CN118135566B - Semi-supervised learning fiber master batch electron microscope image aggregation structure area identification method - Google Patents
Semi-supervised learning fiber master batch electron microscope image aggregation structure area identification method Download PDFInfo
- Publication number
- CN118135566B CN118135566B CN202410547395.0A CN202410547395A CN118135566B CN 118135566 B CN118135566 B CN 118135566B CN 202410547395 A CN202410547395 A CN 202410547395A CN 118135566 B CN118135566 B CN 118135566B
- Authority
- CN
- China
- Prior art keywords
- master batch
- electron microscope
- microscope image
- fiber master
- feature
- 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
- 239000000835 fiber Substances 0.000 title claims abstract description 95
- 239000004594 Masterbatch (MB) Substances 0.000 title claims abstract description 94
- 238000001000 micrograph Methods 0.000 title claims abstract description 81
- 230000002776 aggregation Effects 0.000 title claims abstract description 54
- 238000000034 method Methods 0.000 title claims abstract description 53
- 238000004220 aggregation Methods 0.000 title claims abstract description 49
- 230000002209 hydrophobic effect Effects 0.000 claims abstract description 4
- 238000012549 training Methods 0.000 claims description 31
- 210000002569 neuron Anatomy 0.000 claims description 15
- 238000005070 sampling Methods 0.000 claims description 15
- 239000012634 fragment Substances 0.000 claims description 12
- 230000008569 process Effects 0.000 claims description 11
- 238000012545 processing Methods 0.000 claims description 11
- 238000012937 correction Methods 0.000 claims description 8
- 230000000694 effects Effects 0.000 claims description 7
- 238000001914 filtration Methods 0.000 claims description 7
- 230000003044 adaptive effect Effects 0.000 claims description 4
- 238000013507 mapping Methods 0.000 claims description 4
- 230000006870 function Effects 0.000 claims description 3
- 238000003707 image sharpening Methods 0.000 claims description 3
- 239000011159 matrix material Substances 0.000 claims description 3
- 239000006185 dispersion Substances 0.000 abstract description 6
- 238000009776 industrial production Methods 0.000 abstract description 2
- 239000004595 color masterbatch Substances 0.000 description 9
- 238000005054 agglomeration Methods 0.000 description 5
- 239000002245 particle Substances 0.000 description 5
- VTYYLEPIZMXCLO-UHFFFAOYSA-L Calcium carbonate Chemical compound [Ca+2].[O-]C([O-])=O VTYYLEPIZMXCLO-UHFFFAOYSA-L 0.000 description 4
- 238000013473 artificial intelligence Methods 0.000 description 4
- 239000011521 glass Substances 0.000 description 4
- 238000000227 grinding Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 238000004519 manufacturing process Methods 0.000 description 3
- QAOWNCQODCNURD-UHFFFAOYSA-N Sulfuric acid Chemical compound OS(O)(=O)=O QAOWNCQODCNURD-UHFFFAOYSA-N 0.000 description 2
- 239000003963 antioxidant agent Substances 0.000 description 2
- 229910000019 calcium carbonate Inorganic materials 0.000 description 2
- 239000006229 carbon black Substances 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 239000012628 flowing agent Substances 0.000 description 2
- 238000002360 preparation method Methods 0.000 description 2
- 239000011347 resin Substances 0.000 description 2
- 229920005989 resin Polymers 0.000 description 2
- 239000004753 textile Substances 0.000 description 2
- 241001411320 Eriogonum inflatum Species 0.000 description 1
- 235000021355 Stearic acid Nutrition 0.000 description 1
- 238000005411 Van der Waals force Methods 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 230000003078 antioxidant effect Effects 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 239000012752 auxiliary agent Substances 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000007598 dipping method Methods 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 239000000975 dye Substances 0.000 description 1
- 238000000635 electron micrograph Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000010438 heat treatment Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 239000007788 liquid Substances 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 239000008204 material by function Substances 0.000 description 1
- 238000013508 migration Methods 0.000 description 1
- 230000005012 migration Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- QIQXTHQIDYTFRH-UHFFFAOYSA-N octadecanoic acid Chemical compound CCCCCCCCCCCCCCCCCC(O)=O QIQXTHQIDYTFRH-UHFFFAOYSA-N 0.000 description 1
- OQCDKBAXFALNLD-UHFFFAOYSA-N octadecanoic acid Natural products CCCCCCCC(C)CCCCCCCCC(O)=O OQCDKBAXFALNLD-UHFFFAOYSA-N 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 239000000049 pigment Substances 0.000 description 1
- 239000000843 powder Substances 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000007789 sealing Methods 0.000 description 1
- 239000008117 stearic acid Substances 0.000 description 1
- 238000003756 stirring Methods 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 150000003573 thiols Chemical class 0.000 description 1
- 238000005303 weighing Methods 0.000 description 1
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Image Processing (AREA)
Abstract
The invention belongs to the field of industrial production and intelligent recognition, and relates to a fiber master batch electron microscope image aggregation structure area recognition method for semi-supervised learning, which comprises the steps of inputting a fiber master batch electron microscope image into a trained semi-supervised convolution-graph structure network model, and outputting the fiber master batch electron microscope image with a rectangular boundary frame, wherein the rectangular boundary frame area is an aggregation structure area; the workflow of the semi-supervised convolution-graph structured network model is as follows: the method comprises the steps that firstly, a fiber master batch electron microscope image is processed to obtain a feature A and a feature B, wherein the feature A comprises a hydrophobic feature and an electrostatic feature, and the feature B comprises a texture feature and a structural feature; the second step, the feature A and the feature B are input into a fully-connected network together, and a fiber master batch electron microscope image with a rectangular boundary frame is output by the fully-connected network; the method can obtain the multidimensional key characteristics of the fiber master batch aggregation structure, and finish the accurate identification of the aggregation and dispersion structure in a self-learning mode.
Description
Technical Field
The invention belongs to the field of industrial production and intelligent identification, and relates to a fiber master batch electron microscope image aggregation structure area identification method for semi-supervised learning.
Background
The fiber master batch is an original material in the textile manufacturing process, and comprises the components of carrier resin, pigment, dye, functional powder and auxiliary agent. When a high concentration of the masterbatch (or high concentration dispersion) is in a liquid, the particles in the high concentration dispersion are close to each other due to molecular forces between the particles, which phenomenon is known as agglomeration or aggregation. The agglomeration effect is mainly caused by van der waals forces, charge interactions, repulsive shielding effects, and the like. Particle agglomeration can be dissociated in the processing process, so that the dispersion performance or evaluation of functional materials in the master batch or the fiber is a key technology, in the preparation process of the fiber master batch, a high-resolution electron microscope (scanning electron microscope, SEM) image is often used for researching the microstructure of the fiber master batch, and the electron microscope image can clearly show the form and structure of the fiber master batch, including particle size, shape, surface smoothness, internal agglomeration structure and the like; although information on various aspects on a fine-grained scale can be observed in the electron microscopy image, many of the key information therein, such as a small number of agglomerated structures (3-10 particles forming an agglomerated effect) hidden in the internal structure, may be ignored.
In recent years, a technical framework based on artificial intelligence has obtained breakthrough research results in the fields of textile image classification, chemical fiber process optimization, on-line monitoring of fiber generation and the like; the method mainly utilizes a deep neural network to learn the distribution of the characteristics of the image, such as texture, brightness, color and the like on different pixel areas, and finally uses a small number of full-connection layers to realize the classification or prediction purposes through the characteristic self-learning of a plurality of layers; however, the performance of the method depends on massive and high-quality training samples, complex model structures and complicated super-parameter adjustment to a great extent, and is still questioned to a certain extent in terms of feature migration capability, interpretability and computational complexity, especially in complex feature processing tasks such as microscopic information and high-resolution fiber master batch ultrastructure, the effective features of the data ultrastructure cannot be learned by the existing technical method, so that the final algorithm performance is unstable and even greatly reduced.
The invention of patent application CN111855587A discloses a method for judging the chromatic aberration of color master batches, firstly, weighing 0.100-0.125 g of color master batches by a balance, and adding the color master batches into a grinding bottle; adding a certain amount of sulfuric acid into the grinding bottle, sealing the bottle stopper of the grinding bottle, heating the grinding bottle on a stirrer, and stirring to completely dissolve the color master batch; then dipping color master batch with a glass rod, dissolving the color master batch on a blank glass slide, manufacturing a sample glass, placing the sample glass under a microscope for observation, randomly sampling at least three points in a mirror image, recording images on a computer matched with the microscope, and mapping 0.100mm multiplied by 0.100mm areas in the center of each image to form sampling points; and finally, calculating the average defect number of each sampling point respectively to judge the microscopic uniformity of the color master batch. The invention provides a method for judging the color difference of a color master batch, which can avoid production risk and reduce cost, but the method has the problems of high manual false detection and low recognition efficiency in the actual color master batch recognition process.
Therefore, developing an intelligent method for automatically identifying the aggregation and dispersion ultrastructure of fiber master batch has very important significance for improving identification efficiency and accuracy.
Disclosure of Invention
The invention aims to solve the technical problems in the prior art and provides a method for identifying the image aggregation structure region of a fiber master batch electron microscope for semi-supervised learning.
In order to achieve the above purpose, the technical scheme adopted by the invention is as follows:
A fiber master batch electron microscope image aggregation structure area identification method for semi-supervised learning includes inputting a fiber master batch electron microscope image into a trained semi-supervised convolution-graph structure network model, and outputting the fiber master batch electron microscope image with a rectangular boundary frame, wherein the rectangular boundary frame area is an aggregation structure area;
The workflow of the semi-supervised convolution-graph structured network model is as follows:
the method comprises the steps that firstly, a fiber master batch electron microscope image is processed to obtain a feature A and a feature B, wherein the feature A comprises a hydrophobic feature and an electrostatic feature, and the feature B comprises a texture feature and a structural feature;
The second step, the feature A and the feature B are input into a fully-connected network together, and a fiber master batch electron microscope image with a rectangular boundary frame is output by the fully-connected network;
the step of processing the fiber master batch electron microscope image to obtain the characteristic A is as follows:
(a) Combining the fiber master batch electron microscope image with a local feature mapping mode to obtain local features;
(b) Inputting local features into a graph network, outputting features therefrom ；
(C) For characteristics ofPerforming batch regularization (Batch Normalization) treatment to obtain a feature A;
The step of processing the fiber master batch electron microscope image to obtain the characteristic B is as follows:
(i) Inputting the fiber master batch electron microscope image into a convolution network, and outputting filtering characteristic information by the fiber master batch electron microscope image ；
(Ii) For filtering characteristic informationAnd carrying out self-adaptive weight correction to obtain a characteristic B.
As a preferable technical scheme:
The method for identifying the fiber master batch electron microscope image aggregation structure area for semi-supervised learning comprises the following steps: after the fiber master batch electron microscope image is divided into 4 blocks (patches), each block (patch) is converted into a feature vector, and then the local feature is obtained.
In the above-mentioned fiber master batch electron microscope image aggregation structure area identification method for semi-supervised learning, in the step (b), the structure of the graph network comprises 1 hidden layer and 1 output layer; the hidden layer has an input dimension of 128 and an output dimension of 256; the input dimension of the output layer is 256, and the output dimension is 256; adjacency matrix 16 x 16 in the network.
In the above method for identifying the fiber master batch electron microscope image aggregation structure area for semi-supervised learning, in the step (c), the batch regularization processing adopts the following formula: in which, in the process, Representing characteristicsIs used for the average value of (a),Representing characteristicsIs a function of the variance of (a),A constant is represented (the present invention sets its value to 0.0001).
In the above method for identifying the fiber master batch electron microscope image aggregation structure area for semi-supervised learning, in the step (i), the structure of the convolution network comprises 2 convolution layers and 2 downsampling layers, and each layer is numbered according to the sequence of data arrival;
The convolution kernel of the convolution layer 1 is 11 2 pixels, the number of the output feature images is 32, and the step length of convolution is 3 pixels; the convolution kernel of the convolution layer 2 is 5 2 pixels, the number of the output feature images is 64, and the step length of convolution is 2 pixels; the sampling size of the downsampling layer 1 is 2 2 pixels, and the sampling step length is 2 pixels; the downsampling layer 2 has a sampling size of 2 2 pixels and a sampling step size of 2 pixels.
In the above-mentioned method for identifying fiber master batch electron microscope image aggregation structure region for semi-supervised learning, in the step (ii), the invention adopts an artificial intelligence conventional general attention structure to filter characteristic informationAnd performing adaptive weight correction, wherein the sizes of the adaptive weights before and after correction are 256×256.
In the second step, the full-connection network comprises 3 full-connection layers, the full-connection layer 1 comprises 5184 neurons, the full-connection layer 2 comprises 5184 neurons, the full-connection layer 3 comprises 2592 neurons, wherein the full-connection layer 1 and the full-connection layer 2 are responsible for learning a target area, namely an aggregation structure area of the fiber master batch, the full-connection layer 3 generates a rectangular boundary frame in a confidence threshold mode common to the artificial intelligence field, and the position of the aggregation structure area is determined.
According to the fiber master batch electron microscope image aggregation structure region identification method for semi-supervised learning, the training process of the semi-supervised convolution-graph structure network model is that the connection weight between neurons of a convolution network and the connection weight between neurons of a graph network are adjusted, the training termination condition is that the iteration number g is equal to the maximum iteration number g max, or the number of images correctly identified by the semi-supervised convolution-graph structure network model is more than 87.04% of the total number of the input images.
The fiber master batch electron microscope image aggregation structure area identification method for semi-supervised learning comprises the following training steps of:
(I) Constructing a training set, wherein training samples in the training set are fiber master batch electron microscope images of marked aggregation structure areas;
(II) let iteration number g=1;
Randomly selecting w training samples from the training set, inputting the w training samples into a semi-supervised convolution-graph structure network model, and outputting a fiber master batch electron microscope image with a rectangular boundary frame by the semi-supervised convolution-graph structure network model;
(IV) judging whether the training termination condition is met, and if so, ending; otherwise, let g=g+1, return to step (III).
The fiber master batch electron microscope image aggregation structure area identification method for semi-supervised learning comprises the following steps of: firstly, randomly extracting image fragments with the size of 512 multiplied by 512 from an original fiber master batch electron microscope image with high resolution, marking an aggregation structure area in the image fragments by adopting a manual marking mode, then adjusting the image fragments with the probability of 30 percent in a random horizontal overturning mode, randomly adjusting contrast ratio and randomly adjusting brightness, and finally adding blurring and image sharpening effects into the image fragments.
The beneficial effects are that:
The fiber master batch electron microscope image aggregation structure area identification method for semi-supervised learning can acquire multi-dimensional key characteristics of a fiber master batch aggregation structure, and finish accurate identification of aggregation and dispersion structures in a self-learning mode; the problems of agglomeration evaluation and actual disjointing in the preparation process of the fiber master batch caused by the traditional manual characteristic design mode are avoided.
Drawings
Fig. 1 is a basic network structure diagram of a convolutional network in the present invention;
FIG. 2 is a theoretical block diagram of the fiber master batch electron microscope image aggregation structure area identification method for semi-supervised learning;
FIG. 3 is a fiber master batch electron microscope image;
fig. 4 is an effect diagram of the output of the method of the present invention after identifying fig. 3.
Detailed Description
The application is further described below in conjunction with the detailed description. It is to be understood that these examples are illustrative of the present application and are not intended to limit the scope of the present application. Furthermore, it should be understood that various changes and modifications can be made by one skilled in the art after reading the teachings of the present application, and such equivalents are intended to fall within the scope of the application as defined in the appended claims.
A method for identifying the region of a fiber master batch electron microscope image aggregation structure for semi-supervised learning comprises the following specific steps:
(1) Constructing a semi-supervised convolution-graph structure network model;
The workflow of the semi-supervised convolution-graph structured network model is shown in FIG. 2, and is specifically as follows:
(1.1) respectively processing the fiber master batch electron microscope image to obtain a feature A and a feature B, wherein the feature A comprises a hydrophobic feature and an electrostatic feature, and the feature B comprises a texture feature and a structural feature;
the step of processing the fiber master batch electron microscope image to obtain the characteristic A is as follows:
(a) Combining the fiber master batch electron microscope image with a local feature mapping mode to obtain local features, wherein the method specifically comprises the following steps: dividing a fiber master batch electron microscope image into 4 blocks, and converting each block into a feature vector to obtain local features;
(b) Inputting local features into a graph network, outputting features therefrom ; The structure of the graph network comprises 1 hidden layer and 1 output layer; the hidden layer has an input dimension of 128 and an output dimension of 256; the input dimension of the output layer is 256, and the output dimension is 256; adjacency matrix 16×16 in the graph network;
(c) For characteristics of Performing batch regularization treatment to obtain a feature A; the batch regularization process uses the formula: in which, in the process, Representing characteristicsIs used for the average value of (a),Representing characteristicsIs a function of the variance of (a),A representation constant (the present invention sets its value to 0.0001);
The step of processing the fiber master batch electron microscope image to obtain the characteristic B is as follows:
(i) Inputting the fiber master batch electron microscope image into a convolution network, and outputting filtering characteristic information by the fiber master batch electron microscope image ；
As shown in fig. 1, the structure of the convolutional network includes 2 convolutional layers and 2 downsampling layers, and each layer is numbered according to the sequence of data arrival;
The convolution kernel of the convolution layer 1 is 11 2 pixels, the number of the output feature images is 32, and the step length of convolution is 3 pixels; the convolution kernel of the convolution layer 2 is 5 2 pixels, the number of the output feature images is 64, and the step length of convolution is 2 pixels; the sampling size of the downsampling layer 1 is 2 2 pixels, and the sampling step length is 2 pixels; the sampling size of the downsampling layer 2 is 2 2 pixels, and the sampling step length is 2 pixels;
(ii) Filtering characteristic information by adopting artificial intelligence conventional and universal attention structure Performing self-adaptive weight correction to obtain a characteristic B; the self-adaptive weights before and after correction are 256×256;
(1.2) inputting the characteristic A and the characteristic B into a fully connected network together, and outputting a fiber master batch electron microscope image with a rectangular boundary box by the fully connected network;
wherein the fully-connected network comprises 3 fully-connected layers, wherein the fully-connected layer 1 comprises 5184 neurons, the fully-connected layer 2 comprises 5184 neurons, and the fully-connected layer 3 comprises 2592 neurons;
(2) Training a semi-supervised convolution-graph structure network model;
The training process of the semi-supervised convolution-graph structure network model is that the connection weight between neurons of a convolution network and the connection weight between neurons of a graph network are adjusted, and the training termination condition is that the iteration number g is equal to the maximum iteration number g max, or the percentage of the number of images correctly identified by the semi-supervised convolution-graph structure network model to the total number of the input images is more than 87.04%; the specific training process is as follows:
(1.1) constructing a training set: firstly randomly extracting image fragments with the size of 512 multiplied by 512 from an original fiber master batch electron microscope image with high resolution, marking an aggregation structure area in the image fragments by adopting a manual marking mode, then adjusting the image fragments with the probability of 30 percent, wherein the adjusting mode is random horizontal overturning, random contrast adjustment and random brightness adjustment, and finally adding blurring and image sharpening effects into the image fragments;
(1.2) let iteration number g=1;
Randomly selecting w training samples from the training set, inputting the w training samples into a semi-supervised convolution-graph structure network model, and outputting a fiber master batch electron microscope image with a rectangular bounding box by the semi-supervised convolution-graph structure network model;
(1.4) judging whether the training termination condition is met, and if so, ending; otherwise, let g=g+1, return to step (1.3);
(3) The fiber master batch electron microscope image (shown in figure 3) is input into a trained semi-supervised convolution-graph structure network model, and the fiber master batch electron microscope image (shown in figure 4) with a rectangular bounding box is output, wherein the rectangular bounding box area is an aggregation structure area.
The identification rate of the traditional convolution network to the fiber master batch electron microscope image aggregation structure area is 65%, and a plurality of groups of experiments prove that the fiber master batch electron microscope image aggregation structure area identification method based on semi-supervised learning can accurately identify the fiber master batch electron microscope image aggregation structure area, and the identification rate is 85%;
The identification rate=the number of correctly identified fiber master batch electron microscope images/the total number of fiber master batch electron microscope images×100%, wherein for the same fiber master batch electron microscope image, the invention adopts a manual marking mode to mark an aggregation structure area, and adopts a traditional convolution network or the identification method to obtain the aggregation structure area, wherein the two are the same, namely the fiber master batch electron microscope images are correctly identified, the total number of the fiber master batch electron microscope images is 40, the 40 fiber master batch electron microscope images and the FIG. 3 are electron microscope images of the same fiber master batch with different sections, the fiber master batch consists of PET resin, carbon black, calcium carbonate, antioxidants (thiols) and flowing agents (stearic acid), wherein the carbon black content is 20wt%, the calcium carbonate content is 35wt%, the antioxidant content is 0.01wt%, and the flowing agents content is 0.01wt%.
Claims (10)
1. A recognition method of a fiber master batch electron microscope image aggregation structure area for semi-supervised learning is characterized in that a fiber master batch electron microscope image is input into a trained semi-supervised convolution-graph structure network model, and a fiber master batch electron microscope image with a rectangular boundary frame is output by the fiber master batch electron microscope image, wherein the rectangular boundary frame area is the aggregation structure area;
The workflow of the semi-supervised convolution-graph structured network model is as follows:
the method comprises the steps that firstly, a fiber master batch electron microscope image is processed to obtain a feature A and a feature B, wherein the feature A comprises a hydrophobic feature and an electrostatic feature, and the feature B comprises a texture feature and a structural feature;
The second step, the feature A and the feature B are input into a fully-connected network together, and a fiber master batch electron microscope image with a rectangular boundary frame is output by the fully-connected network;
the step of processing the fiber master batch electron microscope image to obtain the characteristic A is as follows:
(a) Combining the fiber master batch electron microscope image with a local feature mapping mode to obtain local features;
(b) Inputting local features into a graph network, outputting features therefrom ；
(C) For characteristics ofPerforming batch regularization treatment to obtain a feature A;
The step of processing the fiber master batch electron microscope image to obtain the characteristic B is as follows:
(i) Inputting the fiber master batch electron microscope image into a convolution network, and outputting filtering characteristic information by the fiber master batch electron microscope image ；
(Ii) For filtering characteristic informationAnd carrying out self-adaptive weight correction to obtain a characteristic B.
2. The method for identifying the fiber master batch electron microscope image aggregation structure area for semi-supervised learning according to claim 1, wherein the step (a) is specifically as follows: after the fiber master batch electron microscope image is divided into 4 blocks, each block is converted into a feature vector, and then the local feature is obtained.
3. The method for identifying the fiber master batch electron microscope image aggregation structure area for semi-supervised learning according to claim 1, wherein in the step (b), the structure of the graph network comprises 1 hidden layer and 1 output layer; the hidden layer has an input dimension of 128 and an output dimension of 256; the input dimension of the output layer is 256, and the output dimension is 256; adjacency matrix 16 x 16 in the network.
4. The method for identifying the fiber master batch electron microscope image aggregation structure area for semi-supervised learning according to claim 1, wherein in the step (c), a formula adopted by batch regularization processing is as follows: in which, in the process, Representing characteristicsIs used for the average value of (a),Representing characteristicsIs a function of the variance of (a),Representing a constant.
5. The method for identifying the fiber master batch electron microscope image aggregation structure area for semi-supervised learning according to claim 1, wherein in the step (i), the structure of a convolution network comprises 2 convolution layers and 2 downsampling layers, and each layer is numbered according to the sequence of data arrival;
The convolution kernel of the convolution layer 1 is 11 2 pixels, the number of the output feature images is 32, and the step length of convolution is 3 pixels; the convolution kernel of the convolution layer 2 is 5 2 pixels, the number of the output feature images is 64, and the step length of convolution is 2 pixels; the sampling size of the downsampling layer 1 is 2 2 pixels, and the sampling step length is 2 pixels; the downsampling layer 2 has a sampling size of 2 2 pixels and a sampling step size of 2 pixels.
6. The method for identifying the areas of the fiber master batch electron microscope image aggregation structure for semi-supervised learning according to claim 1, wherein in the step (ii), the attention structure is adopted for filtering characteristic informationAnd performing adaptive weight correction, wherein the sizes of the adaptive weights before and after correction are 256×256.
7. The method for identifying the fiber master batch electron microscope image aggregation structure area for semi-supervised learning according to claim 1, wherein in the second step, the full-connection network comprises 3 full-connection layers, the full-connection layer 1 comprises 5184 neurons, the full-connection layer 2 comprises 5184 neurons, and the full-connection layer 3 comprises 2592 neurons.
8. The method for identifying the fiber master batch electron microscope image aggregation structure area according to claim 1, wherein the training process of the semi-supervised convolution-graph structure network model is a process of adjusting connection weights among neurons of a convolution network and connection weights among neurons of a graph network, and the training termination condition is that the iteration number g is equal to the maximum iteration number g max or that the number of images correctly identified by the semi-supervised convolution-graph structure network model accounts for more than 87.04% of the total number of input images.
9. The method for identifying the fiber master batch electron microscope image aggregation structure area for semi-supervised learning according to claim 8, wherein the training steps of the semi-supervised convolution-graph structure network model are as follows:
(I) Constructing a training set, wherein training samples in the training set are fiber master batch electron microscope images of marked aggregation structure areas;
(II) let iteration number g=1;
Randomly selecting w training samples from the training set, inputting the w training samples into a semi-supervised convolution-graph structure network model, and outputting a fiber master batch electron microscope image with a rectangular boundary frame by the semi-supervised convolution-graph structure network model;
(IV) judging whether the training termination condition is met, and if so, ending; otherwise, let g=g+1, return to step (III).
10. The method for identifying the fiber master batch electron microscope image aggregation structure area for semi-supervised learning according to claim 9, wherein the training set is constructed by the following steps: firstly randomly extracting image fragments with the size of 512 multiplied by 512 from an original fiber master batch electron microscope image, marking an aggregation structure area in the image fragments by adopting a manual marking mode, then adjusting the image fragments with the probability of 30 percent, wherein the adjusting mode is random horizontal overturning, random contrast adjustment and random brightness adjustment, and finally adding blurring and image sharpening effects into the image fragments.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410547395.0A CN118135566B (en) | 2024-05-06 | 2024-05-06 | Semi-supervised learning fiber master batch electron microscope image aggregation structure area identification method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410547395.0A CN118135566B (en) | 2024-05-06 | 2024-05-06 | Semi-supervised learning fiber master batch electron microscope image aggregation structure area identification method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN118135566A CN118135566A (en) | 2024-06-04 |
| CN118135566B true CN118135566B (en) | 2024-07-02 |
Family
ID=91246128
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410547395.0A Active CN118135566B (en) | 2024-05-06 | 2024-05-06 | Semi-supervised learning fiber master batch electron microscope image aggregation structure area identification method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN118135566B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111832650A (en) * | 2020-07-14 | 2020-10-27 | 西安电子科技大学 | Image classification method based on generation of confrontation network local aggregation coding semi-supervision |
| AU2020103901A4 (en) * | 2020-12-04 | 2021-02-11 | Chongqing Normal University | Image Semantic Segmentation Method Based on Deep Full Convolutional Network and Conditional Random Field |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112801212B (en) * | 2021-03-02 | 2022-11-18 | 东南大学 | White blood cell classification counting method based on small sample semi-supervised learning |
| CN113469356B (en) * | 2021-06-03 | 2024-06-07 | 江苏大学 | Improved VGG16 network pig identity recognition method based on transfer learning |
-
2024
- 2024-05-06 CN CN202410547395.0A patent/CN118135566B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111832650A (en) * | 2020-07-14 | 2020-10-27 | 西安电子科技大学 | Image classification method based on generation of confrontation network local aggregation coding semi-supervision |
| AU2020103901A4 (en) * | 2020-12-04 | 2021-02-11 | Chongqing Normal University | Image Semantic Segmentation Method Based on Deep Full Convolutional Network and Conditional Random Field |
Also Published As
| Publication number | Publication date |
|---|---|
| CN118135566A (en) | 2024-06-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110232675B (en) | Texture surface defect detection and segmentation device and method in industrial environment | |
| CN107341499B (en) | Fabric defect detection and classification method based on unsupervised segmentation and ELM | |
| CN111325152B (en) | Traffic sign recognition method based on deep learning | |
| CN112862811A (en) | Material microscopic image defect identification method, equipment and device based on deep learning | |
| WO2021007744A1 (en) | Kernel fuzzy c-means fast clustering algorithm with integrated spatial constraints | |
| CN109800754A (en) | A kind of ancient character body classification method based on convolutional neural networks | |
| CN101980298B (en) | Multi-agent genetic clustering algorithm-based image segmentation method | |
| CN109376798B (en) | Titanium dioxide lattice phase classification method based on convolutional neural network | |
| CN111860596B (en) | Unsupervised pavement crack classification method and model building method based on deep learning | |
| CN109409438B (en) | Remote sensing image classification method based on IFCM clustering and variational inference | |
| CN111368742A (en) | Double-yellow traffic marking reconstruction identification method and system based on video analysis | |
| CN115937626A (en) | Automatic generation method of semi-virtual data set based on instance segmentation | |
| CN113392854A (en) | Image texture feature extraction and classification method | |
| US11908118B2 (en) | Visual model for image analysis of material characterization and analysis method thereof | |
| CN118135566B (en) | Semi-supervised learning fiber master batch electron microscope image aggregation structure area identification method | |
| CN113538342B (en) | Convolutional neural network-based aluminum aerosol can coating quality detection method | |
| CN117333655A (en) | Small sample target detection method based on super network and condition aggregation | |
| CN113221814A (en) | Road traffic sign identification method, equipment and storage medium | |
| CN116310358B (en) | Method, storage medium and equipment for detecting bolt loss of railway wagon | |
| CN102737232B (en) | Cleavage cell recognition method | |
| CN116468083A (en) | Transformer-based network generation countermeasure method | |
| Zhang et al. | A deep learning-based approach for the automatic measurement of laser-cladding coating sizes | |
| CN110533098B (en) | Method for identifying loading type of green traffic vehicle compartment based on convolutional neural network | |
| Wu et al. | Rapid detection of defect structures in graphene by the machine learning | |
| CN113642429B (en) | Marine fish identification method based on TPP-TCCNN |
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 |