training.
In the present embodiment, the classifier is carried out about by returning loss function in conjunction with intersection loss function and boundary
Beam is trained the classifier, keeps the classification data obtained more accurate.
Batch gradient descent method is explained, mainly the loss function of the classifier is constantly asked
It leads, the value progress parameter optimization for updating its loss function is constantly iterated, until the loss function L (S of output layerc(θ),
tx, ty, tw, th) meeting preset condition, deconditioning obtains the insulator breakdown identification model for completing training；Wherein, it asks
The formula led is respectively as follows:
In an alternative embodiment, the preset condition is less than or equal to 0.0001.
In the present embodiment, by setting a condition, make the insulator breakdown identification model deconditioning for completing training
Optimization, to obtain the insulator breakdown identification model of suitable optimization.
Referring to Fig. 4, it is provided in an embodiment of the present invention a kind of based on the pyramidal insulator breakdown identification dress of feature
The structural schematic diagram set, it is described to be based on the pyramidal insulator breakdown identification device of feature, comprising:
Image collection module 1, for obtaining the background image containing insulator；
Insulator breakdown identification module 2, it is right for the background image to be inputted preset insulator breakdown identification model
The background image carries out insulator breakdown identification, identifies the faulty insulator in the background image.
Compared with the existing technology, provided in an embodiment of the present invention a kind of based on the pyramidal insulator breakdown identification side of feature
The beneficial effect of method is: described to be based on the pyramidal insulator breakdown recognition methods of feature, comprising: acquisition contains insulator
Background image；The background image is inputted into preset insulator breakdown identification model, insulator is carried out to the background image
Fault identification identifies the faulty insulator in the background image.This method identifies mould by constructing preset insulator breakdown
Type can fast and effeciently identify the failure of insulator, improve fault identification efficiency, it is ensured that the stable operation of electric system.
It is the preferred embodiment of the present invention described in upper, it is noted that those skilled in the art are come
It says, various improvements and modifications may be made without departing from the principle of the present invention, these improvements and modifications are also considered as this
The protection scope of invention.
Claims (10)
1. one kind is based on the pyramidal insulator breakdown recognition methods of feature characterized by comprising
Obtain the background image containing insulator；
The background image is inputted into preset insulator breakdown identification model, insulator breakdown knowledge is carried out to the background image
Not, the faulty insulator in the background image is identified.
2. being based on the pyramidal insulator breakdown recognition methods of feature as described in claim 1, which is characterized in that the insulation
Sub- fault identification model, is constructed by following steps:
It acquires the background image that multiple contain insulator to be pre-processed, obtains image data set；
Convolutional neural networks model is constructed, and feature letter is carried out to described image data set by the convolutional neural networks model
Breath extracts, and obtains the insulator characteristic pattern of multiple different scales；
According to the insulator characteristic pattern construction feature pyramid network, and to each of feature pyramid network gold
Word tower level carries out candidate region division, obtains several candidate frames；
According to the candidate frame and the insulator characteristic pattern, position sensing score chart is constructed, the position sensing score is obtained
Scheme corresponding category score probability, and the corresponding class label of candidate frame according to the category score determine the probability；
Pass through the probability value of each class label in background image described in preset classifier calculated；
Using the candidate frame, the insulator characteristic pattern and the position sensing score chart as training sample, and according to batch
Gradient descent method is trained the preset classifier, obtains the insulator breakdown identification model for completing training.
3. being based on the pyramidal insulator breakdown recognition methods of feature as claimed in claim 2, which is characterized in that the acquisition
Multiple background images for containing insulator are pre-processed, and are obtained image data set, are specifically included:
Amplification processing is carried out to multiple described background images for containing insulator；Wherein the mode of the amplification processing includes to figure
As being rotated, being cut out, saturation degree is adjusted, tone is adjusted and the time for exposure is adjusted；
Image category mark is carried out to amplification treated background image, and place is normalized to the background image after mark
Reason, obtains image data set.
4. being based on the pyramidal insulator breakdown recognition methods of feature as claimed in claim 3, which is characterized in that the convolution
Neural network model includes 5 convolution group modules and 1 average pond layer.
5. being based on the pyramidal insulator breakdown recognition methods of feature as claimed in claim 4, which is characterized in that the basis
The insulator characteristic pattern construction feature pyramid network, and to each of feature pyramid network pyramid level
Candidate region division is carried out, several candidate frames is obtained, specifically includes:
Using top-down path, the insulator obtained according to 4 convolution group modules rear in the convolutional neural networks model
Characteristic pattern construction feature pyramid；
Candidate region is carried out to each of feature pyramid pyramidal layer grade according to preset candidate region network to draw
Point.
6. being based on the pyramidal insulator breakdown recognition methods of feature as claimed in claim 5, which is characterized in that the basis
The candidate frame and the insulator characteristic pattern construct position sensing score chart, it is corresponding to obtain the position sensing score chart
Category score probability, and the corresponding class label of candidate frame according to the category score determine the probability, specifically include:
Increase default convolutional layer in the convolutional neural networks model, generates several position sensing score charts；
Obtain the category score probability of any position sensitivity score chart；
Using the maximum category score probability of numerical value as the corresponding class label of the candidate frame.
7. being based on the pyramidal insulator breakdown recognition methods of feature as claimed in claim 6, which is characterized in that described to pass through
The probability value of each class label in background image described in preset classifier calculated, specifically includes:
According to formulaPreset classifier is constructed, and passes through the classifier
Calculate the probability value of each class label in background image；
Wherein, rc(i, j/ θ)=∑(x, y) ∈ bin (i, j)zI, j, c(x+x0, y+y0/ θ)/n be c-th of classification (i, j) subregion
Pondization response；C is foreground object classification number in background image；zI, j, cFor one in several described position sensing score charts
Position sensing score chart；N indicates the number of pixels in each sub-regions；6 ginseng to learn in convolutional neural networks model
Number；(x0, y0) it is the position sensing point that the c+1 dimension that pondization ballot obtains is carried out after summing up to the candidate frame averagely
The centre coordinate in the number figure upper left corner.
8. being based on the pyramidal insulator breakdown recognition methods of feature as claimed in claim 7, which is characterized in that by the time
Select frame, the insulator characteristic pattern and the position sensing score chart as training sample, and according to batch gradient descent method pair
The preset classifier is trained, and is obtained the insulator breakdown identification model for completing training, is specifically included:
According to formulaObtain the classifier
Loss function；
Wherein,To intersect loss function, Sc(θ) is
The prediction label of candidate frame,For true tag, and its label value is only 0 or 1, and works asWhen being 0, table
Show that the candidate frame is background class label；For boundary recurrence
Loss function,T=(tx, ty, tw, th) sat for the position of the candidate frame of prediction
Mark information, t*=(t* x, t* y, t* w, t* h) be true frame corresponding with candidate frame location coordinate information, x, y, w, h distinguish table
Show frame centre coordinate, width and height；λ is to return weight coefficient, and it is initialized as 10；
According to formulaTo the damage
Function is lost to optimize；Wherein, NclsFor batch of quantity every in the batch gradient descent method, NregIndicate the number of candidate frame
Amount；
When loss function output valve after optimization meets preset condition, stopping is trained the classifier, is completed
Trained insulator breakdown identification model.
9. being based on the pyramidal insulator breakdown recognition methods of feature as claimed in claim 8, which is characterized in that described default
Condition is less than or equal to 0.0001.
10. one kind is based on the pyramidal insulator breakdown identification device of feature characterized by comprising
Image collection module, for obtaining the background image containing insulator；
Insulator breakdown identification module, for the background image to be inputted preset insulator breakdown identification model, to described
Background image carries out insulator breakdown identification, identifies the faulty insulator in the background image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910283167.6A CN110136097A (en) | 2019-04-10 | 2019-04-10 | One kind being based on the pyramidal insulator breakdown recognition methods of feature and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910283167.6A CN110136097A (en) | 2019-04-10 | 2019-04-10 | One kind being based on the pyramidal insulator breakdown recognition methods of feature and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110136097A true CN110136097A (en) | 2019-08-16 |
Family
ID=67569541
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910283167.6A Pending CN110136097A (en) | 2019-04-10 | 2019-04-10 | One kind being based on the pyramidal insulator breakdown recognition methods of feature and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110136097A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112651954A (en) * | 2020-12-30 | 2021-04-13 | 广东电网有限责任公司电力科学研究院 | Method and device for detecting insulator string dropping area |
| CN113363870A (en) * | 2021-06-30 | 2021-09-07 | 东南大学 | Intelligent operation and maintenance method for insulator based on autonomous controllable chip |
| CN114359739A (en) * | 2022-03-18 | 2022-04-15 | 深圳市海清视讯科技有限公司 | Target identification method and device |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107392901A (en) * | 2017-07-24 | 2017-11-24 | 国网山东省电力公司信息通信公司 | A kind of method for transmission line part intelligence automatic identification |
| CN108038409A (en) * | 2017-10-27 | 2018-05-15 | 江西高创保安服务技术有限公司 | A kind of pedestrian detection method |
| CN108596886A (en) * | 2018-04-17 | 2018-09-28 | 福州大学 | Aerial Images insulator based on deep learning falls piece fault rapid detecting method |
| CN108734143A (en) * | 2018-05-28 | 2018-11-02 | 江苏迪伦智能科技有限公司 | A kind of transmission line of electricity online test method based on binocular vision of crusing robot |
| CN109166094A (en) * | 2018-07-11 | 2019-01-08 | 华南理工大学 | A kind of insulator breakdown positioning identifying method based on deep learning |
-
2019
- 2019-04-10 CN CN201910283167.6A patent/CN110136097A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107392901A (en) * | 2017-07-24 | 2017-11-24 | 国网山东省电力公司信息通信公司 | A kind of method for transmission line part intelligence automatic identification |
| CN108038409A (en) * | 2017-10-27 | 2018-05-15 | 江西高创保安服务技术有限公司 | A kind of pedestrian detection method |
| CN108596886A (en) * | 2018-04-17 | 2018-09-28 | 福州大学 | Aerial Images insulator based on deep learning falls piece fault rapid detecting method |
| CN108734143A (en) * | 2018-05-28 | 2018-11-02 | 江苏迪伦智能科技有限公司 | A kind of transmission line of electricity online test method based on binocular vision of crusing robot |
| CN109166094A (en) * | 2018-07-11 | 2019-01-08 | 华南理工大学 | A kind of insulator breakdown positioning identifying method based on deep learning |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112651954A (en) * | 2020-12-30 | 2021-04-13 | 广东电网有限责任公司电力科学研究院 | Method and device for detecting insulator string dropping area |
| CN113363870A (en) * | 2021-06-30 | 2021-09-07 | 东南大学 | Intelligent operation and maintenance method for insulator based on autonomous controllable chip |
| CN114359739A (en) * | 2022-03-18 | 2022-04-15 | 深圳市海清视讯科技有限公司 | Target identification method and device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111353413B (en) | Low-missing-report-rate defect identification method for power transmission equipment | |
| CN108037770B (en) | Unmanned aerial vehicle power transmission line inspection system and method based on artificial intelligence | |
| CN110378909B (en) | Single wood segmentation method for laser point cloud based on Faster R-CNN | |
| CN107392901A (en) | A kind of method for transmission line part intelligence automatic identification | |
| CN110018524B (en) | X-ray security inspection contraband identification method based on vision-attribute | |
| CN109166094A (en) | A kind of insulator breakdown positioning identifying method based on deep learning | |
| CN109614985A (en) | A kind of object detection method based on intensive connection features pyramid network | |
| CN109214308A (en) | A kind of traffic abnormity image identification method based on focal loss function | |
| CN110136097A (en) | One kind being based on the pyramidal insulator breakdown recognition methods of feature and device | |
| CN108734143A (en) | A kind of transmission line of electricity online test method based on binocular vision of crusing robot | |
| CN108038846A (en) | Transmission line equipment image defect detection method and system based on multilayer convolutional neural networks | |
| CN108985135A (en) | A kind of human-face detector training method, device and electronic equipment | |
| CN111881730A (en) | Wearing detection method for on-site safety helmet of thermal power plant | |
| CN110826514A (en) | Construction site violation intelligent identification method based on deep learning | |
| CN106356757A (en) | Method for inspecting electric power lines by aid of unmanned aerial vehicle on basis of human vision characteristics | |
| CN110070530A (en) | A kind of powerline ice-covering detection method based on deep neural network | |
| CN108509976A (en) | The identification device and method of animal | |
| CN110472597A (en) | Rock image rate of decay detection method and system based on deep learning | |
| CN110414400B (en) | Automatic detection method and system for wearing of safety helmet on construction site | |
| CN104992223A (en) | Dense population estimation method based on deep learning | |
| CN113240688A (en) | Integrated flood disaster accurate monitoring and early warning method | |
| CN107833209A (en) | A kind of x-ray image detection method, device, electronic equipment and storage medium | |
| CN108154072A (en) | Insulator breakdown of taking photo by plane based on depth convolutional neural networks detects automatically | |
| CN112950634B (en) | Unmanned aerial vehicle inspection-based wind turbine blade damage identification method, equipment and system | |
| CN111652835A (en) | Method for detecting insulator loss of power transmission line based on deep learning and clustering |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190816 |
|
| RJ01 | Rejection of invention patent application after publication |