CN109815943A - A kind of harmful influence storage stacking picture sample generation method and system - Google Patents
A kind of harmful influence storage stacking picture sample generation method and system Download PDFInfo
- Publication number
- CN109815943A CN109815943A CN201910204031.1A CN201910204031A CN109815943A CN 109815943 A CN109815943 A CN 109815943A CN 201910204031 A CN201910204031 A CN 201910204031A CN 109815943 A CN109815943 A CN 109815943A
- Authority
- CN
- China
- Prior art keywords
- sample
- picture
- picture sample
- generation
- harmful influence
- 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.)
- Granted
Links
Abstract
The present invention relates to a kind of harmful influence storage stacking picture sample generation method and systems, this method comprises: obtaining the true picture sample of the harmful influence storage stacking of the first quantity;GAN algorithm process is carried out to the true picture sample of the first quantity, obtains the generation picture sample of the harmful influence storage stacking of the second quantity;Edge Gradient Feature, Corner Detection and noise filtering are carried out to the generation picture sample of the second quantity, so as to generate the sample distribution of picture sample close to the sample distribution of the true picture sample.Technical solution provided by the invention; a large amount of picture sample is generated by using GAN algorithm; and corner points extraction and noise filtering are carried out to the picture sample of generation; so that sample distribution of the sample distribution of the picture sample generated close to true picture sample; the algorithm has important research significance to Stereo matching needed for harmful influence monitor subsequent, distance monitoring and three-dimensional reduction, and strong technical support is provided on sample for visual token in harmful influence storehouse.
Description
Technical field
The present invention relates to harmful influence storage technique fields, and in particular to a kind of harmful influence storage stacking picture sample generation side
Method and system.
Background technique
There is hazardous chemical characteristics, the improper use such as inflammable and explosive, poisonous and harmful can cause to environment and personal safety
Significant damage.In order to avoid during the storage of these hazardous chemicals due to leakage, fire etc. and to environment or artificial at danger
Evil, needs to carry out strict supervision to it.Stacking in library " five away from " is crucial monitoring parameter, and binocular distance measurement is to solve this to ask
One of technological means of topic.
In face of this problem, need to monitor harmful influence warehouse in real time, early warning, the monitor mode taken at this stage is mainly people
Work video monitoring.But this mode needs to expend a large amount of labours, and is difficult to ensure being perfectly safe for storehouse, in data transmission
It works for that rate is relatively low, is not able to satisfy the requirement of modern enterprise informationization.
Machine vision is quickly grown in artificial intelligence field, computer can by safety, dangerous positive and negative sample learning,
The feature distribution of kinds of goods storage is grasped, realizes automatic monitoring.But machine learning, deep learning require a large amount of data sample,
Traditional sample acquisition mode efficiency is lower, and stacking in harmful influence library " five away from " binocular distance measurement is caused to face harmful influence storage
The problem of stacking picture sample wretched insufficiency.
Summary of the invention
In view of this, it is an object of the invention to overcome the deficiencies of the prior art and provide a kind of harmful influence storage stacking figures
Piece sample generating method and system, to solve the automatic necessary for monitoring harmful influence storage stacking picture sample of harmful influence in the prior art
The problem of wretched insufficiency.
In order to achieve the above object, the present invention adopts the following technical scheme:
A kind of harmful influence storage stacking picture sample generation method, comprising:
Step S1, the true picture sample of the harmful influence storage stacking of the first quantity is obtained;
Step S2, GAN algorithm process is carried out to the true picture sample of first quantity, obtains the danger of the second quantity
The generation picture sample of product storage stacking, the first quantity of the second quantity >;
Step S3, Edge Gradient Feature, Corner Detection and noise is carried out to the generation picture sample of second quantity to filter
Wave, so that sample distribution of the sample distribution for generating picture sample close to the true picture sample.
Preferably, the step S1, comprising:
Using the video of binocular camera shooting harmful influence storage stacking, video flowing is carried out by the extraction frame program prestored and is mentioned
Frame obtains the true picture sample of the harmful influence storage stacking of the first quantity.
Preferably, the step S2 includes:
Step S21, using the harmful influence of first quantity storage stacking picture sample as training set, while one is defined
Noise Z is input in the generator G of production confrontation network as stochastic variable, the noise Z of input is mapped to generation figure
Piece G (Z);
Step S22, picture G (Z) will be generated to be input in the arbiter D of production confrontation network, to judge the generation
Whether picture G (Z) is close to true picture, to screen to the generation picture G (Z).
Preferably, Corner Detection is carried out to the generation picture sample of second quantity in the step S3, specifically:
Using Harris Corner Detection Algorithm, Corner Detection is carried out to the generation picture sample of second quantity.
Preferably, Corner Detection is carried out to the generation picture sample of second quantity in the step S3, comprising:
To any generation picture sample, the sliding on any direction is carried out in picture sample using a fixed window,
Compare the pixel grey scale variation degree before sliding and in the case of sliding latter two in window, if there is the cunning on any direction
It is dynamic, larger grey scale change is suffered from, then determines that there are angle points in the window.
Preferably, noise filtering is carried out to the generation picture sample of second quantity in the step S3, comprising:
To the angle point that Harris Corner Detection Algorithm detects, non-maxima suppression and k-means cluster denoising are carried out, with
Harris Corner Detection Algorithm detected error angle point caused by the irradiation of removal light, shade and other non-stacking articles.
Preferably, the angle point detected to Harris Corner Detection Algorithm carries out k-means cluster denoising, comprising:
To any generation picture sample, three central points are randomly selected in picture sample;
All angle points are traversed, each angle point is divided to nearest central point, obtain multiple clusters;
The average value for calculating each cluster, as new central point;
Judge whether the cluster restrains, if so, otherwise the angle point is determined as error corner point by filing cluster result.
In addition, the invention also provides a kind of harmful influence storage stacking picture samples to generate system, comprising:
Module is obtained, the true picture sample of the harmful influence storage stacking for obtaining the first quantity;
Generation module carries out GAN algorithm process for the true picture sample to first quantity, obtains the second quantity
Harmful influence storage stacking generation picture sample, the first quantity of the second quantity >;
Denoise module, for second quantity generation picture sample carry out Edge Gradient Feature, Corner Detection and
Noise filtering, so that sample distribution of the sample distribution for generating picture sample close to the true picture sample.
Preferably, the acquisition module, is specifically used for:
Using the video of binocular camera shooting harmful influence storage stacking, video flowing is carried out by the extraction frame program prestored and is mentioned
Frame obtains the true picture sample of the harmful influence storage stacking of the first quantity.
Preferably, the generation module has and is used for:
Using the harmful influence of first quantity storage stacking picture sample as training set, while defining a noise Z and making
For stochastic variable, it is input in the generator G of production confrontation network, the noise Z of input is mapped to and generates picture G (Z);
Picture G (Z) will be generated to be input in the arbiter D of production confrontation network, to judge the generation picture G (Z)
Whether close to true picture, to be screened to the generation picture G (Z).
The invention adopts the above technical scheme, at least have it is following the utility model has the advantages that
Generate a large amount of picture sample by using GAN algorithm, and to the picture sample of generation carry out corner points extraction and
Noise filtering, so that the sample distribution of the picture sample generated is close to the sample distribution of true picture sample, the algorithm is to dangerization
Stereo matching needed for product monitor subsequent, distance monitoring and three-dimensional reduction have important research significance, are harmful influence storehouse
Interior visual token provides strong technical support on sample.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention for those of ordinary skill in the art without creative efforts, can be with
It obtains other drawings based on these drawings.
Fig. 1 is a kind of flow chart for harmful influence storage stacking picture sample generation method that one embodiment of the invention provides;
Fig. 2 is the video interception in the simulation harmful influence warehouse that one embodiment of the invention provides;
Fig. 3 is the Harris algorithm Corner Detection effect picture that one embodiment of the invention provides;
Fig. 4 is the generation picture sample denoising front and back accuracy rate comparison diagram that one embodiment of the invention provides;
Fig. 5 is the schematic block that a kind of harmful influence storage stacking picture sample that one embodiment of the invention provides generates system
Figure.
Specific embodiment
To make the object, technical solutions and advantages of the present invention clearer, technical solution of the present invention will be carried out below
Detailed description.Obviously, described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.Base
Embodiment in the present invention, those of ordinary skill in the art are obtained all without making creative work
Other embodiment belongs to the range that the present invention is protected.
Below by drawings and examples, technical scheme of the present invention will be described in further detail.
Referring to Fig. 1, a kind of harmful influence storage stacking picture sample generation method of one embodiment of the invention offer, comprising:
Step S1, the true picture sample of the harmful influence storage stacking of the first quantity is obtained;
Step S2, GAN algorithm process is carried out to the true picture sample of first quantity, obtains the danger of the second quantity
The generation picture sample of product storage stacking, the first quantity of the second quantity >;
Step S3, Edge Gradient Feature, Corner Detection and noise is carried out to the generation picture sample of second quantity to filter
Wave, so that sample distribution of the sample distribution for generating picture sample close to the true picture sample.
It is understood that production confrontation network (GAN, Generative Adversarial Networks) is this
Deep learning model, small sample generates large sample, is in recent years in complex distributions without prison for carrying out the machine learning of next step
Educational inspector practises one of the method for most prospect.Generating confrontation network is the antagonistic process for simulating two models, one of them is to generate
Device G (Generator), captures the distribution of truthful data, and generation approaches the data being really distributed;The other is arbiter D
(Discriminator), judge whether data come from true distribution, output data is from the probability being really distributed, i.e., one 0
Numerical value between~1.
It is understood that technical solution proposed by the present invention, generates a large amount of picture sample by using GAN algorithm,
And corner points extraction and noise filtering are carried out to the picture sample of generation, so that the sample distribution of the picture sample generated is close to very
The sample distribution of real picture sample, the algorithm is to Stereo matching needed for harmful influence monitor subsequent, distance monitoring and three-dimensional reduction
With important research significance, strong technical support is provided on sample for visual token in harmful influence storehouse.
Preferably, the step S1, comprising:
Using the video of binocular camera shooting harmful influence storage stacking, video flowing is carried out by the extraction frame program prestored and is mentioned
Frame obtains the true picture sample of the harmful influence storage stacking of the first quantity.
Preferably, the step S2 includes:
Step S21, using the harmful influence of first quantity storage stacking picture sample as training set, while one is defined
Noise Z is input in the generator G of production confrontation network as stochastic variable, the noise Z of input is mapped to generation figure
Piece G (Z);
Step S22, picture G (Z) will be generated to be input in the arbiter D of production confrontation network, to judge the generation
Whether picture G (Z) is close to true picture, to screen to the generation picture G (Z).
It should be noted that GAN is inspired by the zero-sum game in game theory, generation problem is regarded as arbiter D and generation
The confrontation and game of the two models of device G: generator G (generally refer to be uniformly distributed or normal distribution) from given noise is produced
GCMS computer data, arbiter D differentiate output and the truthful data of generator G.The former attempts to generate closer to true data,
Correspondingly, the latter attempts more ideally to differentiate truthful data and generates data.As a result, two models in confrontation progress, into
Continue to fight after step, the data obtained by production network are also just more and more perfect, and approaching to reality data are desired so as to generate
Obtained data (picture, sequence, video etc.).
GAN exactly will generate distribution by D and push to really be distributed from the perspective of probability distribution, and then excellent again
Change D, until it is close to the generated data and truthful data generated, the equilibrium point Nash is reached, to generate distribution and true distribution
Overlapping generates the very close data being really distributed.
If z is random noise, x is truthful data, and production network and discriminate network can be indicated with G and D respectively,
Middle D is considered as two classifiers, then being indicated using cross entropy, can write:
Wherein the logD (x) of first item indicates judgement of the arbiter to truthful data, Section 2 log (1-D (G (z))) table
Show then to the synthesis of data and judgement.Minimax (Max-min) game in this way, is separately optimized G to cycle alternation
Production network required for being trained with D and discriminate network, until reaching the equilibrium point Nash.
Since most of models based on GAN generated for picture sample are constructed with convolutional network, convolution exists
Information is handled in local neighborhood, therefore convolutional layer is used only and is computationally not suitable for the long-range dependence modeling in image,
Therefore the high-resolution details that traditional convolution GAN is generated only is used as the letter of the space partial points on low resolution characteristic pattern
Number.It is added and notices that the improved production of power module fights network SAGAN, will be introduced into from attention mechanism in convolution GAN, as
The supplement of convolution has very big advantage in multi-class image synthesis task.In SAGAN, it can be used from all
The clue of feature locations generates details, allows to generate the modeling that task carries out attention power drive, long range dependent to image.This
Outside, discriminator can be consistent with each other with the highly detailed feature of the distal portions of check image.Preferably, the present embodiment uses
SAGAN carries out picture sample generation, and the picture sample generated is made to be more concerned about main body kinds of goods part.
Preferably, Corner Detection is carried out to the generation picture sample of second quantity in the step S3, specifically:
Using Harris Corner Detection Algorithm, Corner Detection is carried out to the generation picture sample of second quantity.
Preferably, Corner Detection is carried out to the generation picture sample of second quantity in the step S3, comprising:
To any generation picture sample, the sliding on any direction is carried out in picture sample using a fixed window,
Compare the pixel grey scale variation degree before sliding and in the case of sliding latter two in window, if there is the cunning on any direction
It is dynamic, larger grey scale change is suffered from, then determines that there are angle points in the window.
Preferably, noise filtering is carried out to the generation picture sample of second quantity in the step S3, comprising:
To the angle point that Harris Corner Detection Algorithm detects, non-maxima suppression and k-means cluster denoising are carried out, with
Harris Corner Detection Algorithm detected error angle point caused by the irradiation of removal light, shade and other non-stacking articles.
Preferably, the angle point detected to Harris Corner Detection Algorithm carries out k-means cluster denoising, comprising:
To any generation picture sample, three central points are randomly selected in picture sample;
All angle points are traversed, each angle point is divided to nearest central point, obtain multiple clusters;
The average value for calculating each cluster, as new central point;
Judge whether the cluster restrains, if so, otherwise the angle point is determined as error corner point by filing cluster result.
If window centered on pixel (x, y) upper mobile u unit length in the x-direction, upper mobile v is a in the y-direction
Unit length.Accordingly, the analytic expression that Harris gives grey scale change measurement is formula (1):
Wherein, E (x, y) is the grey scale change amount in window;W is the window of image;I is the gray scale of image.
If monitoring offset of the angle point as window of image, its auto-correlation function E (x, y) also can accordingly change.Its
In: A, B, C are the approximate expressions of second-order equation directional differential, it is also possible to following expression:
Wherein, what h (x, y) was indicated is Gaussian smoothing filter function, and what X, Y were indicated is the differential on single order direction, image
Gray scale with the difference operator of x, is used respectively | and 1 0-1 | and | 1 0-1 |TIt indicates.
Equation E (x, y) can be converted are as follows:
E (x, y)=[x, y] M [x, y]T (5)
If two characteristic values of finally obtained system Metzler matrix are all big, image grayscale auto-correlation functions at this time
Two orthogonal directions extreme value curvature it is larger, that is to say, that this point be angle point.Figure herein based on harris Corner Detection
As defining grey value characteristics, and using grey value characteristics as matching benchmark.
The picture sample that GAN is generated carries out Harris Corner Detection, has carried out non-maximum to traditional Harris algorithm
Inhibit and k-means clusters the optimization for denoising (hereinafter referred to as LK denoising).For light irradiation, shade and other non-stacking articles
When interference, hierarchical cluster removal is carried out using k-means
As follows for script Harris improvement formula, n is data volume, and I is iteration number.Data distance center point distance
Judgement is as shown in formula (7).
X in formula (7)s, ysFor Selection Center point, n is angle point number, xi, yiBe it is non-selected centered on the angle point position put
It sets, d is the distance away from central point.According to above-mentioned traditional Corner Detection LK Denoising Algorithm derivation of equation in addition, formula can be summarized to obtain
(8):
Wherein, rows and col is the length and height of picture pixels, unit px.Setting kSize mainly inhibits location of pixels iteration
Go out frame phenomenon in comparison procedure.δ is to meet receptance function R threshold size, experience value is 20000, filters Harris with this
The erroneous point that algorithm Corner Detection goes out.
In concrete practice, in order to verify the reality of this harmful influence storage stacking picture sample generation method provided by the invention
With property, applicant is tested, and test process and result are specific as follows:
In laboratory environments, small-sized stacking formula simulation harmful influence warehouse has been built, stacking article has been simulated with wooden unit, such as schemes
Shown in 2.The stacking for building 4 20cm*20cm*20cm is full goods state, successively removes a certain amount of cargo, is formed and amounts to 5 kinds not
With the stacking model of cargo state.Small sample acquisition is carried out using Fig. 2 upper right binocular camera shooting video, then writes extraction frame
Program.The resolution ratio of binocular camera is 1920*1080, and signal-to-noise ratio 39db, pixel is 2,000,000.GPU is Xeon E5-
2683v3 processor, 11G video memory K80 independent display card server.Development platform is windows7 operating system, is used
Tensorflow1.12 frame is realized.
It first attempted to carry out GAN study with computer CPU, speed is about that 24 seconds iteration are primary, server CPU training speed
It spends similar.After use server GPU instead and be trained, speed is about that every 2.2 seconds iteration are primary, and speed greatly improves, and can meet sample
This acquisition rate requirement.Final argument is provided that every image viewing training result of 500 grey iterative generations, every 2000 times
Iteration is recorded as a generator version, for generating batch picture.
After tested, the available picture sample of clarity is produced after generator G iteration about 4500 times;It can after iteration 8000 times
It generates and the visibly different generation picture sample of known picture sample.
After 10000 iteration, form of diverse, clearly stacking sample have been produced.And illumination, yin are learnt
The feature of shadow, remaining barrier, and realize in the sample.
The picture sample that true picture sample and GAN are generated compares discovery, all angles of the stacking picture of the two
Feature all height are similar, therefore the generation result for generating confrontation network is reliable, can handle to provide for subsequent monitoring and largely may be used
The picture sample leaned on.
It is as shown in Figure 3 by result shown in Harris Corner Detection to generate picture sample.
As seen from Figure 3, it is smaller to be illuminated by the light shadow effect for generation figure, is only affected by barrier.Angle steel joint LK is denoised
It can remove mainly due to most of noise caused by the barrier of right, accuracy rate is as shown in Figure 4 before and after generating image denoising.
The picture sample of generation, final evaluation index carry out accuracy rate using F-measure value in the prior art respectively
With the calculating of recall rate.Accuracy rate size should be no more than 1, and its value is the bigger the better.Recall rate represents the measurement of covering, and value is got over
It is better close to 1 value.For the contradictory relation of accuracy rate and recall rate, the standard for the evaluation Corner Detection that F-measure can be integrated
True rate, value are about better close to 1.For all generating image calculated result statistics, result such as table one denoises effect and image
Shown in quality evaluation table.
Table one
By table one as it can be seen that the accuracy rate for generating picture sample is lower than true picture sample, but recall rate is than true
Picture sample is closer to 1;Total F-measure value, angle point testing result is it can be found that generate figure after comparison LK algorithm denoising
Piece sample and 1 deviation 13.20%, true picture sample and 1 deviation 15.39%, therefore the generation picture of GAN can be used as image
The sample of processing uses.
In conclusion the present invention increases the sample size of harmful influence storage binocular ranging, benefit by generating confrontation network
300 pictures that frame obtains are mentioned by 10000 iteration with video flowing, generate 10000 new generation samples, it was demonstrated that life
At the operability of the practical application of neural network;
The present invention has carried out Edge Gradient Feature, Corner Detection and LK to the sample of generation and has denoised, the experimental results showed that,
The rate that GAN generates sample compared to video flowing for great amount of samples improves about 8 to 10 times, Corner Detection and LK are denoised etc.
For Processing Algorithm, sample and authentic specimen data deviation very little are generated, effective branch can be provided for subsequent binocular ranging etc.
Support technology.
In addition, the invention also provides a kind of harmful influence storage stacking picture samples to generate system 100, packet referring to Fig. 5
It includes:
Module 101 is obtained, the true picture sample of the harmful influence storage stacking for obtaining the first quantity;
Generation module 102 carries out GAN algorithm process for the true picture sample to first quantity, obtains second
The generation picture sample of the harmful influence storage stacking of quantity, the first quantity of the second quantity >;
Module 103 is denoised, carries out Edge Gradient Feature, Corner Detection for the generation picture sample to second quantity
And noise filtering, so that sample distribution of the sample distribution for generating picture sample close to the true picture sample.
It is understood that technical solution provided by the invention, generates a large amount of picture sample by using GAN algorithm,
And corner points extraction and noise filtering are carried out to the picture sample of generation, so that the sample distribution of the picture sample generated is close to very
The sample distribution of real picture sample, the algorithm is to Stereo matching needed for harmful influence monitor subsequent, distance monitoring and three-dimensional reduction
With important research significance, strong technical support is provided on sample for visual token in harmful influence storehouse.
Preferably, the acquisition module 101, is specifically used for:
Using the video of binocular camera shooting harmful influence storage stacking, video flowing is carried out by the extraction frame program prestored and is mentioned
Frame obtains the true picture sample of the harmful influence storage stacking of the first quantity.
Preferably, the generation module 102 has and is used for:
Using the harmful influence of first quantity storage stacking picture sample as training set, while defining a noise Z and making
For stochastic variable, it is input in the generator G of production confrontation network, the noise Z of input is mapped to and generates picture G (Z);
Picture G (Z) will be generated to be input in the arbiter D of production confrontation network, to judge the generation picture G (Z)
Whether close to true picture, to be screened to the generation picture G (Z).
The above description is merely a specific embodiment, but scope of protection of the present invention is not limited thereto, any
Those familiar with the art in the technical scope disclosed by the present invention, can easily think of the change or the replacement, and should all contain
Lid is within protection scope of the present invention.Therefore, protection scope of the present invention should be based on the protection scope of the described claims.
Term " first ", " second " are used for descriptive purposes only and cannot be understood as indicating or suggesting relative importance.Term " multiple " refers to
Two or more, unless otherwise restricted clearly.
Claims (10)
- The stacking picture sample generation method 1. a kind of harmful influence is stored in a warehouse characterized by comprisingStep S1, the true picture sample of the harmful influence storage stacking of the first quantity is obtained;Step S2, GAN algorithm process is carried out to the true picture sample of first quantity, obtains the harmful influence storehouse of the second quantity Store up the generation picture sample of stacking, the first quantity of the second quantity >;Step S3, Edge Gradient Feature, Corner Detection and noise filtering are carried out to the generation picture sample of second quantity, with Make the sample distribution for generating picture sample close to the sample distribution of the true picture sample.
- 2. the method according to claim 1, wherein the step S1, comprising:Using the video of binocular camera shooting harmful influence storage stacking, video flowing is carried out by the extraction frame program prestored and mentions frame, Obtain the true picture sample of the harmful influence storage stacking of the first quantity.
- 3. the method according to claim 1, wherein the step S2 includes:Step S21, using the harmful influence of first quantity storage stacking picture sample as training set, while a noise is defined Z is input in the generator G of production confrontation network as stochastic variable, the noise Z of input is mapped to and generates picture G (Z);Step S22, picture G (Z) will be generated to be input in the arbiter D of production confrontation network, to judge the generation picture G (Z) whether close to true picture, to be screened to the generation picture G (Z).
- 4. the method according to claim 1, wherein to the generation picture of second quantity in the step S3 Sample carries out Corner Detection, specifically:Using Harris Corner Detection Algorithm, Corner Detection is carried out to the generation picture sample of second quantity.
- 5. according to the method described in claim 4, it is characterized in that, to the generation picture of second quantity in the step S3 Sample carries out Corner Detection, comprising:To any generation picture sample, the sliding on any direction is carried out in picture sample using a fixed window, is compared Pixel grey scale variation degree before sliding and in the case of sliding latter two in window, if there is the sliding on any direction, all There is larger grey scale change, then determines that there are angle points in the window.
- 6. according to the method described in claim 4, it is characterized in that, to the generation picture of second quantity in the step S3 Sample carries out noise filtering, comprising:To the angle point that Harris Corner Detection Algorithm detects, non-maxima suppression and k-means cluster denoising are carried out, with removal Harris Corner Detection Algorithm detected error angle point caused by light irradiation, shade and other non-stacking articles.
- 7. according to the method described in claim 6, it is characterized in that, the angle detected to Harris Corner Detection Algorithm Point carries out k-means cluster denoising, comprising:To any generation picture sample, three central points are randomly selected in picture sample;All angle points are traversed, each angle point is divided to nearest central point, obtain multiple clusters;The average value for calculating each cluster, as new central point;Judge whether the cluster restrains, if so, otherwise the angle point is determined as error corner point by filing cluster result.
- 8. a kind of harmful influence storage stacking picture sample generates system characterized by comprisingModule is obtained, the true picture sample of the harmful influence storage stacking for obtaining the first quantity;Generation module carries out GAN algorithm process for the true picture sample to first quantity, obtains the danger of the second quantity The generation picture sample of change product storage stacking, the first quantity of the second quantity >;Module is denoised, carries out Edge Gradient Feature, Corner Detection and noise for the generation picture sample to second quantity Filtering, so that sample distribution of the sample distribution for generating picture sample close to the true picture sample.
- 9. system according to claim 8, which is characterized in that the acquisition module is specifically used for:Using the video of binocular camera shooting harmful influence storage stacking, video flowing is carried out by the extraction frame program prestored and mentions frame, Obtain the true picture sample of the harmful influence storage stacking of the first quantity.
- 10. system according to claim 8, which is characterized in that the generation module has and is used for:Regard the harmful influence of first quantity storage stacking picture sample as training set, at the same define a noise Z as with Machine variable is input in the generator G of production confrontation network, the noise Z of input is mapped to and generates picture G (Z);Picture G (Z) will be generated to be input in the arbiter D of production confrontation network, whether to judge the generation picture G (Z) Close to true picture, to be screened to the generation picture G (Z).
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910204031.1A CN109815943B (en) | 2019-03-18 | 2019-03-18 | Hazardous chemical storage stacking picture sample generation method and system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910204031.1A CN109815943B (en) | 2019-03-18 | 2019-03-18 | Hazardous chemical storage stacking picture sample generation method and system |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109815943A true CN109815943A (en) | 2019-05-28 |
CN109815943B CN109815943B (en) | 2021-02-09 |
Family
ID=66609344
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910204031.1A Active CN109815943B (en) | 2019-03-18 | 2019-03-18 | Hazardous chemical storage stacking picture sample generation method and system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109815943B (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111275126A (en) * | 2020-02-12 | 2020-06-12 | 武汉轻工大学 | Sample data set generation method, device, equipment and storage medium |
CN112101125A (en) * | 2020-08-21 | 2020-12-18 | 浙江百世技术有限公司 | Method and device for detecting stacking degree of express goods |
CN112258635A (en) * | 2020-10-26 | 2021-01-22 | 北京石油化工学院 | Three-dimensional reconstruction method and device based on improved binocular matching SAD algorithm |
Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102004904A (en) * | 2010-11-17 | 2011-04-06 | 东软集团股份有限公司 | Automatic teller machine-based safe monitoring device and method and automatic teller machine |
CN105224946A (en) * | 2015-09-22 | 2016-01-06 | 成都融创智谷科技有限公司 | A kind of minimizing technology of pseudo-angle point |
CN105512685A (en) * | 2015-12-10 | 2016-04-20 | 小米科技有限责任公司 | Object identification method and apparatus |
CN105809657A (en) * | 2014-12-30 | 2016-07-27 | Tcl集团股份有限公司 | Angular point detection method and device |
CN106780524A (en) * | 2016-11-11 | 2017-05-31 | 厦门大学 | A kind of three-dimensional point cloud road boundary extraction method |
CN107563385A (en) * | 2017-09-02 | 2018-01-09 | 西安电子科技大学 | License plate character recognition method based on depth convolution production confrontation network |
CN108399432A (en) * | 2018-02-28 | 2018-08-14 | 成都果小美网络科技有限公司 | Object detecting method and device |
CN108470187A (en) * | 2018-02-26 | 2018-08-31 | 华南理工大学 | A kind of class imbalance question classification method based on expansion training dataset |
CN109165735A (en) * | 2018-07-12 | 2019-01-08 | 杭州电子科技大学 | Based on the method for generating confrontation network and adaptive ratio generation new samples |
US10209974B1 (en) * | 2017-12-04 | 2019-02-19 | Banjo, Inc | Automated model management methods |
-
2019
- 2019-03-18 CN CN201910204031.1A patent/CN109815943B/en active Active
Patent Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102004904A (en) * | 2010-11-17 | 2011-04-06 | 东软集团股份有限公司 | Automatic teller machine-based safe monitoring device and method and automatic teller machine |
CN105809657A (en) * | 2014-12-30 | 2016-07-27 | Tcl集团股份有限公司 | Angular point detection method and device |
CN105224946A (en) * | 2015-09-22 | 2016-01-06 | 成都融创智谷科技有限公司 | A kind of minimizing technology of pseudo-angle point |
CN105512685A (en) * | 2015-12-10 | 2016-04-20 | 小米科技有限责任公司 | Object identification method and apparatus |
CN106780524A (en) * | 2016-11-11 | 2017-05-31 | 厦门大学 | A kind of three-dimensional point cloud road boundary extraction method |
CN107563385A (en) * | 2017-09-02 | 2018-01-09 | 西安电子科技大学 | License plate character recognition method based on depth convolution production confrontation network |
US10209974B1 (en) * | 2017-12-04 | 2019-02-19 | Banjo, Inc | Automated model management methods |
CN108470187A (en) * | 2018-02-26 | 2018-08-31 | 华南理工大学 | A kind of class imbalance question classification method based on expansion training dataset |
CN108399432A (en) * | 2018-02-28 | 2018-08-14 | 成都果小美网络科技有限公司 | Object detecting method and device |
CN109165735A (en) * | 2018-07-12 | 2019-01-08 | 杭州电子科技大学 | Based on the method for generating confrontation network and adaptive ratio generation new samples |
Non-Patent Citations (2)
Title |
---|
WENKAI CHANG ET AL.: "Real-time segmentation of various insulators using generative adversarial networks", 《IET COMPUTER VISION》 * |
王雅欣等: "基于生成对抗网络的人脸热红外图像生成", 《人工智能》 * |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111275126A (en) * | 2020-02-12 | 2020-06-12 | 武汉轻工大学 | Sample data set generation method, device, equipment and storage medium |
CN112101125A (en) * | 2020-08-21 | 2020-12-18 | 浙江百世技术有限公司 | Method and device for detecting stacking degree of express goods |
CN112101125B (en) * | 2020-08-21 | 2024-05-03 | 浙江百世技术有限公司 | Method and device for detecting stacking degree of express goods |
CN112258635A (en) * | 2020-10-26 | 2021-01-22 | 北京石油化工学院 | Three-dimensional reconstruction method and device based on improved binocular matching SAD algorithm |
Also Published As
Publication number | Publication date |
---|---|
CN109815943B (en) | 2021-02-09 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109815943A (en) | A kind of harmful influence storage stacking picture sample generation method and system | |
CN109360232B (en) | Indoor scene layout estimation method and device based on condition generation countermeasure network | |
CN106530271B (en) | A kind of infrared image conspicuousness detection method | |
CN110321873A (en) | Sensitization picture recognition methods and system based on deep learning convolutional neural networks | |
Rogers et al. | Threat Image Projection (TIP) into X-ray images of cargo containers for training humans and machines | |
CN103679641B (en) | Depth image enhancement method and device | |
CN107784663A (en) | Correlation filtering tracking and device based on depth information | |
CN110378997A (en) | A kind of dynamic scene based on ORB-SLAM2 builds figure and localization method | |
CN111738373B (en) | Multi-sample anti-disturbance generation method and device, storage medium and computing equipment | |
CN106846362A (en) | A kind of target detection tracking method and device | |
CN109934810A (en) | A kind of defect classification method based on improvement population wavelet neural network | |
CN109902715A (en) | A kind of method for detecting infrared puniness target based on context converging network | |
CN108875655A (en) | A kind of real-time target video tracing method and system based on multiple features | |
CN110135344A (en) | Based on the method for detecting infrared puniness target for weighting fixed order expression | |
CN110472699A (en) | A kind of harmful biological motion blurred picture detection method of field of electric force institute based on GAN | |
CN109919095A (en) | The monitoring method and electronic equipment of stacking in harmful influence warehouse | |
CN106815567A (en) | A kind of flame detecting method and device based on video | |
CN105427341B (en) | Complex background video image multi-target detection method based on more variation level sets | |
CN103425958B (en) | A kind of method of motionless analyte detection in video | |
CN110321851A (en) | A kind of prohibited items detection method, device and equipment | |
US10776932B2 (en) | Determining whether ground is to be re-detected | |
CN109949337A (en) | Moving target detecting method and device based on Gaussian mixture model-universal background model | |
CN110992324A (en) | Intelligent dangerous goods detection method and system based on X-ray image | |
CN113409407B (en) | Countermeasure sample defense method for acquiring low-frequency information based on average compression | |
CN115457373A (en) | Anchor-frame-free one-stage insect pest image detection method and device and storage medium |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |