CN109815943B - Hazardous chemical storage stacking picture sample generation method and system - Google Patents
Hazardous chemical storage stacking picture sample generation method and system Download PDFInfo
- Publication number
- CN109815943B CN109815943B CN201910204031.1A CN201910204031A CN109815943B CN 109815943 B CN109815943 B CN 109815943B CN 201910204031 A CN201910204031 A CN 201910204031A CN 109815943 B CN109815943 B CN 109815943B
- Authority
- CN
- China
- Prior art keywords
- picture samples
- picture
- generated
- samples
- chemical storage
- 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.)
- Active
Links
Images
Abstract
The invention relates to a method and a system for generating a dangerous chemical storage stacking picture sample, wherein the method comprises the following steps: acquiring real picture samples of a first number of dangerous chemical warehouse stacks; performing GAN algorithm processing on the first number of real picture samples to obtain a second number of generated picture samples of the hazardous chemical storage stacks; and performing edge feature extraction, corner detection and noise filtering on a second number of generated picture samples to enable the sample distribution of the generated picture samples to be close to the sample distribution of the real picture samples. According to the technical scheme provided by the invention, a large number of picture samples are generated by adopting a GAN algorithm, and edge point extraction and noise filtering are carried out on the generated picture samples, so that the sample distribution of the generated picture samples is close to the sample distribution of a real picture sample.
Description
Technical Field
The invention relates to the technical field of hazardous chemical storage, in particular to a method and a system for generating stack picture samples of hazardous chemical storage.
Background
Dangerous chemicals have the characteristics of flammability, explosiveness, toxicity, harm and the like, and cause great harm to the environment and personal safety when being improperly used. In order to avoid the danger to the environment or people due to leakage, fire, etc. during the storage of these dangerous chemicals, strict supervision is required. The 'five-distance' stacking in the warehouse is a key monitoring parameter, and the binocular vision ranging is one of the technical means for solving the problem.
In order to solve the problem, the dangerous chemical warehouse needs to be monitored and pre-warned in real time, and the monitoring mode adopted at the present stage is mainly manual video monitoring. However, a large amount of labor is consumed in the method, absolute safety of the warehouse is difficult to guarantee, the data transmission and reporting efficiency is low, and the requirement of modern enterprise informatization cannot be met.
Machine vision develops rapidly in the artificial intelligence field, and the computer can master the characteristic distribution of goods storage through learning safe and dangerous positive and negative samples, realizes automatic monitoring. However, machine learning and deep learning both need a large number of data samples, and the traditional sample acquisition mode is low in efficiency, so that the problem that dangerous chemical storage stacking picture samples in a stacking five-distance binocular vision distance measurement plane in a dangerous chemical library are seriously insufficient is caused.
Disclosure of Invention
In view of the above, the present invention provides a method and a system for generating image samples of hazardous chemical storage stacks, so as to solve the problem of serious shortage of image samples of hazardous chemical storage stacks required for hazardous chemical automatic monitoring in the prior art.
In order to achieve the purpose, the invention adopts the following technical scheme:
a hazardous chemical storage stacking picture sample generation method comprises the following steps:
step S1, acquiring real picture samples of a first number of dangerous chemical storage stacks;
step S2, performing GAN algorithm processing on the first number of real picture samples to obtain a second number of generated picture samples of dangerous chemical storage stacks, wherein the second number is larger than the first number;
step S3, performing edge feature extraction, corner detection, and noise filtering on the second number of generated picture samples, so that the sample distribution of the generated picture samples is close to the sample distribution of the real picture samples.
Preferably, the step S1 includes:
the method comprises the steps of shooting videos of dangerous chemical storage stacks by using a binocular camera, carrying out video stream frame extraction through a prestored frame extraction program, and obtaining real picture samples of the dangerous chemical storage stacks in a first quantity.
Preferably, the step S2 includes:
step S21, taking the first quantity of dangerous chemical storage stacking picture samples as a training set, defining a noise Z as a random variable, and inputting the noise Z into a generator G of a generating countermeasure network to map the input noise Z into a generating picture G (Z);
step S22, inputting the generated picture g (z) into the discriminator D of the generative countermeasure network to judge whether the generated picture g (z) is close to the real picture, so as to filter the generated picture g (z).
Preferably, in step S3, the corner detection is performed on the second number of generated picture samples, specifically:
and performing corner detection on the second number of generated picture samples by adopting a Harris corner detection algorithm.
Preferably, the step S3 of performing corner detection on the second number of generated picture samples includes:
and for any generated picture sample, sliding in any direction on the picture sample by using a fixed window, comparing the gray level change degrees of pixels in the window under two conditions of before sliding and after sliding, and if the sliding in any direction has large gray level change, judging that an angular point exists in the window.
Preferably, the noise filtering the second number of generated picture samples in step S3 includes:
and (3) carrying out non-maximum suppression and k-means clustering denoising on the angular points detected by the Harris angular point detection algorithm so as to remove wrong angular points detected by the Harris angular point detection algorithm caused by light irradiation, shadow and other non-stacked articles.
Preferably, the performing k-means clustering denoising on the corners detected by the Harris corner detection algorithm includes:
randomly selecting three central points on any generated picture sample;
traversing all the angular points, and dividing each angular point to the nearest central point to obtain a plurality of clusters;
calculating the average value of each cluster as a new central point;
and judging whether the clustering is converged, if so, filing a clustering result, and otherwise, determining the corner as an error corner.
In addition, the invention also provides a system for generating the dangerous chemical storage stacking picture sample, which comprises the following steps:
the acquiring module is used for acquiring real picture samples of the dangerous chemical warehouse stacks in a first quantity;
the generating module is used for carrying out GAN algorithm processing on the first number of real picture samples to obtain a second number of generated picture samples of the dangerous chemical storage stacks, wherein the second number is larger than the first number;
and the denoising module is used for performing edge feature extraction, corner detection and noise filtering on the second number of generated image samples so as to enable the sample distribution of the generated image samples to be close to the sample distribution of the real image samples.
Preferably, the obtaining module is specifically configured to:
the method comprises the steps of shooting videos of dangerous chemical storage stacks by using a binocular camera, carrying out video stream frame extraction through a prestored frame extraction program, and obtaining real picture samples of the dangerous chemical storage stacks in a first quantity.
Preferably, the generating module has a module for:
taking the first quantity of dangerous chemical storage stacking picture samples as a training set, defining a noise Z as a random variable, and inputting the noise Z into a generator G of a generating countermeasure network to map the input noise Z into a generating picture G (Z);
inputting the generated picture G (Z) into a discriminator D of the generative countermeasure network to judge whether the generated picture G (Z) is close to a real picture or not, and screening the generated picture G (Z).
By adopting the technical scheme, the invention at least has the following beneficial effects:
the method has important research significance for stereoscopic matching, distance monitoring and three-dimensional reduction required by hazardous chemical monitoring, and provides powerful technical support for visual ranging in hazardous chemical storage warehouses on samples.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
Fig. 1 is a flowchart of a method for generating a dangerous chemical storage stacking image sample according to an embodiment of the present invention;
fig. 2 is a video screenshot of a simulated hazardous chemical warehouse according to an embodiment of the present invention;
fig. 3 is a diagram illustrating an angular point detection effect of the Harris algorithm according to an embodiment of the present invention;
FIG. 4 is a comparison graph of accuracy before and after denoising of a generated image sample according to an embodiment of the present invention;
fig. 5 is a schematic block diagram of a hazardous chemical storage stacking image sample generating system according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the technical solutions of the present invention will be described in detail below. It is to be understood that the described embodiments are merely exemplary of the invention, and not restrictive of the full scope of the invention. All other embodiments, which can be derived by a person skilled in the art from the examples given herein without any inventive step, are within the scope of the present invention.
The technical solution of the present invention is further described in detail by the accompanying drawings and embodiments.
Referring to fig. 1, a method for generating a dangerous chemical storage stacking image sample provided in an embodiment of the present invention includes:
step S1, acquiring real picture samples of a first number of dangerous chemical storage stacks;
step S2, performing GAN algorithm processing on the first number of real picture samples to obtain a second number of generated picture samples of dangerous chemical storage stacks, wherein the second number is larger than the first number;
step S3, performing edge feature extraction, corner detection, and noise filtering on the second number of generated picture samples, so that the sample distribution of the generated picture samples is close to the sample distribution of the real picture samples.
It can be understood that generating a deep learning model such as a Generic Adaptive Network (GAN) and generating a large sample for the next step of machine learning are one of the most promising methods for unsupervised learning in complex distribution in recent years. The generation of the countermeasure network is a countermeasure process simulating two models, one of which is a generator G (Generator), capturing the distribution of real data and generating data approximating the real distribution; the other is a discriminator D (discriminator) for judging whether the data comes from the real distribution or not and outputting the probability that the data comes from the real distribution, namely a value between 0 and 1.
The method has the advantages that according to the technical scheme, a large number of picture samples are generated by adopting a GAN algorithm, and edge-point extraction and noise filtering are performed on the generated picture samples, so that the sample distribution of the generated picture samples is close to the sample distribution of real picture samples.
Preferably, the step S1 includes:
the method comprises the steps of shooting videos of dangerous chemical storage stacks by using a binocular camera, carrying out video stream frame extraction through a prestored frame extraction program, and obtaining real picture samples of the dangerous chemical storage stacks in a first quantity.
Preferably, the step S2 includes:
step S21, taking the first quantity of dangerous chemical storage stacking picture samples as a training set, defining a noise Z as a random variable, and inputting the noise Z into a generator G of a generating countermeasure network to map the input noise Z into a generating picture G (Z);
step S22, inputting the generated picture g (z) into the discriminator D of the generative countermeasure network to judge whether the generated picture g (z) is close to the real picture, so as to filter the generated picture g (z).
It should be noted that GAN is inspired by the game theory of zero sum, and the generation problem is regarded as the game and the countermeasure of two models, i.e. the discriminator D and the generator G: the generator G produces synthetic data from a given noise (typically a uniform or normal distribution) and the discriminator D resolves the output of the generator G from the true data. The former attempts to produce data that is closer to the true one, and correspondingly, the latter attempts to more perfectly distinguish the true data from the generated data. Therefore, the two models progress in the countermeasure, and continue to resist after the progress, the data obtained by the generating network is more and more perfect and approaches the real data, so that the data (pictures, sequences, videos and the like) which are expected to be obtained can be generated.
From the perspective of probability distribution, the GAN pushes the generated distribution to the real distribution through D, and then optimizes D until the generated synthetic data is close to the real data and reaches the Nash equilibrium point, so that the generated distribution is overlapped with the real distribution, and data extremely close to the real distribution is generated.
Assuming that z is random noise and x is real data, the generating network and the discriminating network can be represented by G and D, respectively, where D can be regarded as a two-classifier, and then the cross-entropy representation can be written as:
the first term log (x) represents the judgment of the discriminator on the real data, and the second term log (1-D (G (z))) represents the synthesis and judgment of the data. Through such a Max-min game, G and D are respectively optimized to train the required generating network and the required discriminating network circularly and alternately until a Nash equilibrium point is reached.
Since most GAN-based models for picture sample generation are built with convolutional networks, convolution processes information in the local neighborhood and therefore using convolutional layers only is computationally unsuitable for modeling long-term correlation in images, the high-resolution details of conventional convolutional GAN generation are only a function of spatial local points on the low-resolution feature map. Adding an attention module improved generative countermeasure network SAGAN introduces a self-attention mechanism into the convolution GAN, and has great advantages in a multi-category image synthesis task as a complement of the convolution. In SAGAN, cues from all feature locations can be used to generate details, allowing attention-driven, long-correlation modeling of image generation tasks. In addition, the discriminator may check that highly detailed features of the distal portions of the images coincide with each other. Preferably, the present embodiment performs picture sample generation using SAGAN, so that the generated picture sample focuses more on the main item portion.
Preferably, in step S3, the corner detection is performed on the second number of generated picture samples, specifically:
and performing corner detection on the second number of generated picture samples by adopting a Harris corner detection algorithm.
Preferably, the step S3 of performing corner detection on the second number of generated picture samples includes:
and for any generated picture sample, sliding in any direction on the picture sample by using a fixed window, comparing the gray level change degrees of pixels in the window under two conditions of before sliding and after sliding, and if the sliding in any direction has large gray level change, judging that an angular point exists in the window.
Preferably, the noise filtering the second number of generated picture samples in step S3 includes:
and (3) carrying out non-maximum suppression and k-means clustering denoising on the angular points detected by the Harris angular point detection algorithm so as to remove wrong angular points detected by the Harris angular point detection algorithm caused by light irradiation, shadow and other non-stacked articles.
Preferably, the performing k-means clustering denoising on the corners detected by the Harris corner detection algorithm includes:
randomly selecting three central points on any generated picture sample;
traversing all the angular points, and dividing each angular point to the nearest central point to obtain a plurality of clusters;
calculating the average value of each cluster as a new central point;
and judging whether the clustering is converged, if so, filing a clustering result, and otherwise, determining the corner as an error corner.
Let the window centered on pixel (x, y) move u unit lengths in the x direction and v unit lengths in the y direction. Accordingly, Harris gives an analytical expression for the gray scale change metric as formula (1):
wherein E (x, y) is the gray level variation in the window; w is a window of the image; i is the gray scale of the image.
If the corner image window of the image is detected to be shifted, its autocorrelation function E (x, y) is changed accordingly. Wherein: A. b, C is an approximate expression of the directional differential of the second order equation, which can also be expressed as:
where h (x, y) represents a Gaussian smoothing filter function, X, Y represents a differential in the first order, and the difference operators between the image gray levels and x are calculated using | 10-1 | and | 10-1 |TAnd (4) showing.
Equation E (x, y) can be converted to:
E(x,y)＝[x,y]M[x,y]T (5)
if both eigenvalues of the finally obtained system M matrix are large, the extreme curvature of the image gray scale autocorrelation function in two orthogonal directions is large at the moment, that is, the point is an angular point. The gray value feature is defined on the basis of the image detected by the harris corner points, and the gray value feature is used as a matching reference.
And carrying out Harris corner detection on the image sample generated by the GAN, and carrying out non-maximum suppression and optimization of k-means clustering denoising (hereinafter referred to as LK denoising) on the traditional Harris algorithm. Clustering and layering removal using k-means for light illumination, shadows, and other non-stacking item interference
For the original Harris improved formula, n is the data volume, and I is the iteration number. The determination of the distance between the data and the center point is shown in equation (7).
X in formula (7)s，ysTo select the center point, n is the number of the angular points, xi，yiIs the position of an angular point not selected as the center point, and d is the distance from the center point. According to the derivation of the formula of the conventional corner detection and LK denoising algorithm, the formula (8) can be summarized:
where rows and col are the length and height of the picture pixels, in px. And setting kSize to mainly inhibit the frame-out phenomenon in the iterative comparison process of pixel positions. And delta is the size of the threshold value of the response function R, and the empirical value is 20000, so that error points detected by the angular points of the Harris algorithm are filtered.
In specific practice, in order to verify the practicability of the dangerous chemical storage stacking image sample generation method provided by the invention, the applicant performs tests, and the test process and the test result are as follows:
under the laboratory environment, a small stacking type dangerous chemical simulation warehouse is built, and the stacking objects are simulated by wood blocks, as shown in fig. 2. 4 stacking of 20cm by 20cm is built into a full-cargo state, a certain amount of cargos are sequentially removed, and a stacking model of 5 different cargo states in total is formed. And (3) shooting a video by using a binocular camera at the upper right of the picture 2 to acquire a small sample, and then writing a program for extracting frames. The resolution of the binocular camera is 1920 x 1080, its signal-to-noise ratio is 39db, and the pixels are 200 ten thousand. The GPU is a server of a Xeon E5-2683v3 processor and a K80 independent video card of 11G video memory. The development platform is a windows7 operating system and is realized by a Tensorflow1.12 framework.
Firstly, the GAN learning is tried by using a computer CPU, the iteration speed is about 24 seconds, and the training speed of the server CPU is similar. And training by using a server GPU, wherein the speed is about once iteration every 2.2 seconds, the speed is greatly improved, and the requirement of the sample acquisition rate can be met. The final parameter settings were as follows: the image viewing training result is generated once every 500 iterations, and recorded as a generator version every 2000 iterations for generating a batch of pictures.
After testing, the generator G can generate a picture sample with available definition after iterating for about 4500 times; 8000 iterations may result in a generated picture sample that is significantly different from known picture samples.
After 10000 iterations, the stacking sample with various and clear forms can be generated. And features of illumination, shadows, other obstacles have been learned and implemented in the sample.
The comparison between the real picture sample and the picture sample generated by the GAN shows that the characteristics of all angles of the stacked pictures of the real picture sample and the GAN are highly similar, so that the generation result of the generated countermeasure network is reliable, and a large number of reliable picture samples can be provided for subsequent monitoring processing.
The result of generating a picture sample through Harris corner detection is shown in fig. 3.
As can be seen from fig. 3, the generated map is less affected by the illumination shadow and is more affected only by the obstacle. Most of noise points mainly caused by the right obstacle can be removed by denoising the angular point LK, and the accuracy before and after denoising of the generated image is shown in fig. 4.
And (4) calculating the accuracy and the recall rate of the generated picture sample by adopting an F-measure value in the prior art as a final evaluation index. The accuracy should not exceed 1 and the larger the value the better. Recall represents a measure of coverage, with values closer to 1 being better. Aiming at the contradiction between the accuracy and the recall rate, the F-measure can comprehensively evaluate the accuracy of corner detection, and the value is as close as 1 as better. And calculating result statistics of all generated images, wherein the results are shown in a table-denoising effect and image quality evaluation table.
Watch 1
As can be seen from table one, the accuracy of generating the picture sample is lower than that of the real picture sample, but the recall rate is closer to 1 than that of the real picture sample; compared with the corner detection result after the LK algorithm is denoised, the total F-measure value shows that the deviation of the generated picture sample from 1 is 13.20%, and the deviation of the real picture sample from 1 is 15.39%, so that the generated picture of GAN can be used as a sample for image processing.
In conclusion, the quantity of samples for binocular distance measurement of dangerous chemicals storage is increased by generating the countermeasure network, 10000 new generated samples are generated by utilizing 300 pictures obtained by video streaming frame extraction through 10000 iterations, and the operability of practical application of generating the neural network is proved;
the method carries out edge feature extraction, angular point detection and LK denoising on the generated sample, and experimental results show that the velocity of the GAN for generating the sample by aiming at a large number of samples is improved by about 8 to 10 times compared with that of the video stream, and for processing algorithms such as angular point detection, LK denoising and the like, the generated sample has small deviation with real sample data, so that an effective support technology can be provided for subsequent binocular ranging and the like.
In addition, referring to fig. 5, the present invention further provides a system 100 for generating image samples of hazardous chemical substance warehouse stacks, including:
the acquiring module 101 is used for acquiring real picture samples of a first number of dangerous chemical storage stacks;
a generating module 102, configured to perform GAN algorithm processing on the first number of real image samples to obtain a second number of generated image samples of the hazardous chemical storage stacks, where the second number is greater than the first number;
a denoising module 103, configured to perform edge feature extraction, corner detection, and noise filtering on the second number of generated picture samples, so that the sample distribution of the generated picture samples is close to the sample distribution of the real picture samples.
The method has the advantages that according to the technical scheme, a large number of picture samples are generated by adopting a GAN algorithm, and edge-point extraction and noise filtering are performed on the generated picture samples, so that the sample distribution of the generated picture samples is close to the sample distribution of real picture samples.
Preferably, the obtaining module 101 is specifically configured to:
the method comprises the steps of shooting videos of dangerous chemical storage stacks by using a binocular camera, carrying out video stream frame extraction through a prestored frame extraction program, and obtaining real picture samples of the dangerous chemical storage stacks in a first quantity.
Preferably, the generating module 102 has a module for:
taking the first quantity of dangerous chemical storage stacking picture samples as a training set, defining a noise Z as a random variable, and inputting the noise Z into a generator G of a generating countermeasure network to map the input noise Z into a generating picture G (Z);
inputting the generated picture G (Z) into a discriminator D of the generative countermeasure network to judge whether the generated picture G (Z) is close to a real picture or not, and screening the generated picture G (Z).
The above description is only for the specific embodiments of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art can easily conceive of the changes or substitutions within the technical scope of the present invention, and all the changes or substitutions should be covered within the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the appended claims. The terms "first" and "second" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance. The term "plurality" means two or more unless expressly limited otherwise.
Claims (8)
1. A hazardous chemical storage stacking image sample generation method is characterized by comprising the following steps:
step S1, acquiring real picture samples of a first number of dangerous chemical storage stacks;
step S2, performing GAN algorithm processing on the first number of real picture samples to obtain a second number of generated picture samples of dangerous chemical storage stacks, wherein the second number is larger than the first number;
step S3, performing edge feature extraction, corner detection, and noise filtering on the second number of generated picture samples to make the sample distribution of the generated picture samples approximate to the sample distribution of the real picture samples;
in step S3, performing corner detection on the second number of generated picture samples, specifically:
performing corner detection on the second number of generated picture samples by adopting a Harris corner detection algorithm;
noise filtering the second number of generated picture samples in the step S3, including:
and (3) carrying out non-maximum suppression and k-means clustering denoising on the angular points detected by the Harris angular point detection algorithm so as to remove wrong angular points detected by the Harris angular point detection algorithm caused by light irradiation, shadow and other non-stacked articles.
2. The method according to claim 1, wherein the step S1 includes:
the method comprises the steps of shooting videos of dangerous chemical storage stacks by using a binocular camera, carrying out video stream frame extraction through a prestored frame extraction program, and obtaining real picture samples of the dangerous chemical storage stacks in a first quantity.
3. The method according to claim 1, wherein the step S2 includes:
step S21, taking the first quantity of dangerous chemical storage stacking picture samples as a training set, defining a noise Z as a random variable, and inputting the noise Z into a generator G of a generating countermeasure network to map the input noise Z into a generating picture G (Z);
step S22, inputting the generated picture g (z) into the discriminator D of the generative countermeasure network to judge whether the generated picture g (z) is close to the real picture, so as to filter the generated picture g (z).
4. The method according to claim 1, wherein the step S3 of performing corner detection on the second number of generated picture samples comprises:
and for any generated picture sample, sliding in any direction on the picture sample by using a fixed window, comparing the gray level change degrees of pixels in the window under two conditions of before sliding and after sliding, and if the sliding in any direction has large gray level change, judging that an angular point exists in the window.
5. The method according to claim 1, wherein the performing k-means cluster denoising on the corners detected by the Harris corner detection algorithm comprises:
randomly selecting three central points on any generated picture sample;
traversing all the angular points, and dividing each angular point to the nearest central point to obtain a plurality of clusters;
calculating the average value of each cluster as a new central point;
and judging whether the clustering is converged, if so, filing a clustering result, and otherwise, determining the corner as an error corner.
6. The utility model provides a hazardous chemicals storage stack picture sample generation system which characterized in that includes:
the acquiring module is used for acquiring real picture samples of the dangerous chemical warehouse stacks in a first quantity;
the generating module is used for carrying out GAN algorithm processing on the first number of real picture samples to obtain a second number of generated picture samples of the dangerous chemical storage stacks, wherein the second number is larger than the first number;
the de-noising module is used for performing edge feature extraction, corner detection and noise filtering on the second number of generated picture samples so as to enable the sample distribution of the generated picture samples to be close to the sample distribution of the real picture samples;
the corner detection is performed on the second number of generated picture samples, specifically:
performing corner detection on the second number of generated picture samples by adopting a Harris corner detection algorithm;
said noise filtering the second number of generated picture samples comprises:
and (3) carrying out non-maximum suppression and k-means clustering denoising on the angular points detected by the Harris angular point detection algorithm so as to remove wrong angular points detected by the Harris angular point detection algorithm caused by light irradiation, shadow and other non-stacked articles.
7. The system of claim 6, wherein the obtaining module is specifically configured to:
the method comprises the steps of shooting videos of dangerous chemical storage stacks by using a binocular camera, carrying out video stream frame extraction through a prestored frame extraction program, and obtaining real picture samples of the dangerous chemical storage stacks in a first quantity.
8. The system of claim 6, wherein the generating module has means for:
taking the first quantity of dangerous chemical storage stacking picture samples as a training set, defining a noise Z as a random variable, and inputting the noise Z into a generator G of a generating countermeasure network to map the input noise Z into a generating picture G (Z);
inputting the generated picture G (Z) into a discriminator D of the generative countermeasure network to judge whether the generated picture G (Z) is close to a real picture or not, and screening the generated 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 CN109815943A (en) | 2019-05-28 |
| CN109815943B true 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) |
Families Citing this family (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 |
| 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 |
| CN112258635B (en) * | 2020-10-26 | 2023-07-21 | 北京石油化工学院 | Three-dimensional reconstruction method and device based on improved binocular matching SAD algorithm |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
| 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 |
Family Cites Families (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102004904B (en) * | 2010-11-17 | 2013-06-19 | 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 |
| CN107563385B (en) * | 2017-09-02 | 2019-10-25 | 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 |
| CN109165735B (en) * | 2018-07-12 | 2020-06-23 | 杭州电子科技大学 | Method for generating sample picture based on generation of confrontation network and adaptive proportion |
-
2019
- 2019-03-18 CN CN201910204031.1A patent/CN109815943B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105809657A (en) * | 2014-12-30 | 2016-07-27 | Tcl集团股份有限公司 | Angular point detection method and device |
| 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 |
| 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 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109815943A (en) | 2019-05-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110532859B (en) | Remote sensing image target detection method based on deep evolution pruning convolution net | |
| CN109815943B (en) | Hazardous chemical storage stacking picture sample generation method and system | |
| CN109360232B (en) | Indoor scene layout estimation method and device based on condition generation countermeasure network | |
| CN109214337B (en) | Crowd counting method, device, equipment and computer readable storage medium | |
| CN110705376A (en) | Abnormal behavior detection method based on generative countermeasure network | |
| Raghavendra et al. | Presentation attack detection algorithm for face and iris biometrics | |
| CN106384092A (en) | Online low-rank abnormal video event detection method for monitoring scene | |
| EP3012781A1 (en) | Method and apparatus for extracting feature correspondences from multiple images | |
| US20220309635A1 (en) | Computer vision-based anomaly detection method, device and electronic apparatus | |
| CN112036381B (en) | Visual tracking method, video monitoring method and terminal equipment | |
| CN111079518A (en) | Fall-down abnormal behavior identification method based on scene of law enforcement and case handling area | |
| CN114821032A (en) | Special target abnormal state detection and tracking method based on improved YOLOv5 network | |
| CN112329764A (en) | Infrared dim target detection method based on TV-L1 model | |
| CN109919095B (en) | Monitoring method for stacking in hazardous chemical warehouse and electronic equipment | |
| CN111079572A (en) | Forest smoke and fire detection method based on video understanding, storage medium and equipment | |
| Gupta et al. | Analytical global median filtering forensics based on moment histograms | |
| Alaql et al. | Classification of image distortions for image quality assessment | |
| CN115797970B (en) | Dense pedestrian target detection method and system based on YOLOv5 model | |
| CN115205793B (en) | Electric power machine room smoke detection method and device based on deep learning secondary confirmation | |
| CN115375966A (en) | Image countermeasure sample generation method and system based on joint loss function | |
| CN116385281A (en) | Remote sensing image denoising method based on real noise model and generated countermeasure network | |
| CN109949337A (en) | Moving target detecting method and device based on Gaussian mixture model-universal background model | |
| CN115457373A (en) | Anchor-frame-free one-stage insect pest image detection method and device and storage medium | |
| CN113962900A (en) | Method, device, equipment and medium for detecting infrared dim target under complex background | |
| CN113971737A (en) | Object recognition method for robot, electronic device, medium, and program product |
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 |