CN112598034A - Ore image generation method based on generative countermeasure network and computer-readable storage medium - Google Patents
Ore image generation method based on generative countermeasure network and computer-readable storage medium Download PDFInfo
- Publication number
- CN112598034A CN112598034A CN202011462352.0A CN202011462352A CN112598034A CN 112598034 A CN112598034 A CN 112598034A CN 202011462352 A CN202011462352 A CN 202011462352A CN 112598034 A CN112598034 A CN 112598034A
- Authority
- CN
- China
- Prior art keywords
- ore
- ore image
- real
- image
- generation method
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
Abstract
The invention relates to an ore image generation method based on a generative confrontation network, which comprises the following steps: s1, acquiring a real ore image, preprocessing the real ore image to obtain real ore pictures of different types and labeling the real ore pictures; s2, constructing a condition generating type countermeasure network, and adding conditions based on the label of the selected real ore image; and S3, generating an ore image based on the condition and the selected real ore image by adopting the condition generating type confrontation network. The invention also relates to a computer-readable storage medium. By implementing the ore image generation method based on the generative confrontation network and the computer readable storage medium, a large number of vivid ore pictures can be generated aiming at the data types with smaller occupation in the unbalanced data sets in the ore samples so as to expand the training set.
Description
Technical Field
The present invention relates to the field of ore images, and more particularly, to an ore image generation method and a computer-readable storage medium based on a generative confrontation network.
Background
The ore resource is a very important natural resource in China and is an important material basis for the development of the economic society. The problems of low efficiency and poor precision exist in the field of ore separation all the time, and the problems are solved greatly and effectively due to the occurrence of deep learning. However, problems exist, such as the acquisition and training of ore data requires a large number of samples, and in addition, the identification accuracy is not enough due to the natural lack of real sample data of certain types of ores. For example, in tungsten ore classification, the number of ores with high ore particle sizes is extremely rare, so that the distribution of the ores in the collected ore image training set is very low, and the classification precision of the model can be influenced. However, due to its high value, the accuracy of such ores is subject to strict requirements, which creates a contradiction. For a specific ore image generation method, there has been no method other than taking a picture at the mine site for a while.
In the application of deep learning to sort ore pictures, there are some conventional data enhancement methods: mainly from the perspective of physical transformations and color transformations. Such as: random cropping, flipping, color dithering, adding noise, rotation, translation, etc., to the original image, a large number of similar pictures can be generated to expand the data set.
In the field of ore image generation, at present, the ore is mainly photographed and sampled on a mine site, namely data of real photographing. No method for generating an ore picture has been proposed. The true data is most reliable, but the time and the labor are defects. And the requirement is large in the training of the neural network. And a large amount of data needs to be collected again to train a new model every time the ore of a certain mine is modeled. Therefore, a large amount of manpower and material resources are consumed each time. Moreover, in some ore types with small quantity distribution, the data of the part is naturally lacked, so that the problem of unbalanced data set is caused, and therefore, the identification precision of the ore type needs to be improved when the deep learning model is trained. The data enhancement method mainly based on physical transformation can increase the generalization capability of the model to a certain extent, but the data enhancement effect is not obvious for specific ore types lacking per se.
In the application of deep learning, the problem of unbalanced training samples exists, and unbalanced training data means that the number of samples in different classes is greatly different. The consequence of this is that the accuracy of the model classification for the missing sample classes is greatly affected. The problem of this type of training sample imbalance is particularly acute in the field of ore sorting. The ore minerals are formed by the migration and aggregation of chemical elements through the processes of geological action and the like. The natural result is that the distribution of different forms of the ore produced in the mine and even the ore of the same kind have no balanced rule. Therefore, in the aspect of applying deep learning to solve the problem of ore sorting, the problem of unbalanced training samples is inevitable.
Disclosure of Invention
The present invention is directed to a method and a computer-readable storage medium for generating an ore image based on a generative confrontation network, which can generate a large number of realistic ore images for the data types with a smaller percentage of unbalanced data sets in an ore sample to expand a training set, and the method and the medium for generating an ore image based on a generative confrontation network
The technical scheme adopted by the invention for solving the technical problems is as follows: an ore image generation method based on a generative confrontation network is constructed, and comprises the following steps:
s1, acquiring a real ore image, preprocessing the real ore image to obtain real ore pictures of different types and labeling the real ore pictures;
s2, constructing a condition generating type countermeasure network, and adding conditions based on the label of the selected real ore image;
and S3, generating an ore image based on the condition and the selected real ore image by adopting the condition generating type confrontation network.
In the ore image generation method based on a generative confrontation network according to the present invention, the step S1 further includes:
s11, acquiring an actual ore picture and dividing the picture containing a plurality of ores into a real ore image only containing one ore in an image dividing mode;
s12, performing pixel whitening on the background of the real ore image;
and S13, classifying the ores according to the ore spot characteristics in the real ore image and labeling the ores according to the categories.
In the method for generating an ore image based on a generating type antagonistic network, in step S2, the condition generating type antagonistic network includes a generator and an arbiter, and the same condition is added to the generator and the arbiter, the condition is a label of a selected real ore image, and the overall loss function of the condition generating type antagonistic network is as follows:
wherein D represents a discriminator, the input of which is a real ore image x, and the output of which is 1 or 0; g represents a generator, the input of which is a one-dimensional random noise vector z, the output of which is G (z), and the training aims to make the distribution of G (z) as close as possible to the distribution p of a real ore imagedataY denotes a label of the selected real ore image, pzRepresenting the noise distribution.
In the ore image generation method based on the generative confrontation network, a plurality of Gaussian components are adopted to optimize the conditional generative confrontation network, and the overall loss function of the optimized conditional generative confrontation network is as follows:
where N is the z dimension, σiλ represents a weight as a standard deviation of the ith gaussian component.
In the ore image generation method based on the generative countermeasure network, the number of Gaussian components is more than 3.
In the ore image generation method based on the generative countermeasure network, the generator comprises a first convolution layer, a second convolution layer, a third convolution layer and a full connection layer, the convolution kernel size of the first convolution layer is 5 × 5, the adopted activation function is a Relu function and comprises 64 channels, the convolution kernel size of the second convolution layer is 5 × 5, the adopted activation function is a Relu function and comprises 128 channels, the convolution kernel size of the third convolution layer is 3 × 3, and the adopted activation function is a Tanh function and comprises 256 channels.
In the ore image generation method based on the generative countermeasure network, the discriminator comprises a first convolution layer, a second convolution layer, a third convolution layer and a full-connection layer, the convolution kernel size of the first convolution layer is 5 × 5, the adopted activation function is a Relu function and comprises 64 channels, the convolution kernel size of the second convolution layer is 3 × 3, the adopted activation function is a Relu function and comprises 128 channels, the convolution kernel size of the third convolution layer is 5 × 5, and the adopted activation function is a Sigmiod function and comprises 256 channels.
In the ore image generation method based on the generative confrontation network, the size of the real ore image is 56 × 56, the training batch is a group of 128 samples, the maximum iteration number is 1000, and an Adam optimizer is adopted in a gradient optimization algorithm.
In the method for generating an ore image based on a generative confrontation network according to the present invention, the method further comprises:
and S4, adding the ore image into a trained ore classifier to serve as a classification training set.
Another technical solution adopted by the present invention to solve the technical problem is to configure a computer-readable storage medium, on which a computer program is stored, wherein the computer program, when executed by a processor, implements the ore image generation method based on a generative confrontation network.
By implementing the ore image generation method based on the generative confrontation network and the computer readable storage medium, a large number of vivid ore pictures can be generated aiming at the data types with smaller occupation in the unbalanced data sets in the ore samples so as to expand the training set. Furthermore, the samples can be characterized by normal distribution by adding a Gaussian component, so that the diversity of the samples is increased. Further, the generated ore image is used in the ore classifier, so that the classification accuracy can be increased.
Drawings
The invention will be further described with reference to the accompanying drawings and examples, in which:
FIG. 1 is a flow chart of a first preferred embodiment of the method of ore image generation based on a generative confrontation network of the present invention;
FIGS. 2A-2D are schematic illustrations of different kinds of real ore images according to a preferred embodiment of the present invention;
FIG. 3 is a schematic diagram of a generative countermeasure network;
FIG. 4 is a schematic diagram of the structure of a condition generating countermeasure network of the preferred embodiment of the present invention;
FIG. 5 is a schematic structural diagram of a condition generating countermeasure network incorporating Gaussian components according to yet another preferred embodiment of the present invention;
FIG. 6 is a data set of real ore images prior to processing using a generative confrontation network based ore image generation method according to the present invention;
FIG. 7 is an ore image generated from the data set shown in FIG. 6 using a generative countermeasure network-based ore image generation method according to the present invention;
fig. 8 is a flowchart of a second preferred embodiment of the ore image generation method based on a generative confrontation network of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
The Generative adaptive Network (Generative adaptive Network) proposed by Goodfellow et al is a Generative model, and its main ideas are as follows: structurally inspired by two-player zero-sum games in game theory (i.e. the sum of the benefits of two players is zero, and the result of one player is just the loss of the other player), it consists of a generator G and a discriminator D. G captures a mathematical distribution model of the real data samples and generates new data samples from the learned distribution model. Fig. 3 is a schematic structural diagram of a generative countermeasure network. The optimization of GAN is actually a very small maximization problem, whose objective function is defined as:
the ore picture generation method provided by the invention is improved on the basis of generation of the countermeasure network, and a large number of vivid ore pictures can be generated aiming at the data types which are small in unbalanced data set in the ore sample by constructing the condition generation type countermeasure network and adding conditions based on the label of the selected real ore image so as to expand the training set. For example, in the invention, the problem that part of ore types are extremely rare in ore sorting can be solved by selecting conditions to generate the countermeasure network. Conditional GAN model control conditions are implemented by adding the same condition Y (e.g., a label of data) to the generator and the arbiter. Further, on the basis of the conditional generation type confrontation network, the conditional generation type confrontation network is optimized by adopting a plurality of Gaussian components, so that the samples are characterized by normal distribution, and the diversity of the samples is increased.
Fig. 1 is a flow chart of a first preferred embodiment of the ore image generation method based on a generative confrontation network of the present invention. As shown in fig. 1, in step S1, a real ore image is acquired and preprocessed to obtain different kinds of real ore pictures and labeled. In the preferred embodiment of the invention, an actual ore picture is obtained, and the picture containing a plurality of ores is divided into a real ore image only containing one ore in an image dividing mode; then carrying out pixel whitening on the background of the real ore image; and finally, classifying according to the ore spot characteristics of the ores in the real ore image, and labeling according to the classification. Preferably, as shown in fig. 2A-2D, it is a real ore image classified according to weak spots, massive black spots, massive spots, and gradual black spots, in that order.
In step S2, a condition generating countermeasure network is constructed, and a condition is added based on the label of the selected real ore image. In a preferred embodiment of the invention, after obtaining the real ore image and its specific classification, a condition generating countermeasure network can be constructed as required.
Fig. 4 is a schematic structural diagram of a condition generating countermeasure network of the preferred embodiment of the present invention. As shown in fig. 4, the condition generating countermeasure network adds the same condition Y, i.e., the label of the selected real ore image in the present application, to the generator and the discriminator. In a preferred embodiment of the invention, referring to the real ore images shown in fig. 2A-2D, we find that what is missing is a real ore image with a gradual black spot, so we again refer to it as a label. And generating an ore image with a gradual black spot through a conditional generation type antagonistic network.
In contrast to the conventional GAN, the conditional generative countermeasure network only modifies the overall loss function of the former, and the new overall loss function is of the formula
D represents a discriminator, the input of which is a real ore image x, and the output of which is 1 or 0; g represents a generator, the input of which is a one-dimensional random noise vector z, the output of which is G (z), and the training aims to make the distribution of G (z) as close as possible to the distribution p of a real ore imagedataY denotes a label of the selected real ore image, pzRepresenting the noise distribution.
Preferably, the generator G includes a first convolutional layer having a convolutional kernel size of 5 × 5, an activation function adopted is a Relu function and includes 64 channels, a second convolutional layer having a convolutional kernel size of 5 × 5, an activation function adopted is a Relu function and includes 128 channels, and a third convolutional layer having a convolutional kernel size of 3 × 3, an activation function adopted is a Tanh function and includes 256 channels. The discriminator D comprises a first convolution layer, a second convolution layer, a third convolution layer and a full-connection layer, wherein the convolution kernel size of the first convolution layer is 5 x 5, the adopted activation function is a Relu function and comprises 64 channels, the convolution kernel size of the second convolution layer is 3 x 3, the adopted activation function is a Relu function and comprises 128 channels, the convolution kernel size of the third convolution layer is 5 x 5, and the adopted activation function is a Sigmiod function and comprises 256 channels. The size of the real ore image is 56 x 56, the training batch is a group of 128 samples, the maximum iteration number is 1000, and an Adam optimizer is adopted in the gradient optimization algorithm.
With the conditional generative confrontation network described above, in step S3, we can obtain an ore image with gradual black spots, however, generator G describes the distribution of training data samples with a single distribution, resulting in insufficient reflection of sample data diversity. In order to solve the problem that the generated data sample features of training are single, the aim of enhancing the data set is difficult to achieve.
Therefore, in the preferred embodiment of the present invention, we use multiple gaussian components to optimize the conditional generative confrontation network, which essentially characterizes the samples by using multiple normal distributions, so as to constrain the diversity of the generated samples. The structure of the CGAN model with the Gaussian components added is shown in FIG. 5.
As mentioned above, the essence of Gaussian component (GMM) is to describe the diversity characteristics of the whole sample by using m (m ≧ 3) normal distributions, and after training and learning, to establish a mixed distribution model composed of m components (i.e., m normal distributions). On one hand, a multi-component mixed model can better describe the diversity characteristics of the sample, and on the other hand, the diversity of the data characteristics is restricted by each component, so that a new sample generated by the mixed model has diversity and keeps the similarity of the characteristics with the original sample. The goal of generator G in GAN is to make pdata (G (z)) the distribution as close to the sample as possible, where pdata (G (z)) is the distribution describing G (z). According to the probability multiplication formula, pdata (g (z), z) can be written as a known prior distribution density function pz (z), multiplied by pdata (g (z) | z). In combination with the above analysis, the diversity of the prior distribution is improved, so that the diversity of G (z) is improved, and the purpose of generating the diversity of the sample is achieved.
First, assume that the prior distributed density function pz (z) is a GMM with m components, while assuming the covariance matrix of each gaussian component as a diagonal matrix.
pdata(G(z))＝∫zp(G(z),z)dz＝∫zpdata(G(z)∣z)pz(z)dz
Wherein N (x; mu)i，σi) The probability density function of the gaussian component is expressed in a specific form as follows, and in the process of GAN training, since the parameter pi i cannot be optimized, pi i is set to be 1/m to simplify the calculation:
then, a one-dimensional random noise vector zz ═ mu obeying prior distribution is generated by using a repeated parameter adjusting technologyi+σiδ; delta-N (0, 1); wherein is mui、σiMean and standard deviation of the ith gaussian component. The iterative parameter tuning technique has the advantage that the parameters of the gaussian component can be considered as part of the network parameters and then trained and optimized together with the network parameters.
Derived from the above
Wherein, u ═ u1, u2, …, uN ] T, σ ═ σ 1, σ 2, …, σ N ] T, m is the number of gaussian components, and N is the dimension of z. The number of gaussian components is closely related to the diversity of the generated samples. To prevent the value of σ from becoming 0 in the experiment, an L2 regularization term for σ is added to the loss function of the generator G, the modified generator loss function being given by the equation:
in step S3, an ore image is generated based on the condition and the selected real ore image using the condition generating countermeasure network. In a preferred embodiment of the invention, the following hardware platform is selected: the processor is an Intel (R) core (TM) i5 CPU, the main frequency is 2.4GHz, the memory is 16GB, and the display card is NVIDIA GeForce GTX 2060; a software platform: a WIN 1064 bit operating system and a pytoch deep learning framework based on python. The actual ore picture of the Qibaoshan lead-zinc ore is selected, the size of the picture is 56 x 56, and the number of the picture is 8719. Partially as shown in fig. 6. An image of a portion of ore having a gradual black spot generated using the foregoing conditional generation type antagonistic network is shown in fig. 7.
Fig. 8 is a flowchart of a second preferred embodiment of the ore image generation method based on a generative confrontation network of the present invention. In the preferred embodiment shown in fig. 8, the method further comprises a step S4 of adding the ore images to a trained ore classifier as a classification training set in addition to the steps S1-S3 shown in fig. 1. In the preferred embodiment, the ore picture generated by the conditional generation type confrontation network is added into the trained ore classifier to obtain the predicted value of the test ore type, so that the data enhancement effect is achieved, and the accuracy of ore classification is obviously improved. The accuracy calculation formula is as follows.
Wherein TP, TN, FP, and FN represent true, false positive, and false negative classes, respectively. True class, true negative class, false positive class, and false negative class respectively indicate the number of positive classes judged correctly, the number of negative classes judged correctly, the number of positive classes judged incorrectly, and the number of negative classes judged incorrectly). Table 1 shows a comparison of the accuracy of the training results before and after the ore pictures generated by the condition generating type confrontation network are added, and it is found that the accuracy is significantly improved after the condition generating type confrontation network is added.
Table 1
The performance of the sorting model can be greatly improved by taking the generated ore image as a data enhancement means. And for the types lacking in the number in the unevenly distributed data sets, the condition generating type countermeasure network is used for specifically generating the rare ore image types, so that the purpose of expanding the data sets is achieved, and the classification precision of the ores is greatly improved. In addition, the ore image is generated by using the condition generating type countermeasure network and applied to the ore sorting model, so that the generalization capability of the model is improved due to the increase of training samples in addition to the expansion of the data set. Manpower and material resources for the on-site re-photographing modeling process are saved to a certain extent.
By implementing the ore image generation method based on the generative confrontation network, a large number of vivid ore images can be generated aiming at the data types with small occupation in the unbalanced data set in the ore sample so as to expand the training set. Furthermore, the samples can be characterized by normal distribution by adding a Gaussian component, so that the diversity of the samples is increased. Further, the generated ore image is used in the ore classifier, so that the classification accuracy can be increased.
Accordingly, the present invention can be realized in hardware, software, or a combination of hardware and software. The present invention can be realized in a centralized fashion in at least one computer system, or in a distributed fashion where different elements are spread across several interconnected computer systems. Any kind of computer system or other apparatus adapted for carrying out the methods of the present invention is suited. A typical combination of hardware and software could be a general purpose computer system with a computer program that, when being loaded and executed, controls the computer system such that it carries out the methods described herein.
The present invention also relates to a computer readable storage medium having stored thereon a computer program having all the features enabling the implementation of the method of the invention, when installed in a computer system. The computer program in this document refers to: any expression, in any programming language, code or notation, of a set of instructions intended to cause a system having an information processing capability to perform a particular function either directly or after either or both of the following: a) conversion to other languages, codes or symbols; b) reproduced in a different format.
By implementing the computer-readable storage medium, a large number of vivid ore pictures can be generated aiming at the data types which occupy small data in the unbalanced data set in the ore sample so as to expand the training set. Furthermore, the samples can be characterized by normal distribution by adding a Gaussian component, so that the diversity of the samples is increased. Further, the generated ore image is used in the ore classifier, so that the classification accuracy can be increased.
While the invention has been described with reference to specific embodiments, it will be understood by those skilled in the art that various changes may be made and equivalents may be substituted without departing from the scope of the invention. In addition, many modifications may be made to adapt a particular situation or material to the teachings of the invention without departing from its scope. Therefore, it is intended that the invention not be limited to the particular embodiment disclosed, but that the invention will include all embodiments falling within the scope of the appended claims.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents and improvements made within the spirit and principle of the present invention are intended to be included within the scope of the present invention.
Claims (10)
1. An ore image generation method based on a generative confrontation network, which is characterized by comprising the following steps:
s1, acquiring a real ore image, preprocessing the real ore image to obtain real ore pictures of different types and labeling the real ore pictures;
s2, constructing a condition generating type countermeasure network, and adding conditions based on the label of the selected real ore image;
and S3, generating an ore image based on the condition and the selected real ore image by adopting the condition generating type confrontation network.
2. The ore image generation method based on generative countermeasure network according to claim 1, wherein the step S1 further comprises:
s11, acquiring an actual ore picture and dividing the picture containing a plurality of ores into a real ore image only containing one ore in an image dividing mode;
s12, performing pixel whitening on the background of the real ore image;
and S13, classifying the ores according to the ore spot characteristics in the real ore image and labeling the ores according to the categories.
3. The method as claimed in claim 2, wherein in step S2, the conditional generative confrontation network comprises a generator and a discriminator, and the same condition is added to the generator and the discriminator, the condition is a label of the selected real ore image, and the overall loss function of the conditional generative confrontation network is:
wherein D represents a discriminator, the input of which is a real ore image x, and the output of which is 1 or 0; g represents a generator, the input of which is a one-dimensional random noise vector z, the output of which is G (z), and the training aims to make the distribution of G (z) as close as possible to the distribution p of a real ore imagedataY denotes a label of the selected real ore image, pzRepresenting the noise distribution.
4. The method as claimed in claim 3, wherein in step S2, the conditional generative countermeasure network is optimized by using a plurality of gaussian components, and the overall loss function of the optimized conditional generative countermeasure network is:
where N is the z dimension, σiλ represents a weight as a standard deviation of the ith gaussian component.
5. The method as claimed in claim 4, wherein the number of Gaussian components is greater than 3.
6. The generative countermeasure network-based ore image generation method according to claim 4, wherein the generator comprises a first convolutional layer having a convolution kernel size of 5 x 5, an activation function adopted as a Relu function and comprising 64 channels, a second convolutional layer having a convolution kernel size of 5 x 5, an activation function adopted as a Relu function and comprising 128 channels, a third convolutional layer having a convolution kernel size of 3 x 3, and an activation function adopted as a Tanh function and comprising 256 channels.
7. The generative countermeasure network-based ore image generation method according to claim 6, wherein the discriminator comprises a first convolution layer having a convolution kernel size of 5 x 5, an activation function of Relu function and comprising 64 channels, a second convolution layer having a convolution kernel size of 3 x 3, an activation function of Relu function and comprising 128 channels, a third convolution layer having a convolution kernel size of 5 x 5, and an activation function of Sigmiod function and comprising 256 channels.
8. The method as claimed in claim 7, wherein the size of the real ore image is 56 x 56, the training batch is a set of 128 samples, the maximum number of iterations is 1000, and the Adam optimizer is used in the gradient optimization algorithm.
9. The ore image generation method based on generative countermeasure network according to any one of claims 1 to 8, further comprising:
and S4, adding the ore image into a trained ore classifier to serve as a classification training set.
10. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, implements the method for ore image generation based on a generative confrontation network according to any one of claims 1 to 9.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011462352.0A CN112598034B (en) | 2020-12-09 | 2020-12-09 | Ore image generation method based on generative countermeasure network and computer-readable storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011462352.0A CN112598034B (en) | 2020-12-09 | 2020-12-09 | Ore image generation method based on generative countermeasure network and computer-readable storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112598034A true CN112598034A (en) | 2021-04-02 |
| CN112598034B CN112598034B (en) | 2022-10-11 |
Family
ID=75192657
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011462352.0A Active CN112598034B (en) | 2020-12-09 | 2020-12-09 | Ore image generation method based on generative countermeasure network and computer-readable storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112598034B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108564611A (en) * | 2018-03-09 | 2018-09-21 | 天津大学 | A kind of monocular image depth estimation method generating confrontation network based on condition |
| CN110046653A (en) * | 2019-03-22 | 2019-07-23 | 赣州好朋友科技有限公司 | A kind of white tungsten method for separating and system based on XRT ray |
| US20190302290A1 (en) * | 2018-03-27 | 2019-10-03 | Westerngeco Llc | Generative adversarial network seismic data processor |
| CN111582348A (en) * | 2020-04-29 | 2020-08-25 | 武汉轻工大学 | Method, device, equipment and storage medium for training condition generating type countermeasure network |
| CN111652864A (en) * | 2020-05-29 | 2020-09-11 | 沈阳铸造研究所有限公司 | Casting defect image generation method for generating countermeasure network based on conditional expression |
-
2020
- 2020-12-09 CN CN202011462352.0A patent/CN112598034B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108564611A (en) * | 2018-03-09 | 2018-09-21 | 天津大学 | A kind of monocular image depth estimation method generating confrontation network based on condition |
| US20190302290A1 (en) * | 2018-03-27 | 2019-10-03 | Westerngeco Llc | Generative adversarial network seismic data processor |
| CN110046653A (en) * | 2019-03-22 | 2019-07-23 | 赣州好朋友科技有限公司 | A kind of white tungsten method for separating and system based on XRT ray |
| CN111582348A (en) * | 2020-04-29 | 2020-08-25 | 武汉轻工大学 | Method, device, equipment and storage medium for training condition generating type countermeasure network |
| CN111652864A (en) * | 2020-05-29 | 2020-09-11 | 沈阳铸造研究所有限公司 | Casting defect image generation method for generating countermeasure network based on conditional expression |
Non-Patent Citations (2)
| Title |
|---|
| 陈文兵等: "基于条件生成式对抗网络的数据增强方法", 《计算机应用》 * |
| 魏伯阳等: "基于条件生成对抗网络的成像测井图像裂缝计算机识别", 《石油物探》 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112598034B (en) | 2022-10-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US20070065003A1 (en) | Real-time recognition of mixed source text | |
| CN111582348A (en) | Method, device, equipment and storage medium for training condition generating type countermeasure network | |
| CN109934826A (en) | A kind of characteristics of image dividing method based on figure convolutional network | |
| CN110232373A (en) | Face cluster method, apparatus, equipment and storage medium | |
| CN111401374A (en) | Model training method based on multiple tasks, character recognition method and device | |
| CN107368827A (en) | Character identifying method and device, user equipment, server | |
| CN111428557A (en) | Method and device for automatically checking handwritten signature based on neural network model | |
| CN111062036A (en) | Malicious software identification model construction method, malicious software identification medium and malicious software identification equipment | |
| CN111191789A (en) | Model training method, system, chip, electronic device and medium | |
| CN106778714A (en) | LDA face identification methods based on nonlinear characteristic and model combination | |
| CN112035345A (en) | Mixed depth defect prediction method based on code segment analysis | |
| JP3482191B2 (en) | Method for address reading | |
| US6571013B1 (en) | Automatic method for developing custom ICR engines | |
| CN112598034B (en) | Ore image generation method based on generative countermeasure network and computer-readable storage medium | |
| CN108830302B (en) | Image classification method, training method, classification prediction method and related device | |
| Thiesson et al. | Computationally efficient methods for selecting among mixtures of graphical models | |
| CN111783688A (en) | Remote sensing image scene classification method based on convolutional neural network | |
| CN116502898A (en) | Enterprise risk portrait generation method and device based on neural network | |
| US6466926B1 (en) | Method for optimizing a recognition dictionary to distinguish between patterns that are difficult to distinguish | |
| CN112686099A (en) | Automatic image-text data identification method and system suitable for power consumption business expansion application | |
| CN112257787A (en) | Image semi-supervised classification method based on generation type dual-condition confrontation network structure | |
| CN113657453B (en) | Detection method based on harmful website generating countermeasure network and deep learning | |
| CN116958954B (en) | License plate recognition method, device and storage medium based on key points and bypass correction | |
| WO1997033250A1 (en) | Method for optimizing a recognition dictionary to distinguish between patterns that are difficult to distinguish | |
| Shmurun et al. | Extraction of fingerprint orientation maps using a radial basis function recognition accelerator |
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 |