CN109598349A - Overhead transmission line fault detection data sample batch processing training method based on classification stochastical sampling - Google Patents
Overhead transmission line fault detection data sample batch processing training method based on classification stochastical sampling Download PDFInfo
- Publication number
- CN109598349A CN109598349A CN201811407820.7A CN201811407820A CN109598349A CN 109598349 A CN109598349 A CN 109598349A CN 201811407820 A CN201811407820 A CN 201811407820A CN 109598349 A CN109598349 A CN 109598349A
- Authority
- CN
- China
- Prior art keywords
- classification
- data
- sample
- batch
- data sample
- 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
- 238000012549 training Methods 0.000 title claims abstract description 37
- 238000001514 detection method Methods 0.000 title claims abstract description 33
- 238000012545 processing Methods 0.000 title claims abstract description 24
- 238000000034 method Methods 0.000 title claims abstract description 23
- 230000005540 biological transmission Effects 0.000 title claims abstract description 19
- 238000005070 sampling Methods 0.000 title claims abstract description 13
- 238000013135 deep learning Methods 0.000 abstract description 11
- 230000007547 defect Effects 0.000 description 19
- 238000010586 diagram Methods 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 4
- 238000007689 inspection Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 2
- 230000006399 behavior Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000006798 recombination Effects 0.000 description 1
- 238000005215 recombination Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Hardware Design (AREA)
- Evolutionary Computation (AREA)
- Geometry (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Image Analysis (AREA)
Abstract
The overhead transmission line fault detection data sample batch processing training method based on classification stochastical sampling that the invention discloses a kind of, steps are as follows: S1, unmanned plane acquire data, determine the corresponding classification of each data sample, category stores data, then counts the quantity of the included sample of every one kind;S2, deep learning algorithm model batch processing quantity N is determined according to the quantity that every a kind of sample includesbatchNumerical value;S3, it is to guarantee that each classification sample size is identical in batching data, randomly selects N from every one kindbatch/ n data sample;S4, the data being drawn into are combined into one group of batching data, are trained after upsetting sequence for model batch processing;S5, to the training of this batch of data after, repeat the above steps S3 and S4 carry out the training of batch processing next time.The present invention ensure that identical for classification sample size each in trained every batch of data in data plane, effectively improve the generalization ability of training pattern.
Description
Technical field
The present invention relates to machine learning techniques fields, and in particular to a kind of overhead transmission line based on classification stochastical sampling
Fault detection data sample batch processing training method.
Background technique
Overhead transmission line is largely directly characterized as electric system important component, operation conditions
The operation conditions of entire power grid.For the reliability for ensuring operation of power networks, patrol officer need to periodically examine overhead transmission line
It looks into.Currently, there are mainly two types of detection modes: traditional detection mode and intelligent measurement mode.Traditional detection mode relies primarily on people
Work carries out, firstly, traditional detection mode generally requires to expend a large amount of man power and material;Secondly, high voltage transmission line scale it is big,
It has a very wide distribution, the environment of surrounding has many potential danger, is unfavorable for artificial detection, so that traditional method for inspecting is applicable in model
It encloses and greatly reduces.For this problem, unmanned plane inspection technology is widely used in succession, they carry photograph or camera shooting
Equipment is flown along power transmission line corridor, and shooting at close range route obtains Aerial Images.Its shortcoming needs artificial searching picture
The defects of position, and mark out come.What traditional image labeling was manually performed, the understanding and mark of image are aligned
Really, but under big data environment, artificial mark workload is huge, and for the missing of some widgets, detection leakage phenomenon is very
Common phenomenon.
In recent years, with the rise of deep learning, scholar proposes the defect automatic marking technology based on deep learning.
Automatic marking technology because its excellent performance, meet modern intelligent and automation requirement the advantages that due to be gradually taken seriously.
The important branch that deep learning learns as robot is the important channel for making computer have intelligence, main research meter
The learning behavior of the mankind is simulated, realized to calculation machine how, identifies existing knowledge, obtains new knowledge, constantly improves performance and improve certainly
Body.Therefore, deep learning is combined with modern inspection technology, have in terms of electric power defects detection very high convenience with it is excellent
More property.
In recent years, with the further investigation of deep learning, the target detection based on deep learning achieves astonishing achievements.
However, electric power defects detection and target detection are there are while similitude, the two also has otherness to a certain degree.Wherein the most
Significantly surely belong to positive and negative data sample imbalance, imbalance here in electric power defects detection and refers to that the quantity of defect classification is remote
Less than the quantity of normal category sample.Traditional algorithm of target detection often assumes the sample number that each classification of training sample includes
Amount be it is same or similar, i.e., number of samples of all categories be it is balanced, but the defects of field of power system detection be not to be inconsistent
This hypothesis is closed, the electric system most of the time is in safe and stable operating status, the sample of defect or fault category
It only occupies the minority, is largely normal.In general, this unbalanced training sample will lead to training pattern on test set
Stress the more normal category of number of samples, and the defect classification that " despising " number of samples is fewer, however, in reality
The correct meaning for distinguishing minority class defect sample is much higher than whole classification accuracy, this segmental defect sample there is a serious shortage of to not
Balanced sort increases difficulty.
Therefore, in order to further enhance model generalization ability, solve network training when unmanned plane inspection take photo by plane in figure not
Balance sample handles problem, makes and being correspondingly improved in original deep learning algorithm model batch processing training, is just able to satisfy existing
For intelligent patrol detection requirement.It is necessary in view of the above-mentioned problems, proposing a kind of overhead transmission line defect based on classification stochastical sampling
Detection data sample batch processing training method.
Summary of the invention
It is an object of the invention to solve the problems, such as in the electric power defects detection based on deep learning class imbalance this, mention
A kind of overhead transmission line fault detection data sample batch processing training method based on classification stochastical sampling out, is effectively relieved electricity
The weight number of samples on the upside of test set of training pattern caused by defect categorical data is on the low side in power defects detection is more normal
Classification.
The purpose of the present invention can be reached by adopting the following technical scheme that:
A kind of overhead transmission line fault detection data sample batch processing training method based on classification stochastical sampling, including
Following steps:
S1, determine that collected overhead transmission line fault detection data sample includes several classifications, category is marked
Note, then counts the quantity of the included sample of every one kind, is denoted as N respectively1、N2…Nn, n is the quantity of classification type;
S2, suitable algorithm model batchparameters N is chosen according to the quantity for counting every a kind of sample in step S1batch
Numerical value;
S3, it is to guarantee that each classification sample size is identical in batching data, randomly selects N from every one kindbatch/ n
Data sample;
S4, the data sample randomly selected in step S3 is combined, is trained after upsetting sequence as batch processing
Data sample;
S5, to the training of this batch of data after, repeat the above steps S3 and step S4 carry out the training of batch processing next time.
Detailed process is as follows by step S1 as a preferred technical solution:
The data that unmanned plane collection in worksite arrives are arranged, satisfactory picture sample is picked out, subsequent category stores number
According to sample, the quantity of every included sample of one kind is counted, is denoted as N respectively1、N2…Nn, finally corresponding mark is added for every picture
Label;
Wherein, the classification type of the overhead transmission line fault detection data sample includes normally, missing, loosens,
The quantity n of classification type is equal to 3 at this time.
Detailed process is as follows by step S2 as a preferred technical solution:
Suitable algorithm model batchparameters (N is chosen according to the quantity for counting every a kind of sample in step S1batch)
Numerical value, it should be noted that the N of selectionbatchMake Nbatch/ n is integer, to facilitate subsequent processing;
Detailed process is as follows by step S3 as a preferred technical solution:
N is randomly selected from every a kind of data samplebatch/ n data sample, it is ensured that be used for each batch processing of deep learning
Data in each classification sample size for including it is identical;
Detailed process is as follows by step S4 as a preferred technical solution:
The step S3 data sample randomly selected is combined, the number trained after sequence as batch processing is upset
According to sample;
Detailed process is as follows by step S5 as a preferred technical solution:
The data after sequence will be upset in S4 and be used for batch processing training deep learning defects detection model, until epicycle training
Terminate, repeats the above steps, carry out the training of next round.
The present invention has the following advantages and effects with respect to the prior art:
1, the method for the present invention ensure that in data plane for classification sample size phase each in trained every batch of data
Together, the generalization ability of training pattern is effectively increased;
2, the method for the present invention uses stochastical sampling, it is ensured that the chance that each data is drawn into is identical, and method is easy to real
It is existing, it can canbe used on line in memory.
Detailed description of the invention
Fig. 1 is typical sample exemplary diagram, wherein Fig. 1 (a) is that the closed pin of sample is normal, schematic diagram of missing;Fig. 1 (b)
It is the schematic diagram that the closed pin of sample loosens;
Fig. 2 is exemplary sample block diagram;
Fig. 3 is example training flow chart.
Specific embodiment
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with the embodiment of the present invention
In attached drawing, technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described embodiment is
A part of the embodiment of the present invention, instead of all the embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art
Every other embodiment obtained without making creative work, shall fall within the protection scope of the present invention.
Embodiment
As shown in Figure 1, power connection fitting closed pin have normally with defect two states, defect can be subdivided into loosening with
Fall off two states, and exemplary sample block diagram shown in Fig. 2 is summarized simply as follows stochastical sampling, recombination, trains these three ranks
Section, algorithm flow chart shown in Fig. 3 illustrate the model training flow chart of closed pin defects detection, comprising the following steps:
S1, normal, missing will be divided by the unmanned plane figure of screening, loosen 3 classes, category store picture not it
Label is added, then counting each classification includes that sample size is as follows: normal class N1=780, lack class N2=530, loosen class
N3=270;
S2, according to the data volume of all categories in step S1 for including, enable Nbatch=30, algorithm other parameters default is constant;
S3, input data sample and respective labels, from randomly selecting 10 data sample integration in each classification
Come, upsets the batch processing training after sequence for model;
S4, after training to the batch data, judge whether to reach specified training batch, if not having, repeat step S3
Until reaching specified training batch;
After S5, specified trained batch to be achieved, epicycle training terminates, and repeats step S3, S4 until reaching specified exercise wheel
Number.
The above embodiment is a preferred embodiment of the present invention, but embodiments of the present invention are not by above-described embodiment
Limitation, other any changes, modifications, substitutions, combinations, simplifications made without departing from the spirit and principles of the present invention,
It should be equivalent substitute mode, be included within the scope of the present invention.
Claims (4)
1. a kind of overhead transmission line fault detection data sample batch processing training method based on classification stochastical sampling, feature
Be, the training method the following steps are included:
S1, the classification type that collected overhead transmission line fault detection data sample includes is determined, category is marked,
The quantity for then counting each included sample of classification type respectively, is denoted as N1、N2…Nn, n is the quantity of classification type;
S2, batchparameters N is chosen according to the quantity of each included sample of classification typebatchNumerical value;
S3, N is randomly selected from the sample of each classification typebatch/ n data sample;
S4, the data sample randomly selected in step S3 is combined, is upset after sequence as batch processing training
Data sample;
S5, to the training of this batch of data after, repeat the above steps S3 and step S4 carry out the training of batch processing next time.
2. the overhead transmission line fault detection data sample batch processing according to claim 1 based on classification stochastical sampling
Training method, which is characterized in that the classification type of the overhead transmission line fault detection data sample includes normal, scarce
It loses, loosen.
3. the overhead transmission line fault detection data sample batch processing according to claim 1 based on classification stochastical sampling
Training method, which is characterized in that the batchparameters NbatchNumerical value selection to make Nbatch/ n is integer.
4. the overhead transmission line fault detection data sample batch processing according to claim 1 based on classification stochastical sampling
Training method, which is characterized in that completed in the step S1 by adding corresponding label to picture every in data sample
Classification type is marked.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811407820.7A CN109598349A (en) | 2018-11-23 | 2018-11-23 | Overhead transmission line fault detection data sample batch processing training method based on classification stochastical sampling |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811407820.7A CN109598349A (en) | 2018-11-23 | 2018-11-23 | Overhead transmission line fault detection data sample batch processing training method based on classification stochastical sampling |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109598349A true CN109598349A (en) | 2019-04-09 |
Family
ID=65959606
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811407820.7A Pending CN109598349A (en) | 2018-11-23 | 2018-11-23 | Overhead transmission line fault detection data sample batch processing training method based on classification stochastical sampling |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109598349A (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110324207A (en) * | 2019-07-10 | 2019-10-11 | 深圳市智物联网络有限公司 | A kind of detection method and device of data collection station |
CN112529033A (en) * | 2020-09-22 | 2021-03-19 | 陕西土豆数据科技有限公司 | Method for solving data imbalance of remote sensing image multi-classification scene segmentation algorithm |
CN114091665A (en) * | 2020-07-30 | 2022-02-25 | 北京四维图新科技股份有限公司 | Method for training deep neural network and network thereof |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106548196A (en) * | 2016-10-20 | 2017-03-29 | 中国科学院深圳先进技术研究院 | A kind of random forest sampling approach and device for non-equilibrium data |
CN107578071A (en) * | 2017-10-13 | 2018-01-12 | 北京工业大学 | The unbalanced method of solution data based on Epoch |
CN108197697A (en) * | 2017-12-29 | 2018-06-22 | 汕头大学 | A kind of dynamic method for resampling of trained deep neural network |
-
2018
- 2018-11-23 CN CN201811407820.7A patent/CN109598349A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106548196A (en) * | 2016-10-20 | 2017-03-29 | 中国科学院深圳先进技术研究院 | A kind of random forest sampling approach and device for non-equilibrium data |
CN107578071A (en) * | 2017-10-13 | 2018-01-12 | 北京工业大学 | The unbalanced method of solution data based on Epoch |
CN108197697A (en) * | 2017-12-29 | 2018-06-22 | 汕头大学 | A kind of dynamic method for resampling of trained deep neural network |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110324207A (en) * | 2019-07-10 | 2019-10-11 | 深圳市智物联网络有限公司 | A kind of detection method and device of data collection station |
CN114091665A (en) * | 2020-07-30 | 2022-02-25 | 北京四维图新科技股份有限公司 | Method for training deep neural network and network thereof |
CN112529033A (en) * | 2020-09-22 | 2021-03-19 | 陕西土豆数据科技有限公司 | Method for solving data imbalance of remote sensing image multi-classification scene segmentation algorithm |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109598349A (en) | Overhead transmission line fault detection data sample batch processing training method based on classification stochastical sampling | |
CN110033453A (en) | Based on the power transmission and transformation line insulator Aerial Images fault detection method for improving YOLOv3 | |
CN114240878A (en) | Routing inspection scene-oriented insulator defect detection neural network construction and optimization method | |
CN103926512B (en) | The method of simulation cross-linked polyethylene insulated cable shelf depreciation and analog detection platform | |
CN109711377A (en) | Standardize examinee's positioning and method of counting in the single-frame images of examination hall monitoring | |
CN111008641B (en) | Power transmission line tower external force damage detection method based on convolutional neural network | |
CN114898232B (en) | Photovoltaic power station unmanned aerial vehicle inspection method and system based on photovoltaic group string data analysis | |
CN106326808A (en) | Method for detecting bird nests in power transmission line poles based on unmanned plane images | |
CN106817708B (en) | A kind of inspection cluster construction method and device | |
CN109919925A (en) | Printed circuit board intelligent detecting method, system, electronic device and storage medium | |
CN108334815A (en) | Method for inspecting, on off state recognition methods and the system of second power equipment | |
CN110008877A (en) | A kind of substation isolating-switch detection recognition method based on Faster RCNN | |
CN112229845A (en) | Unmanned aerial vehicle high-precision winding tower intelligent inspection method based on visual navigation technology | |
CN110909660A (en) | Plastic bottle detection and positioning method based on target detection | |
CN111103307A (en) | Pcb defect detection method based on deep learning | |
CN109614888B (en) | Deep learning defect detection model training method based on overhead transmission line defect auxiliary data set | |
CN112488020A (en) | Water environment pollution condition detection and evaluation device based on unmanned aerial vehicle aerial photography data | |
CN114821852A (en) | Power grid defect depth identification inspection robot control system based on characteristic pyramid | |
CN111091534A (en) | Target detection-based pcb defect detection and positioning method | |
CN114494916A (en) | Black-neck crane monitoring and tracking method based on YOLO and DeepsORT | |
CN116563841B (en) | Detection method and detection device for power distribution network equipment identification plate and electronic equipment | |
CN109598293A (en) | Unmanned plane inspection based on classification balanced sample is taken photo by plane image data sample batch processing training method | |
CN109324264A (en) | A kind of discrimination method and device of distribution network line impedance data exceptional value | |
Lian et al. | Insulator defect detection algorithm based on improved YOLOv5 | |
CN106250924B (en) | A kind of newly-increased category detection method based on multi-instance learning |
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: 20190409 |
|
RJ01 | Rejection of invention patent application after publication |