CN113792666A - Concrete classification method and system based on scanning electron microscope images - Google Patents
Concrete classification method and system based on scanning electron microscope images Download PDFInfo
- Publication number
- CN113792666A CN113792666A CN202111085821.6A CN202111085821A CN113792666A CN 113792666 A CN113792666 A CN 113792666A CN 202111085821 A CN202111085821 A CN 202111085821A CN 113792666 A CN113792666 A CN 113792666A
- Authority
- CN
- China
- Prior art keywords
- electron microscope
- scanning electron
- network
- concrete
- training
- 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
- 239000004567 concrete Substances 0.000 title claims abstract description 115
- 238000001000 micrograph Methods 0.000 title claims abstract description 80
- 238000000034 method Methods 0.000 title claims abstract description 52
- 238000012549 training Methods 0.000 claims abstract description 101
- 238000000605 extraction Methods 0.000 claims abstract description 31
- 238000012360 testing method Methods 0.000 claims abstract description 17
- 238000011176 pooling Methods 0.000 claims description 16
- 238000005498 polishing Methods 0.000 claims description 6
- 238000012935 Averaging Methods 0.000 claims description 4
- 238000005303 weighing Methods 0.000 claims description 2
- 238000001878 scanning electron micrograph Methods 0.000 description 19
- 238000004458 analytical method Methods 0.000 description 8
- 238000010586 diagram Methods 0.000 description 8
- 230000006870 function Effects 0.000 description 6
- 230000000694 effects Effects 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 4
- 238000013508 migration Methods 0.000 description 3
- 230000005012 migration Effects 0.000 description 3
- 239000000203 mixture Substances 0.000 description 3
- 238000013526 transfer learning Methods 0.000 description 3
- 238000013459 approach Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000004626 scanning electron microscopy Methods 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- 238000009825 accumulation Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 238000004451 qualitative analysis Methods 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 238000004381 surface treatment Methods 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
Images
Classifications
-
- 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
- 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
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- 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
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Molecular Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Probability & Statistics with Applications (AREA)
- Analysing Materials By The Use Of Radiation (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to a concrete classification method and a system based on scanning electron microscope images, wherein the method comprises the following steps: acquiring images of a plurality of concrete samples by using a scanning electron microscope to obtain a plurality of scanning electron microscope images, and constructing an original data set; the raw data set includes a number of the scanning electron microscope images and a truth class corresponding to the scanning electron microscope images; dividing an original data set into a training set and a testing set; training the deep convolutional network by using the training set to obtain a trained deep convolutional network; the deep convolutional network comprises a feature extraction network and a feature classification network; and inputting the concrete sample to be tested into the trained deep convolution network, and determining the category of the concrete sample to be tested. The method improves the accuracy and universality of concrete classification.
Description
Technical Field
The invention relates to the field of concrete classification, in particular to a concrete classification method and system based on scanning electron microscope images.
Background
The microstructure of concrete is closely related to performance, and the characterization of the microstructure often depends on Scanning Electron Microscopy (SEM). At present, the application of SEM images is generally limited to qualitative analysis, such as observation of microstructure morphology and chemical component distribution of concrete, and the analysis has great dependence on the experience of operators, needs strong professional knowledge accumulation, and results are easily influenced by the subjectivity of the operators. On the other hand, according to the past experience of operators, all the microstructure information hidden in the SEM images is not enough to be analyzed, and it is difficult to establish the connection between the microscopic and macroscopic properties, and to classify the concrete according to the concrete SEM images.
With the development of image processing technology and machine learning, researchers have proposed different algorithms and systems for concrete SEM image analysis. However, most of the methods have obvious disadvantages, and rely on specially designed artificial features to a great extent, the artificial features are analyzed and compared with the concrete SEM images to obtain concrete categories, and the concrete categories cannot be included due to subjective artificial features, so that the concrete categories are inaccurate, and the existing concrete classification method has the problem of low universality.
Disclosure of Invention
The invention aims to provide a concrete classification method and system based on a scanning electron microscope image, and aims to solve the problems that the concrete classification method in the prior art is inaccurate in classification and low in universality.
In order to achieve the purpose, the invention provides the following scheme:
a concrete classification method based on a scanning electron microscope image comprises the following steps:
acquiring images of a plurality of concrete samples by using a scanning electron microscope to obtain a plurality of scanning electron microscope images, and constructing an original data set; the raw data set comprises a plurality of scanning electron microscope images and real values corresponding to the scanning electron microscope images; dividing an original data set into a training set and a testing set;
training the deep convolutional network by using the training set to obtain a trained deep convolutional network; the deep convolutional network comprises a feature extraction network and a feature classification network;
and inputting the concrete sample to be tested into the trained deep convolution network, and determining the category of the concrete sample to be tested.
Optionally, the acquiring the images of the plurality of concrete samples by using the scanning electron microscope further includes, before the obtaining of the images of the plurality of scanning electron microscopes:
cutting the concrete block to obtain a plurality of concrete samples;
and polishing the plurality of concrete samples to obtain a plurality of smooth concrete samples.
Optionally, the training of the deep convolutional network by using the training set is performed to obtain a trained deep convolutional network, which specifically includes:
inputting the scanning electron microscope images in the training set into a deep convolution network in batches to obtain a first prediction result; the first prediction result is the probability of the classification to which the scanning electron microscope image in the training set belongs;
comparing the first prediction result with the real value, and outputting a first comparison result in a cross entropy function mode;
according to the first comparison result, the difference between the first prediction result and the true value is measured, whether the training reaches the maximum iteration times is judged, and a first judgment result is obtained;
if the first judgment result is that the training reaches the maximum iteration number, finishing the training and outputting a trained deep convolutional network; and if the first judgment result is that the training does not reach the maximum iteration number, adjusting the weight of the deep convolution network by using a back propagation algorithm to obtain an adjusted deep convolution network, taking the adjusted deep convolution network as the deep convolution network, and returning to the step of inputting the scanning electron microscope images in the training set into the deep convolution network in batches to obtain a first prediction result of a first prediction type.
Optionally, the inputting the scanning electron microscope images in the training set into a deep convolutional network in batches to obtain a first prediction result specifically includes:
inputting the scanning electron microscope image into the feature extraction network, and extracting the image features of the scanning electron microscope image;
and inputting the image features into the feature classification network, and outputting a first prediction result.
Optionally, after the deep convolutional network is trained by using the training set to obtain a trained deep convolutional network, the method further includes:
inputting the scanning electron microscope images in the test set into the trained deep convolution network in sequence, and outputting a second prediction result; the second prediction result is the probability of the classification to which the scanning electron microscope image in the test set belongs;
comparing the second prediction result with the real value, and judging whether the second prediction result is correct;
counting the number of correct second prediction results and the number of wrong second prediction results;
calculating the prediction accuracy rate and the recall rate according to the number of correct second prediction results and the number of wrong second prediction results;
using formulas
Weighing the trained deep convolutional network; wherein, F1score is a measure of the trained deep convolutional network.
A concrete classification system based on scanning electron microscope images comprises:
the image acquisition module is used for acquiring images of a plurality of concrete samples by using a scanning electron microscope to obtain a plurality of scanning electron microscope images and construct an original data set; the raw data set includes a number of the scanning electron microscope images and a truth class corresponding to the scanning electron microscope images; dividing an original data set into a training set and a testing set;
the training module is used for training the deep convolutional network by utilizing the training set to obtain a trained deep convolutional network; the deep convolutional network comprises a feature extraction network and a feature classification network;
and the classification module is used for inputting the concrete sample to be detected into the trained deep convolution network and determining the category of the concrete sample to be detected.
Optionally, the method further includes: a sample acquisition module; the sample acquisition module specifically includes:
the sample cutting submodule is used for cutting the concrete block to obtain a plurality of concrete samples;
and the sample grinding submodule is used for polishing a plurality of concrete samples to obtain a plurality of smooth concrete samples.
Optionally, the training module specifically includes:
the prediction submodule is used for inputting the scanning electron microscope images in the training set into a deep convolution network in batches to obtain a first prediction result; the first prediction result is the probability of the classification to which the scanning electron microscope image in the training set belongs;
the comparison submodule is used for comparing the first prediction result with the real value and outputting a first comparison result in a cross entropy function mode;
the judgment submodule is used for measuring the difference between the first prediction result and the true value according to the first comparison result and judging whether the training reaches the maximum iteration number or not to obtain a first judgment result;
the output submodule is used for finishing the training and outputting the trained deep convolutional network if the first judgment result is that the training reaches the maximum iteration times;
and the return submodule is used for adjusting the weight of the deep convolution network by using a back propagation algorithm to obtain an adjusted deep convolution network if the first judgment result is that the training does not reach the maximum iteration number, taking the adjusted deep convolution network as the deep convolution network, and returning to the step of inputting the scanning electron microscope images in the training set into the deep convolution network in batches to obtain a first prediction result.
Optionally, the feature extraction network includes 5 feature extraction sub-networks connected in sequence;
the feature extraction sub-network comprises a convolution layer, a nonlinear layer and a regularization layer which are connected in sequence, and a convolution layer, a nonlinear layer, a regularization layer and a pooling layer which are connected in sequence;
the convolution layer is used for extracting image characteristics of the scanning electron microscope image;
the nonlinear layer is used for increasing the nonlinearity of the feature extraction network;
the regularization layer is used for averaging the image features into 0 and variance into 1;
the pooling layer is used to reduce the dimensionality of the image features.
Optionally, the feature classification network includes a first fully-connected layer, a nonlinear layer, a random discard layer, a second fully-connected layer, and a classification layer, which are connected in sequence;
the first full connection layer and the second full connection layer are stacked and used for classifying the image features output by the feature extraction network;
the non-linear layer is used for increasing the non-linearity of the feature classification network;
the random discarding layer is used for randomly discarding any parameter in the first full connection layer and the second full connection layer in proportion;
the classification layer is used for normalizing the classification result output by the full connection layer and outputting a first prediction result in a probability form.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects:
the method comprises the steps of obtaining a scanning electron microscope image of a concrete sample, utilizing the scanning electron microscope image to train a deep convolution network, adopting the deep convolution network to slide from the image to extract features in the training process, automatically correcting the features to further obtain the trained deep convolution network, and utilizing the trained deep convolution network to extract and classify the features of the concrete. The method can classify all types of concrete without comparing specially designed artificial features with the concrete SEM image, thereby improving the accuracy and the universality of the concrete classification method.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without inventive exercise.
FIG. 1 is a flowchart of a concrete classification method based on scanning electron microscope images according to the present invention;
FIG. 2 is a flow chart of training a deep convolutional network provided by the present invention;
FIG. 3 is a flow chart of the verification of a trained deep convolutional network provided by the present invention;
FIG. 4 is a block diagram of a concrete classification system based on scanning electron microscope images according to the present invention;
FIG. 5 is a block diagram of a deep convolutional network provided by the present invention;
FIG. 6 is a schematic diagram of training and verifying a deep convolutional network in practical application;
FIG. 7 is a diagram illustrating training, validation, and transfer learning of a deep convolutional network in practical application of the present invention.
Description of the symbols: 1-a convolutional layer; RELU-nonlinear layer; BN-regularization layer; a POOL-pooling layer; FC-full connectivity layer; DROP-random discard layer; SOFTMAX-classification layer.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The invention aims to provide a concrete classification method and system based on a scanning electron microscope image, and aims to solve the problems that the concrete classification method in the prior art is inaccurate in classification and low in universality.
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.
The invention belongs to the field of concrete material performance analysis, and discloses a method for analyzing a concrete microstructure image by using an image processing technology and a deep learning method.
The invention provides an efficient and accurate analysis or classification method for the analysis and classification of the concrete SEM images, and the concrete with different mix proportions can be classified based on the SEM images.
The invention mainly aims to provide a set of deep convolution network and a training method thereof, and a method for classifying SEM images of concrete samples with different mix proportions by using the deep convolution network.
The invention applies the computer vision technology, namely the convolutional neural network to the identification of the microscopic electron microscope photo of the concrete, thereby realizing the classification of the concrete with different mixing proportions according to the microstructure of the concrete. In order to train the deep convolutional network by using the concrete microscopic scanning electron microscope picture, a concrete sample needs to be processed by a specific method so as to obtain a proper microscopic picture.
Fig. 1 is a flowchart of a concrete classification method based on a scanning electron microscope image, as shown in fig. 1, the method includes:
step 101: and acquiring images of the concrete samples by using a scanning electron microscope to obtain a plurality of scanning electron microscope images, and constructing an original data set. The raw data set includes a number of the scanning electron microscope images and a truth class corresponding to the scanning electron microscope images; the raw data set is divided into a training set and a test set.
In a specific embodiment, before the step 101, the method further includes:
cutting the concrete block to obtain a plurality of concrete samples; and polishing the plurality of concrete samples to obtain a plurality of smooth concrete samples.
Step 102: and training the deep convolutional network by using the training set to obtain the trained deep convolutional network. The deep convolutional network comprises a feature extraction network and a feature classification network.
In a specific embodiment, fig. 2 is a flowchart of training a deep convolutional network provided in the present invention, and as shown in fig. 2, the step 102 specifically includes:
step 201: and inputting the scanning electron microscope images in the training set into a deep convolution network in batches to obtain a first prediction result. The first prediction result is a probability of a classification to which the scanning electron microscope image in the training set belongs.
Step 202: and comparing the first prediction result with the real value, and outputting a first comparison result in a cross entropy function mode.
Step 203: and measuring the difference between the first prediction result and the true value according to the first comparison result and judging whether the training reaches the maximum iteration number to obtain a first judgment result.
Step 204: and if the first judgment result is that the training reaches the maximum iteration number, finishing the training and outputting the well-trained deep convolutional network. If the first judgment result is that the training does not reach the maximum iteration number, the weight of the deep convolutional network is adjusted by using a back propagation algorithm to obtain an adjusted deep convolutional network, the adjusted deep convolutional network is used as the deep convolutional network, and the step 201 is returned.
The step 201 specifically includes:
inputting the scanning electron microscope image into the feature extraction network, and extracting the image features of the scanning electron microscope image; and inputting the image features into the feature classification network, and outputting a first prediction result.
Step 103: and inputting the concrete sample to be tested into the trained deep convolution network, and determining the category of the concrete sample to be tested.
In a specific embodiment, after the step 102, a process of verifying the trained deep convolutional network is further included, and fig. 3 is a flowchart of verifying the trained deep convolutional network provided by the present invention, as shown in fig. 3, the verification process is as follows:
step 301: inputting the scanning electron microscope images in the test set into the trained deep convolution network in sequence, and outputting a second prediction result; the second prediction result is a probability of the classification to which the scanning electron microscope image in the test set belongs.
Step 302: and comparing the second prediction result with the real value to judge whether the second prediction result is correct. Specifically, the label with the highest probability in the second prediction result is set to 1, and the other labels are set to 0. The true value label is 1, whether the label of the prediction result is consistent with the label of the true value is judged, and if so, the second prediction result is correct; and if the two prediction results are not consistent, the second prediction result is wrong.
Step 303: and counting the number of correct second prediction results and the number of wrong second prediction results.
Step 304: and calculating the prediction accuracy and the recall rate according to the correct number of the second prediction results and the wrong number of the second prediction results.
Step 305: using formulas
And measuring the trained deep convolutional network. Wherein, F1score is a measure of the trained deep convolutional network.
Fig. 6 is a schematic diagram of training and verifying a deep convolutional network in practical application, and as shown in fig. 6, a concrete sample to be analyzed is firstly cut, then the surface of the concrete sample is polished to a mirror effect, and finally an image is acquired through a scanning electron microscope under a specific magnification and resolution. The resulting SEM images are collected and then the unwanted boundaries are cut away, and the true classes corresponding to the SEM images constitute the original dataset needed for training. The raw data set was calculated as 2: the ratio of 8 is randomly divided into a test set and a training set, the training set is used for training the deep convolutional network, and the trained deep convolutional network verifies the accuracy of the method on the test set.
For the training phase of the deep convolutional network, parameters in the deep convolutional network are adjusted by using a back propagation algorithm. SEM images in the training set are input into the network in batches, the prediction categories output by the feature classification network are compared with the real categories, the difference between the prediction categories and the real categories is returned in a cross entropy mode, and then the value of each parameter of the deep convolutional network which needs to be adjusted is calculated according to back propagation, so that the difference is reduced, and the smaller the difference is, the more accurate the prediction is. This process is repeated and the training phase ends when the difference between the predicted and actual classes no longer changes significantly. In the testing stage, the SEM images of the test set are sequentially input into the deep convolutional network without back propagation, namely the parameters of the deep convolutional network are adjusted, the accuracy and the recall ratio are respectively calculated by counting the number of correct predictions and wrong predictions, finally, the quality of the trained deep convolutional network is measured according to F1score, and the higher the F1score is, the better the performance of the trained deep convolutional network is.
F1score can be calculated by the following formula:
fig. 7 is a schematic diagram of training, verifying, and migration learning of a deep convolutional network in practical application of the present invention, and as shown in fig. 7, by applying the classification system proposed by the present invention to other component or process samples, through the application of migration learning, a deep convolutional network trained on one type of sample can be applied to another type of sample, even though the components and processes of the two types of samples are different. Firstly, two concrete samples with different components are collected to form two different original data sets, namely a first original data set and a second original data set, and then a deep convolutional network trained on the first original data set is put on the second original data set in a supervised learning mode for training, so that the advantages of the training comprise: first, such an approach may reduce training time compared to training from zero using only the second raw data set, and the parameters of the resulting deep convolutional network on the first raw data set may be treated as good priors, thereby reducing the time for the training to reach stability; second, such migration learning may improve the accuracy of the model compared to using only the second raw data set. Because the two raw data sets expand the data volume of the training set compared to using only the second raw data set, more samples are provided for training, reducing the impact of overfitting.
The analysis method based on the deep network provided by the invention avoids the influence of subjective factors. The convolution layer is adopted to slide from the SEM image to extract image characteristics, so that the influence caused by image noise or stretching rotation is reduced, and the robustness of the algorithm is enhanced. The deep convolutional network is trained in a back propagation mode, parameters of the deep convolutional network are automatically corrected without manually designing features, and therefore the universality of the algorithm is improved.
In a first aspect of the invention, a concrete SEM image analysis method based on a deep convolutional network is provided, and comprises sample collection, surface treatment, image acquisition, network training and sample analysis.
In a second aspect of the invention, a concrete SEM image classification method based on a deep convolutional network is provided, where the deep convolutional network includes a feature extraction network and a feature classification network. And inputting the collected SEM image into a feature extraction network to obtain corresponding image features, inputting the image features into a feature classification network, and outputting a classification result.
In a third aspect of the invention, the general utility of the proposed deep convolutional network is enhanced by applying the network trained on a specific concrete sample dataset to different concrete samples by means of a transfer learning method. Traditional artificial intelligence image recognition requires tens of thousands of pictures for training. This is not practical for scanning electron microscope pictures. By utilizing transfer learning, the foundation of previous training can be kept, so that the new training effect is better, and a good effect is achieved by using less pictures.
At present, no traditional technology capable of classifying concrete according to concrete microscopic images is available.
Fig. 4 is a system block diagram of a concrete classification system based on a scanning electron microscope image, as shown in fig. 4, including:
the image acquisition module 401 is configured to acquire images of a plurality of concrete samples by using a scanning electron microscope to obtain a plurality of scanning electron microscope images, and construct an original data set; the raw data set includes a number of the scanning electron microscope images and a truth class corresponding to the scanning electron microscope images; dividing an original data set into a training set and a testing set;
and the training module 402 is configured to train the deep convolutional network by using the training set to obtain a trained deep convolutional network.
And the classification module 403 is configured to input the concrete sample to be tested into the trained deep convolutional network, and determine the category of the concrete sample to be tested.
In one embodiment, the method further comprises: a sample acquisition module; the sample acquisition module specifically includes:
and the sample cutting submodule is used for cutting the concrete block to obtain a plurality of concrete samples.
And the sample grinding submodule is used for polishing a plurality of concrete samples to obtain a plurality of smooth concrete samples.
In a specific embodiment, the training module 402 specifically includes:
and the prediction submodule is used for inputting the scanning electron microscope images in the training set into the deep convolution network in batches to obtain a first prediction result.
And the comparison submodule is used for comparing the first prediction result with the real value and outputting a first comparison result in a cross entropy function mode.
And the judgment submodule is used for measuring the difference between the first prediction result and the true value according to the first comparison result and judging whether the training reaches the maximum iteration number or not to obtain a first judgment result.
And the output submodule is used for finishing the training and outputting the trained deep convolutional network if the first judgment result is that the training reaches the maximum iteration number.
And the return submodule is used for adjusting the weight of the deep convolution network by using a back propagation algorithm to obtain an adjusted deep convolution network if the first judgment result is that the training does not reach the maximum iteration number, taking the adjusted deep convolution network as the deep convolution network, and returning to the step of inputting the scanning electron microscope images in the training set into the deep convolution network in batches to obtain a first prediction result.
Fig. 5 is a structural diagram of a deep convolutional network provided by the present invention, and as shown in fig. 5, in an embodiment, the deep convolutional network includes a feature extraction network and a feature classification network, and the feature extraction network includes 5 feature extraction sub-networks connected in sequence.
The feature extraction sub-network comprises a convolution layer, a nonlinear layer and a regularization layer which are connected in sequence, and a convolution layer, a nonlinear layer, a regularization layer and a pooling layer which are connected in sequence.
The convolution layer is used for extracting image characteristics of the scanning electron microscope image; the nonlinear layer is used for increasing the nonlinearity of the feature extraction network; the regularization layer is used for averaging the image features into 0 and variance into 1; the pooling layer is used to reduce the dimensionality of the image features.
The feature classification network comprises a first full-connection layer, a nonlinear layer, a random discarding layer, a second full-connection layer and a classification layer which are sequentially connected.
The first full connection layer and the second full connection layer are stacked and used for classifying the image features output by the feature extraction network; the non-linear layer is used for increasing the non-linearity of the feature classification network; the random discarding layer is used for randomly discarding any parameter in the first full connection layer and the second full connection layer in proportion; the classification layer is used for normalizing the classification result output by the full connection layer and outputting a first prediction result in a probability form.
The deep convolutional network can be regarded as a combination of a feature extraction network and a feature classification network, the feature extraction network is composed of a convolutional layer 1 (volumetric layer), a nonlinear layer relu (relu layer), a regularization layer bn (bn) and a pooling layer pool (pooling layer). The convolution kernel size in all the convolution layers 1 was 3 × 3, and the image features were input into the next layer by sliding the extracted image features on the SEM image of the concrete. By superimposing the two convolutional layers 1, the same field (the area in the input space that affects a particular unit of the network) as one large convolutional kernel will be obtained while requiring fewer parameters, i.e., 2 × 3 × 3 — 18 parameters. After passing through the convolutional layer 1, the SEM image of the concrete will pass through the pooling layer RELU and the regularization layer in sequence. The pooling layer RELU will add non-linearity in the deep convolutional network, and for many classification problems, a linear classifier alone cannot meet the requirements. The regularization layer has the function of averaging the features of the input image into 0 and varying into 1, so that the training speed can be increased, the stability of the training can be improved, and data explosion can be avoided. After two convolutional layers 1, the image features will pass through a pooling layer POOL, which acts to reduce the dimensionality of the image features. The invention uses a maximum pooling approach, i.e. only the maximum parameters for the selected 2 x 2 area will be retained, while other parameters will be discarded. So after a pooling layer POOL, the 2 × 2 area will be reduced to 1 × 1 area, and for 224 × 224 image features, it will become 112 × 112 after a layer of pooling. In the entire feature extraction network, there are a total of 10 convolutional layers 1 and 5 pooling layers POOL, so the size of the final image feature is 7 × 7.
After the 7 × 7 image features are extracted by the feature extraction network, the image features are input to the feature classification network. The characteristic classification network is composed of two layers of full connection layers FC, and according to a universal approximation theory, the stacked full connection layers FC can approximate any function to achieve the classification effect. A random drop layer drop is placed between two fully connected layers FC, which will drop any parameters in the fully connected layers FC randomly in a certain proportion to avoid overfitting of the network. The two full-link layers FC are followed by a classification layer softmax (softmax layer), and the output classification result is normalized to output the first prediction result in the form of probability.
The concrete classification method provided by the invention classifies different scanning electron microscope images, the images are obtained by photographing different concrete samples, and the different concrete samples have different mixing ratios and different microstructures, so that the microscope images are different. But at present, even an experienced person cannot directly confirm which mix ratio concrete the SEM picture is.
The deep convolution network utilizes scanning electron microscope pictures to distinguish different concretes, and the detection of the proportion of the concrete is realized, which cannot be realized by naked eyes and experience.
The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. For the system disclosed by the embodiment, the description is relatively simple because the system corresponds to the method disclosed by the embodiment, and the relevant points can be referred to the method part for description.
The principles and embodiments of the present invention have been described herein using specific examples, which are provided only to help understand the method and the core concept of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the above, the present disclosure should not be construed as limiting the invention.
Claims (10)
1. A concrete classification method based on a scanning electron microscope image is characterized by comprising the following steps:
acquiring images of a plurality of concrete samples by using a scanning electron microscope to obtain a plurality of scanning electron microscope images, and constructing an original data set; the raw data set includes a number of the scanning electron microscope images and a truth class corresponding to the scanning electron microscope images; dividing an original data set into a training set and a testing set;
training the deep convolutional network by using the training set to obtain a trained deep convolutional network; the deep convolutional network comprises a feature extraction network and a feature classification network;
and inputting the concrete sample to be tested into the trained deep convolution network, and determining the category of the concrete sample to be tested.
2. The concrete classification method based on scanning electron microscope images according to claim 1, wherein before the image acquisition of the concrete samples by the scanning electron microscope to obtain the scanning electron microscope images, the method further comprises:
cutting the concrete block to obtain a plurality of concrete samples;
and polishing the plurality of concrete samples to obtain a plurality of smooth concrete samples.
3. The concrete classification method based on scanning electron microscope images according to claim 2, characterized in that the training of the deep convolutional network by using the training set to obtain the trained deep convolutional network specifically comprises:
inputting the scanning electron microscope images in the training set into a deep convolution network in batches to obtain a first prediction result;
comparing the first prediction result with the real value, and outputting a first comparison result in a cross entropy function mode;
according to the first comparison result, the difference between the first prediction result and the true value is measured, whether the training reaches the maximum iteration times is judged, and a first judgment result is obtained;
if the first judgment result is that the training reaches the maximum iteration number, finishing the training and outputting a trained deep convolutional network; and if the first judgment result is that the training does not reach the maximum iteration number, adjusting the weight of the deep convolution network by using a back propagation algorithm to obtain an adjusted deep convolution network, taking the adjusted deep convolution network as the deep convolution network, and returning to the step of inputting the scanning electron microscope images in the training set into the deep convolution network in batches to obtain a first prediction result.
4. The concrete classification method based on scanning electron microscope images according to claim 3, wherein the step of inputting the scanning electron microscope images in the training set into a deep convolutional network in batches to obtain a first prediction result specifically comprises:
inputting the scanning electron microscope image into the feature extraction network, and extracting the image features of the scanning electron microscope image;
and inputting the image features into the feature classification network, and outputting a first prediction result.
5. The concrete classification method based on scanning electron microscope images according to claim 1, characterized in that, after the deep convolutional network is trained by using the training set to obtain the trained deep convolutional network, the method further comprises:
inputting the scanning electron microscope images in the test set into the trained deep convolution network in sequence, and outputting a second prediction result; the second prediction result is the probability of the classification to which the scanning electron microscope image in the test set belongs;
comparing the second prediction result with the real value, and judging whether the second prediction result is correct;
counting the number of correct second prediction results and the number of wrong second prediction results;
calculating the prediction accuracy rate and the recall rate according to the number of correct second prediction results and the number of wrong second prediction results;
using formulas
Weighing the trained deep convolutional network; wherein F1score is a measure of the trained deep convolutional network.
6. The utility model provides a concrete classification system based on scanning electron microscope image which characterized in that includes:
the image acquisition module is used for acquiring images of a plurality of concrete samples by using a scanning electron microscope to obtain a plurality of scanning electron microscope images and construct an original data set; the raw data set includes a number of the scanning electron microscope images and a truth class corresponding to the scanning electron microscope images; dividing an original data set into a training set and a testing set;
the training module is used for training the deep convolutional network by utilizing the training set to obtain a trained deep convolutional network; the deep convolutional network comprises a feature extraction network and a feature classification network;
and the classification module is used for inputting the concrete sample to be detected into the trained deep convolution network and determining the category of the concrete sample to be detected.
7. The concrete classification system based on scanning electron microscope images as claimed in claim 6, further comprising: a sample acquisition module; the sample acquisition module specifically includes:
the sample cutting submodule is used for cutting the concrete block to obtain a plurality of concrete samples;
and the sample grinding submodule is used for polishing a plurality of concrete samples to obtain a plurality of smooth concrete samples.
8. The concrete classification method based on scanning electron microscope images according to claim 7, characterized in that the training module specifically comprises:
the prediction submodule is used for inputting the scanning electron microscope images in the training set into a deep convolution network in batches to obtain a first prediction result; the first prediction result is the probability of the classification to which the scanning electron microscope image in the training set belongs;
the comparison submodule is used for comparing the first prediction result with the real value and outputting a first comparison result in a cross entropy function mode;
the judgment submodule is used for measuring the difference between the first prediction result and the true value according to the first comparison result and judging whether the training reaches the maximum iteration number or not to obtain a first judgment result;
the output submodule is used for finishing the training and outputting the trained deep convolutional network if the first judgment result is that the training reaches the maximum iteration times;
and the return submodule is used for adjusting the weight of the deep convolution network by using a back propagation algorithm to obtain an adjusted deep convolution network if the first judgment result is that the training does not reach the maximum iteration number, taking the adjusted deep convolution network as the deep convolution network, and returning to the step of inputting the scanning electron microscope images in the training set into the deep convolution network in batches to obtain a first prediction result.
9. The concrete classification method based on the scanning electron microscope images as claimed in claim 6, characterized in that the feature extraction network comprises 5 feature extraction sub-networks connected in sequence;
the feature extraction sub-network comprises a convolution layer, a nonlinear layer and a regularization layer which are connected in sequence, and a convolution layer, a nonlinear layer, a regularization layer and a pooling layer which are connected in sequence;
the convolution layer is used for extracting image characteristics of the scanning electron microscope image;
the nonlinear layer is used for increasing the nonlinearity of the feature extraction network;
the regularization layer is used for averaging the image features into 0 and variance into 1;
the pooling layer is used to reduce the dimensionality of the image features.
10. The concrete classification method based on the scanning electron microscope images is characterized in that the feature classification network comprises a first full connection layer, a nonlinear layer, a random discarding layer, a second full connection layer and a classification layer which are connected in sequence;
the first full connection layer and the second full connection layer are stacked and used for classifying the image features output by the feature extraction network;
the non-linear layer is used for increasing the non-linearity of the feature classification network;
the random discarding layer is used for randomly discarding any parameter in the first full connection layer and the second full connection layer in proportion;
the classification layer is used for normalizing the classification result output by the full connection layer and outputting a first prediction result in a probability form.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111085821.6A CN113792666B (en) | 2021-09-16 | 2021-09-16 | Concrete classification method and system based on scanning electron microscope image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111085821.6A CN113792666B (en) | 2021-09-16 | 2021-09-16 | Concrete classification method and system based on scanning electron microscope image |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113792666A true CN113792666A (en) | 2021-12-14 |
| CN113792666B CN113792666B (en) | 2024-06-18 |
Family
ID=79183565
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111085821.6A Active CN113792666B (en) | 2021-09-16 | 2021-09-16 | Concrete classification method and system based on scanning electron microscope image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113792666B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114295050A (en) * | 2021-12-24 | 2022-04-08 | 重庆创筑科技有限公司 | Multi-size detection method and system for precast concrete plate based on convolutional neural network |
| US11481891B1 (en) * | 2021-07-12 | 2022-10-25 | The Florida International University Board Of Trustees | Systems and methods for quantifying concrete surface roughness |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105740894A (en) * | 2016-01-28 | 2016-07-06 | 北京航空航天大学 | Semantic annotation method for hyperspectral remote sensing image |
| CN110619355A (en) * | 2019-08-28 | 2019-12-27 | 武汉科技大学 | Automatic steel material microstructure identification method based on deep learning |
| CN110991402A (en) * | 2019-12-19 | 2020-04-10 | 湘潭大学 | Skin disease classification device and method based on deep learning |
| CN111340133A (en) * | 2020-03-10 | 2020-06-26 | 四川长虹电器股份有限公司 | Image classification processing method based on deep convolutional neural network |
| CN111368909A (en) * | 2020-03-03 | 2020-07-03 | 温州大学 | Vehicle logo identification method based on convolutional neural network depth features |
| CN111612066A (en) * | 2020-05-21 | 2020-09-01 | 成都理工大学 | Remote sensing image classification method based on depth fusion convolutional neural network |
| CN112560968A (en) * | 2020-12-21 | 2021-03-26 | 齐鲁工业大学 | HER2 image classification method and system based on convolution and residual error network |
-
2021
- 2021-09-16 CN CN202111085821.6A patent/CN113792666B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105740894A (en) * | 2016-01-28 | 2016-07-06 | 北京航空航天大学 | Semantic annotation method for hyperspectral remote sensing image |
| CN110619355A (en) * | 2019-08-28 | 2019-12-27 | 武汉科技大学 | Automatic steel material microstructure identification method based on deep learning |
| CN110991402A (en) * | 2019-12-19 | 2020-04-10 | 湘潭大学 | Skin disease classification device and method based on deep learning |
| CN111368909A (en) * | 2020-03-03 | 2020-07-03 | 温州大学 | Vehicle logo identification method based on convolutional neural network depth features |
| CN111340133A (en) * | 2020-03-10 | 2020-06-26 | 四川长虹电器股份有限公司 | Image classification processing method based on deep convolutional neural network |
| CN111612066A (en) * | 2020-05-21 | 2020-09-01 | 成都理工大学 | Remote sensing image classification method based on depth fusion convolutional neural network |
| CN112560968A (en) * | 2020-12-21 | 2021-03-26 | 齐鲁工业大学 | HER2 image classification method and system based on convolution and residual error network |
Non-Patent Citations (1)
| Title |
|---|
| 张丽秀 等: "基于SVM混凝土电镜图像类型识别", 《沈阳建筑大学学报（自然科学版）》, vol. 36, no. 1, pages 1 - 7 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11481891B1 (en) * | 2021-07-12 | 2022-10-25 | The Florida International University Board Of Trustees | Systems and methods for quantifying concrete surface roughness |
| CN114295050A (en) * | 2021-12-24 | 2022-04-08 | 重庆创筑科技有限公司 | Multi-size detection method and system for precast concrete plate based on convolutional neural network |
| CN114295050B (en) * | 2021-12-24 | 2024-05-17 | 重庆创筑科技有限公司 | Precast concrete slab multi-size detection method and system based on convolutional neural network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113792666B (en) | 2024-06-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112861720B (en) | Remote sensing image small sample target detection method based on prototype convolutional neural network | |
| CN109325550B (en) | No-reference image quality evaluation method based on image entropy | |
| CN105512684B (en) | Logo automatic identifying method based on principal component analysis convolutional neural networks | |
| US8687879B2 (en) | Method and apparatus for generating special-purpose image analysis algorithms | |
| CN107563999A (en) | A kind of chip defect recognition methods based on convolutional neural networks | |
| CN111160249A (en) | Multi-class target detection method of optical remote sensing image based on cross-scale feature fusion | |
| CN110400293B (en) | No-reference image quality evaluation method based on deep forest classification | |
| CN110717554B (en) | Image recognition method, electronic device, and storage medium | |
| CN113792666B (en) | Concrete classification method and system based on scanning electron microscope image | |
| CN108446616B (en) | Road extraction method based on full convolution neural network ensemble learning | |
| CN109993236A (en) | Few sample language of the Manchus matching process based on one-shot Siamese convolutional neural networks | |
| CN109657600B (en) | Video area removal tampering detection method and device | |
| CN111325237B (en) | Image recognition method based on attention interaction mechanism | |
| CN111343147A (en) | Network attack detection device and method based on deep learning | |
| CN110751644A (en) | Road surface crack detection method | |
| CN116363440A (en) | Deep learning-based identification and detection method and system for colored microplastic in soil | |
| Xiang et al. | Research on splicing image detection algorithms based on natural image statistical characteristics | |
| CN111275126A (en) | Sample data set generation method, device, equipment and storage medium | |
| CN106682604B (en) | Blurred image detection method based on deep learning | |
| Aferi et al. | Cotton texture segmentation based on image texture analysis using gray level co-occurrence matrix (GLCM) and Euclidean distance | |
| CN116579985A (en) | Wafer defect detection method based on refindet model | |
| CN109919056A (en) | A kind of face identification method based on discriminate principal component analysis | |
| CN113469084B (en) | Hyperspectral image classification method based on contrast generation countermeasure network | |
| CN107679528A (en) | A kind of pedestrian detection method based on AdaBoost SVM Ensemble Learning Algorithms | |
| CN105023016B (en) | Target apperception method based on compressed sensing classification |
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 |