CN108830311A - A kind of infrared image photovoltaic array recognition methods based on Adaboost - Google Patents
A kind of infrared image photovoltaic array recognition methods based on Adaboost Download PDFInfo
- Publication number
- CN108830311A CN108830311A CN201810555978.2A CN201810555978A CN108830311A CN 108830311 A CN108830311 A CN 108830311A CN 201810555978 A CN201810555978 A CN 201810555978A CN 108830311 A CN108830311 A CN 108830311A
- Authority
- CN
- China
- Prior art keywords
- sample
- training
- classifier
- photovoltaic array
- value
- 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
- 238000000034 method Methods 0.000 title claims abstract description 51
- 238000012549 training Methods 0.000 claims abstract description 73
- 238000001514 detection method Methods 0.000 claims abstract description 35
- 230000004069 differentiation Effects 0.000 claims description 3
- 239000000284 extract Substances 0.000 claims description 3
- 238000005259 measurement Methods 0.000 claims description 3
- 238000012795 verification Methods 0.000 claims description 3
- 238000012986 modification Methods 0.000 claims description 2
- 230000004048 modification Effects 0.000 claims description 2
- 238000013507 mapping Methods 0.000 claims 1
- 241000196324 Embryophyta Species 0.000 description 6
- 230000000694 effects Effects 0.000 description 5
- 239000011148 porous material Substances 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 238000012216 screening Methods 0.000 description 3
- 230000001174 ascending effect Effects 0.000 description 2
- 238000010191 image analysis Methods 0.000 description 2
- 238000003331 infrared imaging Methods 0.000 description 2
- 238000012423 maintenance Methods 0.000 description 2
- 238000010248 power generation Methods 0.000 description 2
- 239000013589 supplement Substances 0.000 description 2
- 241000208340 Araliaceae Species 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 206010037660 Pyrexia Diseases 0.000 description 1
- 238000009825 accumulation Methods 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000000903 blocking effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 210000003850 cellular structure Anatomy 0.000 description 1
- 230000006378 damage Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 239000000428 dust Substances 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 210000003608 fece Anatomy 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 239000000047 product Substances 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 238000013083 solar photovoltaic technology Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/211—Selection of the most significant subset of features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/285—Selection of pattern recognition techniques, e.g. of classifiers in a multi-classifier system
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The infrared image photovoltaic array recognition methods based on Adaboost that the invention discloses a kind of, including preparing training required positive sample and negative sample；Using LBP feature training Weak Classifier, Weak Classifier training strong classifier is utilized；Utilize obtained multiple strong classifiers training cascade classifier；Picture to be detected is tentatively identified using cascade classifier；Background process is carried out to preliminary recognition result, obtains the final result of photovoltaic array identification.The present invention improves the accuracy rate of array identification by carrying out photovoltaic array Preliminary detection using Adaboost and combining using the method for removing background based on temperature characterisitic；Avoiding occur when ambient temperature is higher than array according to the method that temperature characterisitic identifies array is misrecognition phenomenon that array, array are identified as background by Background Recognition.
Description
Technical field
The infrared image photovoltaic array recognition methods based on Adaboost that the present invention relates to a kind of, belongs in photovoltaic plant
Array detects automatically and analysis technical field.
Background technique
In recent years, it is constantly progressive as the worldwide petrochemical energy becomes closer to exhausted and solar photovoltaic technology, light
Volt power generation enters the stage of rapid development.China was since 2013, continuous 4 years global photovoltaic industries of taking the lead in race, wherein concentrating
Formula photovoltaic plant accounts for 85% or more.One large-sized photovoltaic power station is generally made of the photovoltaic array of many a queueing disciplines, often
A array is made of the regularly arranged battery of certain amount again.These photovoltaic plants are in for a long time in outdoor environment, very
It is easy blocking by dust, weeds, birds droppings, building and other photovoltaic solar plates, and then leads to the generation of hot spot effect.Heat
Spot will lead to a large amount of fevers in photovoltaic module part, can not convert light energy into electric energy, so not only strong influence power generation effect
Rate, or even the problem of photovoltaic array causes fire can be burnt.Therefore, detection hot spot has photovoltaic plant in daily maintenance in time
There is important meaning.
The method of array hot spot detection at present is broadly divided into two major classes：One kind is to constitute circuit based on photovoltaic cell component
C-V characteristic, another kind of is the temperature characterisitic based on infrared imaging.First kind needs build physical circuit outside photovoltaic array, lead to
It crosses monitoring C-V characteristic and judges whether there is hot spot.This method not only increases the maintenance cost of photovoltaic plant and is building electricity
The physical structure of solar battery itself can be destroyed when road.Second class detects hot spot based on the temperature characterisitic of infrared imaging,
It relies primarily on unmanned plane and carries infrared camera Direct Acquisition temperature data, and detect ground automatically by the method for infrared image analysis
The hot spot in face, accuracy is higher and will not generate destruction to photovoltaic array internal structure.Currently based on the heat of infrared image analysis
Spot detection method is generally divided into two steps：The first step identifies each photovoltaic array in infrared array image；Second step is based on infrared
Image photovoltaic array recognition result detects hot spot.The first step plays a crucial role in the method, tional identification array
It is using temperature data Direct Recognition array, occurring in ambient temperature and close array temperature is battle array by Background Recognition
Column, array are identified as the misrecognition phenomenon of background, once array identification is when the error occurs, second step hot spot detection effect
It can have a greatly reduced quality, so the accuracy rate for improving array identification has great importance for improving hot spot detection effect.
Summary of the invention
In order to solve the above-mentioned technical problems, the present invention provides a kind of, and the infrared image photovoltaic array based on Adaboost is known
Other method.
In order to achieve the above object, the technical scheme adopted by the invention is that：
A kind of infrared image photovoltaic array recognition methods based on Adaboost, includes the following steps,
Prepare training required positive sample and negative sample, wherein negative sample includes incomplete photovoltaic array picture and non-light
Photovoltaic array picture, positive sample include complete photovoltaic array picture；
Using LBP feature training Weak Classifier, Weak Classifier training strong classifier is utilized；
Utilize obtained multiple strong classifiers training cascade classifier；
Picture to be detected is tentatively identified using cascade classifier；
Background process is carried out to preliminary recognition result, obtains the final result of photovoltaic array identification.
Positive sample and negative sample prepare according to certain ratio, incomplete photovoltaic array picture and non-photovoltaic battle array in negative sample
Column picture meets certain ratio.
Training Weak Classifier process be,
Calculate positive sample and negative sample integrogram；
The LBP value for calculating each pixel obtains the LBP characteristic value for reflecting the pixel adjacent domain texture；
All LBP characteristic values are put into a feature set；
Construct Multiway Tree Structure LBP feature Weak Classifier, each branch correspond to some determine LBP characteristic value, weak point
Class device definition is,
Wherein, fi(x) characteristic value that the corresponding LBP of each branch x is determined, f are indicatedi(x) value is bigger, is photovoltaic array
Possibility is bigger, xkIt is k-th of component of vector x, i is the ith feature of window to be detected, and point of Weak Classifier is indicated with n
Branch, n are integer, and the value range of n is 0~L, anIndicate the parameter obtained by training,
Wherein, N is the sum of training sample, wjFor the weight of j-th of training sample, yjFor j-th of training sample classification mark
Label,For k-th of component of j-th of training sample,When equal with n,It is 1, is otherwise 0；
Work as anWhen > 0, illustrates that a possibility that j-th of training sample is positive sample is greater than negative sample, threshold θ is set；
Work as fi(x) >=θ when, be judged as positive sample, be otherwise negative sample.
Threshold θ is the value that a dynamic updates, and specific setting up procedure is,
The weight of all negative samples is set and for Q；
The LBP characteristic value of selected j-th of training sample sets all positive samples and negative sample judged as threshold value
Weight and respectively S+And S-；
Calculate ej=S++(Q-S-)
All samples are traversed, e is obtained1,...,eN；
Calculate e=min { e1,...,eN, the corresponding experienced sample LBP characteristic value of e is optimal threshold.
Using Weak Classifier training strong classifier process be,
Training sample set S is defined, wherein positive sample and negative sample number are respectively numPos and numNeg, the weight of positive sample
Initial value is 1/ (2 × numPos), and the weight initial value of negative sample is 1/ (2 × numNeg)；
T iteration is carried out, each process is：The corresponding Weak Classifier of each feature is constructed first, and it is flat to select wherein weight
Square error andThe smallest Weak Classifier ht(x)；Then according to formulaWeight is updated, wherein wt+1,j,wt,jPower when respectively j-th training sample t+1 times and t iteration
Weight, β is according to formula β=Jwse/(1-Jwse) undated parameter being calculated, e 'j=1 indicates that training sample classification is wrong
Accidentally, e 'j=0 indicates that training sample classification is correct；Finally positive sample and negative sample weight are normalized；
The classification error rate of each weak typing is its weight αt, obtaining strong classifier is,
Wherein,
Training cascade classifier process be,
1) number for defining strong classifier is M, i.e., the maximum number of plies of cascade classifier is M, and training each of obtains strong point
The minimum detection rate of class device and maximum false detection rate are respectively dminAnd fmax, select a positive sample of numPos ' and a negative sample of numNeg '
This, the ratio between negative sample and positive sample are
Rario=numPos '/numNeg ', the initial false detection rate of cascade classifier are F0=1；
2) training：
21) j '=1 is defined；
22) training jth ' a strong classifier, makes its verification and measurement ratio dj′Greater than dmin, while false detection rate fj′Less than fmax；
23)Fj′=fj′*Fj′-1, judge Fj′≤Fj′-1It is whether true, terminate if setting up, otherwise judges whether j ' is equal to M,
Terminate if being equal to, otherwise j '=j '+1；
24) positive sample is verified, the positive sample number by cascade classifier is M ', positive sample in modification next round training
M ' is assigned to numPos ' by this number, is verified to negative sample, and the negative sample of correct rejection is removed, and supplements negative sample,
Make negative sample number numNeg '=numPos '/rario, goes to step 22；
3) obtained cascade classifier is saved.
During tentatively identifying picture to be detected, needs first to scale detection window and picture to be detected, make detection window
It is identical with the size of picture to be detected, then it is screened by cascade classifier.
The background process is gone to be,
Foreground and background differentiation is carried out to preliminary recognition result, extracts foreground target；
Calculate foreground target temperature mean value, when temperature mean value be higher than temperature threshold, then be determined as photovoltaic array, otherwise for
Shade；
The pixel value for being determined as photovoltaic array region is set as 1, shadow region pixel value is set as 0；
The region for being 1 to pixel value calculates local variance, local variance figure is obtained, using Otsu algorithm to local variogram
Thresholding has obtained the binary image of removal shade and background；
Screen temperature distribution range；
Using opening operation, background interference in binary image is eliminated, disconnects the connection between photovoltaic module；
The connected region of photovoltaic module is found, and the number of altitude is carried out to photovoltaic module, to the photovoltaic for being less than average height
Component is marked with average height.
The beneficial effects obtained by the present invention are as follows：1, the present invention carries out photovoltaic array using the Adaboost algorithm of LBP feature
Preliminary identification, improves the accuracy of photovoltaic array coarse positioning；2, the present invention is at the beginning of by carrying out photovoltaic array using Adaboost
Step is detected and is combined using the method based on temperature characterisitic removal background, and the accuracy rate of array identification is improved；Avoid root
What is occurred when ambient temperature is higher than array according to the method for temperature characterisitic identification array is that array, array are identified as by Background Recognition
The misrecognition phenomenon of background；3, the present invention only carries out the processing work of temperature data to the region that Preliminary detection is array rather than right
The temperature data of entire image is handled, and reduces calculation amount in this way, improves efficiency.
Detailed description of the invention
Fig. 1 is flow chart of the invention；
Fig. 2 is the flow chart of training cascade classifier；
Fig. 3 is the flow chart tentatively identified.
Specific embodiment
The invention will be further described below in conjunction with the accompanying drawings.Following embodiment is only used for clearly illustrating the present invention
Technical solution, and not intended to limit the protection scope of the present invention.
As shown in Figure 1, a kind of infrared image photovoltaic array recognition methods based on Adaboost, includes the following steps：
Step 1, prepare training required positive sample and negative sample.
Negative sample includes incomplete photovoltaic array picture and non-photovoltaic array picture, and positive sample includes complete photovoltaic array figure
Piece；Positive sample and negative sample prepare according to certain ratio, incomplete photovoltaic array picture and non-photovoltaic array figure in negative sample
Piece meets certain ratio.
Aforementioned proportion can be adjusted according to different situations, and in operation, the ratio of positive sample and negative sample is 1:3~1:5,
The ratio of incomplete photovoltaic array picture and non-photovoltaic array picture is 1:3~1:4.Prepare 802 positive samples, 3530 negative samples
This, wherein 850 incomplete photovoltaic array pictures and 2680 non-photovoltaic array pictures, all dimension of pictures are normalized to grow
The size of wide 60 pixel of 160 pixels, all pictures are jpg format.
Step 2, all it is the attribute of regular geometric pattern according to photovoltaic array, using LBP feature training Weak Classifier, utilizes
Weak Classifier trains strong classifier, specific as shown in Figure 2.
The process of training Weak Classifier is as follows：
201) positive sample and negative sample integrogram are calculated.
202) the LBP value for calculating each pixel, obtains the LBP characteristic value for reflecting the pixel adjacent domain texture；
In 3 × 3 window, using window center pixel LBP value as threshold value, the pixel LBP value and threshold in 8 fields of surrounding
Value compares, and is 1, otherwise as 0 if pixel LBP value is greater than threshold value, just has 8 unsigned numbers in such window, can be with
The pixel LBP value at center is indicated with 8 bits, this 8 bit is referred to as LBP characteristic value, this LBP characteristic value
It can be used to reflect the texture information in 3 × 3 regions.
203) all LBP characteristic values are put into a feature set, are used during training Weak Classifier.
204) the LBP feature Weak Classifier of Multiway Tree Structure is constructed, each branch corresponds to the LBP characteristic value of some determination,
Weak Classifier definition is：
Wherein, fi(x) characteristic value that the corresponding LBP of each branch x is determined, f are indicatedi(x) value is bigger, is photovoltaic array
Possibility is bigger, xkIt is k-th of component of vector x, i is the ith feature of window to be detected, and point of Weak Classifier is indicated with n
Branch, n are integer, and the value range of n is 0~L, and L+1 is branch's number, generally 256, anIndicate the ginseng obtained by training
Number, an∈ { 1, -1 }, training formula are：
Wherein, N is the sum of training sample, wjFor the weight of j-th of training sample, yjFor j-th of training sample classification mark
Label,For k-th of component of j-th of training sample,When equal with n,It is 1, is otherwise 0.
205) work as anWhen > 0, illustrates that a possibility that j-th of training sample is positive sample is greater than negative sample, threshold θ is set；
Work as fi(x) >=θ when, be judged as positive sample, be otherwise negative sample.
fi(x) a possibility that value is bigger, is photovoltaic array is also higher, its value is carried out ascending order arrangement；Above-mentioned threshold θ is simultaneously
A non-fixed value, but the value that a dynamic updates, specific setting up procedure are as follows：
211) weight of all negative samples is set and it is Q.
212) the LBP characteristic value of j-th of training sample is selected as threshold value, sets all positive samples judged and negative sample
This weight and respectively S+And S-；
213) e is calculatedj=S++(Q-S-)
214) all samples are traversed, e is obtained1,...,eN；
215) e=min { e is calculated1,...,eN, the corresponding experienced sample LBP characteristic value of e is optimal threshold.
The process of single Weak Classifier is subjected to T iteration, by constantly increasing the weight of error sample, obtains T points
The best Weak Classifier of class ability forms Weak Classifier collection, then by adjusting the weight of each Weak Classifier, is changing weight
When simultaneously positive sample and negative sample are normalized, this makes it possible to obtain a strong classifier.Specific training is strong
The process of classifier is as follows：
221) training sample set S={ (x is defined1,y1)(x2,y3)...(xN,yN), xjFor j-th of training sample, work as yj=
When 1, x is indicatedjFor positive sample, work as yjWhen=- 1, x is indicatedjFor negative sample, in S positive sample and negative sample number be respectively numPos and
NumNeg, the weight initial value of positive sample are 1/ (2 × numPos), and the weight initial value of negative sample is 1/ (2 × numNeg).
222) T iteration is carried out, each process is：The corresponding Weak Classifier of each feature is constructed first, is selected and is wherein weighed
Be worth square error andThe smallest Weak Classifier ht(x)；Then basis
FormulaWeight is updated, wherein wt+1,j,wt,jWhen respectively j-th training sample t+1 times and t iteration
Weight, β is according to formula β=Jwse/(1-Jwse) undated parameter being calculated, e 'j=1 indicates training sample classification
Mistake, e 'j=0 indicates that training sample classification is correct；Finally positive sample and negative sample weight are normalized.
223) the classification error rate of each weak typing is its weight αt, obtaining strong classifier is,
Wherein,
Step 3, obtained multiple strong classifiers training cascade classifier is utilized.
Strong classifier has been obtained by previous step is trained, but in order to improve detection rates, reduces false detection rate,
Continue to train cascade classifier on the basis of strong classifier.Cascade classifier is exactly the sieve that sieve pore is ascending from level to level, often
One layer of effect is the negative sample that can be weeded out upper layer sieve pore and leave, and can just be determined as photovoltaic array by all sieve pores.
Detailed process is as follows for training cascade classifier：
31) number for defining strong classifier is M, i.e., the maximum number of plies of cascade classifier is M, and training each of obtains strong point
The minimum detection rate of class device and maximum false detection rate are respectively dminAnd fmax, the minimum detection rate D of entire cascade classifierminIt is M
Strong classifier dminProduct, maximum false detection rate FmaxIt is exactly the maximum false detection rate f that classifies M strongmaxAccumulation, it is a just to select numPos '
Sample and a negative sample of numNeg ', the ratio between negative sample and positive sample are rario=numPos '/numNeg ', cascade classifier
Initial false detection rate is F0=1.
32) training；
321) j '=1 is defined；
322) training jth ' a strong classifier, makes its verification and measurement ratio dj′Greater than dmin, while false detection rate fj′Less than fmax；
323)Fj′=fj′*Fj′-1, judge Fj′≤Fj′-1It is whether true, terminate if setting up, otherwise judges whether j ' is equal to
M terminates, otherwise j '=j '+1 if being equal to；
324) positive sample is verified, the positive sample number by cascade classifier is M ', is modified in next round training just
M ' is assigned to numPos ' by number of samples, is verified to negative sample, and the negative sample of correct rejection is removed, and supplements negative sample
This, makes negative sample number numNeg '=numPos '/rario, goes to step 22.
33) obtained cascade classifier is saved.
Step 4, picture to be detected is tentatively identified using cascade classifier.
As shown in figure 3, tentatively identifying that the process of picture to be detected is as follows：
41) detection window and picture to be detected are scaled, keeps detection window identical with the size of picture to be detected.
The size that will test window first is initialized as the size of positive sample, then according to 1.1 times of size to detection window
Mouth amplifies；Then picture to be detected is subjected to scaled down, i.e., original picture to be detected is reduced 1.1 times every time；
Aforesaid operations are repeated until the size of detection window is identical with the size of picture to be detected.
42) load cascade classifier carries out target detection, and detection window is needed by cascade classifier, in this process
Constantly screening removes some non-photovoltaic array pictures and incomplete photovoltaic array picture, until passing through all layers of screening ability
It is considered as complete photovoltaic array picture.
Step 5, background process is carried out to preliminary recognition result, obtains the final result of photovoltaic array identification.
Go background process as follows：
51) gradient is carried out to preliminary recognition result and gradient image is calculated, then carry out thresholding using Otsu algorithm
Processing；Foreground and background differentiation can be carried out by above-mentioned processing, extracts foreground target.
52) calculate foreground target temperature mean value, when temperature mean value be higher than temperature threshold, then be determined as photovoltaic array, it is no
It is then shade.
53) pixel value for being determined as photovoltaic array region is set as 1, shadow region pixel value is set as 0.
54) region for being 1 to pixel value calculates the local variance in 2 × 2 regions, obtains local variance figure, utilizes Otsu
Algorithm has obtained the binary image of removal shade and background to local variogram thresholding.
55) temperature distribution range is screened.
It is similar to the priori knowledge of Gaussian Profile, the temperature of array region according to array temperature distribution in preliminary recognition result
Distribution should be averaged near peak value max it is contemplated that the temperature for being likely to occur hot spot on photovoltaic array is apparently higher than array
The case where temperature, so selecting [max-4, max+12] when screening temperature distribution range.
56) opening operation is utilized, background interference in binary image is eliminated, disconnects the connection between photovoltaic module.
57) connected region of photovoltaic module is found (if pixel a, in the eight neighborhood of pixel b, a is to be connected to b.From
From the point of view of visually, the point to communicate with each other forms a region, and disconnected point forms different regions, such a institute
The set for thering is the point to communicate with each other to constitute, referred to as a connected region.Because the operation before passing through has been removed background, leave
It is photovoltaic module, so detecting connected region i.e. photovoltaic module connected region.), and the number of altitude is carried out to photovoltaic module,
The photovoltaic module for being less than average height is marked with average height.
It is above-mentioned to put by carrying out photovoltaic array Preliminary detection using Adaboost and removing background using based on temperature characterisitic
Method combine, improve array identification accuracy rate；It avoids and identifies the method for array in background temperature according to temperature characterisitic
It is misrecognition phenomenon that array, array are identified as background by Background Recognition that degree, which is higher than occur when array,.
The above is only a preferred embodiment of the present invention, it is noted that for the ordinary skill people of the art
For member, without departing from the technical principles of the invention, several improvement and deformations can also be made, these improvement and deformations
Also it should be regarded as protection scope of the present invention.
Claims (8)
1. a kind of infrared image photovoltaic array recognition methods based on Adaboost, it is characterised in that：Include the following steps,
Prepare training required positive sample and negative sample, wherein negative sample includes incomplete photovoltaic array picture and non-photovoltaic battle array
Column picture, positive sample include complete photovoltaic array picture；
Using LBP feature training Weak Classifier, Weak Classifier training strong classifier is utilized；
Utilize obtained multiple strong classifiers training cascade classifier；
Picture to be detected is tentatively identified using cascade classifier；
Background process is carried out to preliminary recognition result, obtains the final result of photovoltaic array identification.
2. a kind of infrared image photovoltaic array recognition methods based on Adaboost according to claim 1, feature exist
In：Positive sample and negative sample prepare according to certain ratio, incomplete photovoltaic array picture and non-photovoltaic array figure in negative sample
Piece meets certain ratio.
3. a kind of infrared image photovoltaic array recognition methods based on Adaboost according to claim 1, feature exist
In：Training Weak Classifier process be,
Calculate positive sample and negative sample integrogram；
The LBP value for calculating each pixel obtains the LBP characteristic value for reflecting the pixel adjacent domain texture；
All LBP characteristic values are put into a feature set；
The LBP feature Weak Classifier of Multiway Tree Structure is constructed, each branch corresponds to the LBP characteristic value of some determination, Weak Classifier
Definition is,
Wherein, fi(x) characteristic value that the corresponding LBP of each branch x is determined, f are indicatedi(x) value is bigger, is the possibility of photovoltaic array
Bigger, the x of propertykIt is k-th of component of vector x, i is the ith feature of window to be detected, and the branch of Weak Classifier, n are indicated with n
For integer, the value range of n is 0~L, anIndicate the parameter obtained by training,
Wherein, N is the sum of training sample, wjFor the weight of j-th of training sample, yjFor j-th of training sample class label,For k-th of component of j-th of training sample,When equal with n,It is 1, is otherwise 0；
Work as anWhen > 0, illustrates that a possibility that j-th of training sample is positive sample is greater than negative sample, threshold θ is set；
Work as fi(x) >=θ when, be judged as positive sample, be otherwise negative sample.
4. a kind of infrared image photovoltaic array recognition methods based on Adaboost according to claim 3, feature exist
In：Threshold θ is the value that a dynamic updates, and specific setting up procedure is,
The weight of all negative samples is set and for Q；
The LBP characteristic value of selected j-th of training sample sets the weight of all positive samples and negative sample judged as threshold value
Respectively S+And S-；
Calculate ej=S++(Q-S-)
All samples are traversed, e is obtained1,...,eN；
Calculate e=min { e1,...,eN, the corresponding experienced sample LBP characteristic value of e is optimal threshold.
5. a kind of infrared image photovoltaic array recognition methods based on Adaboost according to claim 1, feature exist
In：Using Weak Classifier training strong classifier process be,
Training sample set S is defined, wherein positive sample and negative sample number are respectively numPos and numNeg, the weight initial value of positive sample
For 1/ (2 × numPos), the weight initial value of negative sample is 1/ (2 × numNeg)；
T iteration is carried out, each process is：The corresponding Weak Classifier of each feature is constructed first, selects wherein weight square mistake
Difference andThe smallest Weak Classifier ht(x)；Then according to formulaWeight is updated, wherein wt+1,j,wt,jPower when respectively j-th training sample t+1 times and t iteration
Weight, β is according to formula β=Jwse/(1-Jwse) undated parameter being calculated, e 'j=1 indicates that training sample classification is wrong
Accidentally, e 'j=0 indicates that training sample classification is correct；Finally positive sample and negative sample weight are normalized；
The classification error rate of each weak typing is its weight αt, obtaining strong classifier is,
Wherein,
6. a kind of infrared image photovoltaic array recognition methods based on Adaboost according to claim 1, feature exist
In：Training cascade classifier process be,
1) number for defining strong classifier is M, i.e., the maximum number of plies of cascade classifier is M, each strong classifier that training obtains
Minimum detection rate and maximum false detection rate be respectively dminAnd fmax, a positive sample of numPos ' and a negative sample of numNeg ' are selected,
The ratio between negative sample and positive sample are rario=numPos '/numNeg ', and the initial false detection rate of cascade classifier is F0=1；
2) training：
21) j '=1 is defined；
22) training jth ' a strong classifier, makes its verification and measurement ratio dj′Greater than dmin, while false detection rate fj′Less than fmax；
23)Fj′=fj′*Fj′-1, judge Fj′≤Fj′-1It is whether true, terminate if setting up, otherwise judge whether j ' is equal to M, if waiting
In then terminating, otherwise j '=j '+1；
24) positive sample is verified, the positive sample number by cascade classifier is M ', positive sample number in modification next round training
M ' is assigned to numPos ' by mesh, is verified to negative sample, and the negative sample of correct rejection is removed, and is supplemented negative sample, is made to bear
Sample number numNeg '=numPos '/rario, goes to step 22；
3) obtained cascade classifier is saved.
7. a kind of infrared image photovoltaic array recognition methods based on Adaboost according to claim 1, feature exist
In：It during tentatively identifying picture to be detected, needs first to scale detection window and picture to be detected, makes detection window and to be checked
The size of mapping piece is identical, is then screened by cascade classifier to it.
8. a kind of infrared image photovoltaic array recognition methods based on Adaboost according to claim 1, feature exist
In：The background process is gone to be,
Foreground and background differentiation is carried out to preliminary recognition result, extracts foreground target；
Calculate foreground target temperature mean value, when temperature mean value be higher than temperature threshold, then be determined as photovoltaic array, be otherwise yin
Shadow；
The pixel value for being determined as photovoltaic array region is set as 1, shadow region pixel value is set as 0；
The region for being 1 to pixel value calculates local variance, local variance figure is obtained, using Otsu algorithm to local variogram threshold value
Change, has obtained the binary image of removal shade and background；
Screen temperature distribution range；
Using opening operation, background interference in binary image is eliminated, disconnects the connection between photovoltaic module；
The connected region of photovoltaic module is found, and the number of altitude is carried out to photovoltaic module, to the photovoltaic module for being less than average height
It is marked with average height.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810555978.2A CN108830311A (en) | 2018-06-01 | 2018-06-01 | A kind of infrared image photovoltaic array recognition methods based on Adaboost |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810555978.2A CN108830311A (en) | 2018-06-01 | 2018-06-01 | A kind of infrared image photovoltaic array recognition methods based on Adaboost |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108830311A true CN108830311A (en) | 2018-11-16 |
Family
ID=64146761
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810555978.2A Pending CN108830311A (en) | 2018-06-01 | 2018-06-01 | A kind of infrared image photovoltaic array recognition methods based on Adaboost |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108830311A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109711416A (en) * | 2018-11-23 | 2019-05-03 | 西安天和防务技术股份有限公司 | Target identification method, device, computer equipment and storage medium |
| CN110428440A (en) * | 2019-07-23 | 2019-11-08 | 浙江树人学院（浙江树人大学） | A kind of shadow detection method based on gray variance |
| CN111860531A (en) * | 2020-07-28 | 2020-10-30 | 西安建筑科技大学 | Raise dust pollution identification method based on image processing |
| CN113780359A (en) * | 2021-08-16 | 2021-12-10 | 佛山科学技术学院 | Method and device for identifying insulator in infrared image and readable storage medium |
| WO2022225449A1 (en) * | 2021-04-19 | 2022-10-27 | Quantified Energy Labs Pte. Ltd. | Automated imaging of photovoltaic devices using an aerial vehicle and automated flight of the aerial vehicle for performing the same |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101577033A (en) * | 2009-05-26 | 2009-11-11 | 官洪运 | Multiband infrared image-type fire detecting system and fire alarm system thereof |
| CN102184384A (en) * | 2011-04-18 | 2011-09-14 | 苏州市慧视通讯科技有限公司 | Face identification method based on multiscale local phase quantization characteristics |
| CN106600777A (en) * | 2016-12-09 | 2017-04-26 | 济南赛英立德电子科技有限公司 | Infrared array number-of-personnel sensor-based counting method and apparatus |
| CN106960202A (en) * | 2017-04-11 | 2017-07-18 | 广西师范大学 | A kind of smiling face's recognition methods merged based on visible ray with infrared image |
| CN107036715A (en) * | 2017-03-30 | 2017-08-11 | 智来光电科技（苏州）有限公司 | A kind of infrared image is without catch Nonuniformity Correction device and its bearing calibration |
| CN107316036A (en) * | 2017-06-09 | 2017-11-03 | 广州大学 | A kind of insect recognition methods based on cascade classifier |
-
2018
- 2018-06-01 CN CN201810555978.2A patent/CN108830311A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101577033A (en) * | 2009-05-26 | 2009-11-11 | 官洪运 | Multiband infrared image-type fire detecting system and fire alarm system thereof |
| CN102184384A (en) * | 2011-04-18 | 2011-09-14 | 苏州市慧视通讯科技有限公司 | Face identification method based on multiscale local phase quantization characteristics |
| CN106600777A (en) * | 2016-12-09 | 2017-04-26 | 济南赛英立德电子科技有限公司 | Infrared array number-of-personnel sensor-based counting method and apparatus |
| CN107036715A (en) * | 2017-03-30 | 2017-08-11 | 智来光电科技（苏州）有限公司 | A kind of infrared image is without catch Nonuniformity Correction device and its bearing calibration |
| CN106960202A (en) * | 2017-04-11 | 2017-07-18 | 广西师范大学 | A kind of smiling face's recognition methods merged based on visible ray with infrared image |
| CN107316036A (en) * | 2017-06-09 | 2017-11-03 | 广州大学 | A kind of insect recognition methods based on cascade classifier |
Non-Patent Citations (1)
| Title |
|---|
| 葛开标: "基于改进的LBP特征的AdaBoost算法与肤色检测相结合的人脸检测", 《中国优秀硕士论文全文数据库》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109711416A (en) * | 2018-11-23 | 2019-05-03 | 西安天和防务技术股份有限公司 | Target identification method, device, computer equipment and storage medium |
| CN109711416B (en) * | 2018-11-23 | 2021-08-06 | 西安天和防务技术股份有限公司 | Target identification method and device, computer equipment and storage medium |
| CN110428440A (en) * | 2019-07-23 | 2019-11-08 | 浙江树人学院（浙江树人大学） | A kind of shadow detection method based on gray variance |
| CN111860531A (en) * | 2020-07-28 | 2020-10-30 | 西安建筑科技大学 | Raise dust pollution identification method based on image processing |
| WO2022225449A1 (en) * | 2021-04-19 | 2022-10-27 | Quantified Energy Labs Pte. Ltd. | Automated imaging of photovoltaic devices using an aerial vehicle and automated flight of the aerial vehicle for performing the same |
| CN113780359A (en) * | 2021-08-16 | 2021-12-10 | 佛山科学技术学院 | Method and device for identifying insulator in infrared image and readable storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108830311A (en) | A kind of infrared image photovoltaic array recognition methods based on Adaboost | |
| Mayr et al. | Weakly supervised segmentation of cracks on solar cells using normalized L p norm | |
| CN110070530B (en) | Transmission line icing detection method based on deep neural network | |
| CN111798412B (en) | Intelligent diagnosis method and system for defects of power transformation equipment based on infrared image | |
| CN109359697A (en) | Graph image recognition methods and inspection system used in a kind of power equipment inspection | |
| Rahman et al. | CNN-based deep learning approach for micro-crack detection of solar panels | |
| CN107481237B (en) | A kind of infrared array image hot spot detection method based on multiframe temperature characterisitic | |
| CN112258490A (en) | Low-emissivity coating intelligent damage detection method based on optical and infrared image fusion | |
| Zuo et al. | An insulator defect detection algorithm based on computer vision | |
| CN111539330B (en) | Transformer substation digital display instrument identification method based on double-SVM multi-classifier | |
| CN110766016B (en) | Code-spraying character recognition method based on probabilistic neural network | |
| CN111310756A (en) | Damaged corn particle detection and classification method based on deep learning | |
| CN115115634B (en) | Photovoltaic array hot spot detection method based on infrared image | |
| CN108537170A (en) | A kind of power equipment firmware unmanned plane inspection pin missing detection method | |
| Prabhakaran et al. | Deep Learning-Based Model for Defect Detection and Localization on Photovoltaic Panels. | |
| CN110245592A (en) | A method of for promoting pedestrian's weight discrimination of monitoring scene | |
| CN104573713A (en) | Mutual inductor infrared image recognition method based on image textual features | |
| CN114782442A (en) | Photovoltaic cell panel intelligent inspection method and system based on artificial intelligence | |
| CN109308685A (en) | Infrared photovoltaic array dividing method based on Threshold segmentation and K mean cluster | |
| CN108510487A (en) | A kind of photovoltaic module remote monitoring device | |
| Liu et al. | A CISG method for internal defect detection of solar cells in different production processes | |
| CN114581407B (en) | Self-adaptive defect detection method for photovoltaic module | |
| CN111144280A (en) | Monitoring video leaf occlusion detection method | |
| CN113781448B (en) | Intelligent defect identification method for photovoltaic power station assembly based on infrared image analysis | |
| CN113160236B (en) | Image identification method for shadow shielding of photovoltaic cell |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20181116 |