CN111461259B - Image classification method and system based on red-black morphological wavelet pooling network - Google Patents
Image classification method and system based on red-black morphological wavelet pooling network Download PDFInfo
- Publication number
- CN111461259B CN111461259B CN202010339964.4A CN202010339964A CN111461259B CN 111461259 B CN111461259 B CN 111461259B CN 202010339964 A CN202010339964 A CN 202010339964A CN 111461259 B CN111461259 B CN 111461259B
- Authority
- CN
- China
- Prior art keywords
- red
- black
- pooling
- wavelet
- subset
- 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
- 238000011176 pooling Methods 0.000 title claims abstract description 77
- 238000000034 method Methods 0.000 title claims abstract description 48
- 230000000877 morphologic effect Effects 0.000 title claims description 36
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 22
- 230000008569 process Effects 0.000 claims abstract description 22
- 238000012549 training Methods 0.000 claims abstract description 9
- 238000011065 in-situ storage Methods 0.000 claims description 13
- 230000003628 erosive effect Effects 0.000 claims description 9
- 230000007797 corrosion Effects 0.000 claims description 6
- 238000005260 corrosion Methods 0.000 claims description 6
- 238000010586 diagram Methods 0.000 claims description 6
- 238000013528 artificial neural network Methods 0.000 abstract description 5
- 230000006872 improvement Effects 0.000 abstract description 2
- 230000006870 function Effects 0.000 description 5
- 238000012545 processing Methods 0.000 description 5
- 238000012360 testing method Methods 0.000 description 5
- 230000008901 benefit Effects 0.000 description 4
- 238000001514 detection method Methods 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 239000013589 supplement Substances 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 238000007792 addition Methods 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 238000003709 image segmentation Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 239000000047 product Substances 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
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)
- Bioinformatics & Computational Biology (AREA)
- Computational Linguistics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Evolutionary Biology (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention provides an image classification method and system based on a red-black morphology wavelet pooling network, comprising the steps of preparing an image data set to be classified; fusing red-black morphology wavelets with the maximum pooling, constructing a lifting scheme pooling layer, and extracting the characteristics of the images to be classified and performing downsampling; constructing and training a lifting scheme convolutional neural network based on an image data set to be classified, and realizing classification through a Softmax classifier, wherein the lifting scheme convolutional neural network comprises a lifting scheme pooling layer; the lifting scheme pooling layer comprises two branches, namely a red-black form wavelet branch and a maximum pooling branch, and the final output y of the lifting scheme pooling layer is obtained by fusing the output of the y red-black form wavelet and the output of the maximum pooling branch; the red-black morphology wavelet branch comprises a lifting process in the vertical and horizontal directions and a lifting process in the diagonal direction. The invention can effectively improve the recognition capability of the neural network and has obvious improvement on the classification task of the reference image.
Description
Technical Field
The invention belongs to the technical field of image processing and deep learning, and particularly relates to a scheme for classifying images based on a red-black morphological wavelet pooling network.
Background
Convolutional Neural Networks (CNNs) are popular methods for computer vision tasks such as image classification, object detection, and image segmentation. The goal of CNN is to extract discriminant features at different levels of abstraction to enhance intra-class consistency and inter-class variability. In the previous layers, the CNN extracts some bottom features, such as edges, endpoints, corner points, etc., and in the subsequent layers, the CNN combines these bottom features to obtain high-level features. Finally, CNNs distinguish one category from other categories based on these high-level features. However, the usual max-pooling layer only extracts the maximum value for a fixed size window, regardless of the structure of the previous layer, which may lead to loss of some of the structure and geometry in the feature map. Furthermore, noise in the image can prevent CNNs from learning useful features.
The second generation wavelet transform based on lifting scheme is an effective way to solve these problems. With lifting schemes, several non-linear wavelets have been constructed. For example, heijmans and Goutsias propose morphological wavelets in which important geometric information is well preserved in low resolution images. This is because the signal analysis operators (predictor and updater) are nonlinear. An example of a morphological wavelet is a maximal lifted morphological wavelet, which is useful in pattern recognition applications because it preserves the local maxima of the processed image. The two-dimensional morphological wavelet can be expanded into a one-dimensional morphological wavelet by tensor product, but anisotropy is introduced, and the image cannot be described in any direction. It can detect vertical, horizontal and diagonal features, but has less feature detection capability for other directions. In contrast, the red-black wavelet proposed by Uytterhoeven and Bulthel is non-linear and inseparable, which is constructed from a two-dimensional lifting scheme. The red and black wavelets are more suitable for the structure of the two-dimensional natural image, and can better detect the characteristics of all directions. In addition to the good characteristics of non-linearity and inseparability, the second generation wavelet transform can reduce noise in the original image by removing the high frequency component after wavelet decomposition. Therefore, the invention proposes to introduce red and black wavelets to promote the morphological wavelets to adapt to the structure of the two-dimensional natural image, and can enhance the representation capability of wavelet transformation.
Disclosure of Invention
The invention aims to combine the advantages of morphological wavelet and red-black wavelet, construct red-black morphological wavelet as the supplement of the maximum pooling layer, and provide a new image classification scheme.
The technical scheme of the invention is an image classification method based on a red-black morphological wavelet pooling network, which comprises the steps of preparing an image data set to be classified; fusing red-black morphology wavelets with the maximum pooling, constructing a lifting scheme pooling layer, and extracting the characteristics of the images to be classified and performing downsampling; constructing and training a lifting scheme convolutional neural network based on an image data set to be classified, and realizing classification through a Softmax classifier, wherein the lifting scheme convolutional neural network comprises a lifting scheme pooling layer;
the lifting scheme pooling layer comprises two branches, namely a red-black morphology wavelet branch and a maximum pooling branch, and the final output y of the lifting scheme pooling layer is obtained by fusing the output r of the red-black morphology wavelet and the output m of the maximum pooling branch; the red-black morphology wavelet branch comprises a vertical and horizontal lifting process and a diagonal lifting process,
the vertical and horizontal lifting process comprises dividing an input characteristic diagram x into a Red subset and a Black subset according to the following rule to realize splitting, then predicting and updating based on morphological erosion operation,
Red＝{x ij |x ij ∈x,i％2＝j％2}
Black＝{x ij |x ij ∈x,i％2≠j％2}
wherein x is ij Is the element in the feature map x at coordinates (i, j);
the vertical-angle lifting process comprises dividing an output set obtained in the vertical and horizontal lifting processes into Blue subsets and Green subsets according to the following rule to realize splitting, then predicting and updating based on morphological erosion operation,
Blue＝{x ij |x ij ∈x,i％2＝0,j％2＝0}
Green＝{x ij |x ij ∈x,i％2≠0,j％2≠0}
wherein x is ij Is the element in the feature map x at coordinates (i, j).
And the final output of the lifting scheme pooling layer is obtained by fusing the output r of the red-black morphological wavelet and the output m of the maximum pooling branch, the implementation mode is as follows,
y＝α·r+(1-α)·m
wherein the scalar alpha is a weight parameter and takes a value between 0 and 1.
In addition, in the lifting process in the vertical and horizontal directions, prediction and updating based on morphological erosion operation are realized as follows,
1) Firstly, taking the minimum value of 4 adjacent red subset samples to predict the corresponding element in the black subset, and replacing the original black subset element with the predicted result in situ;
2) Then, the predicted result is eroded to update the red subset, and the updated result replaces the original red subset in situ and is used as the output of the lifting process in the vertical and horizontal directions.
In addition, in the lifting process in the diagonal direction, prediction and updating based on morphological erosion operation are realized as follows,
1) Firstly, taking the minimum value of 4 adjacent blue subset samples to predict the corresponding elements in the green subset, and replacing the original green subset elements with the predicted results in situ;
2) Then, the predicted result is subjected to corrosion treatment to update the blue subset, and the updated result replaces the original blue subset in situ and is used as the output of the step.
Moreover, the lifting scheme convolutional neural network is realized by adopting a local binary threshold network structure, and sequentially comprises an input layer, a convolutional layer C1, a lifting scheme pooling layer L1, a convolutional layer C2, a lifting scheme pooling layer L2, a full-connection layer F1, a full-connection layer F2 and an output layer.
Moreover, the output layer is realized by adopting an Euclidean radial basis function and is obtained through a softmax function.
The invention provides an image classification system based on a red-black morphological wavelet pooling network, which is used for the image classification method based on the red-black morphological wavelet pooling network.
The invention combines the advantages of morphological wavelet and red-black wavelet, constructs a new pooling layer, and provides a lifting scheme convolutional neural network method based on the red-black morphological wavelet, which can keep the feature map information as much as possible during dimension reduction. The pooling layer of the lifting scheme provided by the invention is the fusion of red-black morphology wavelets and the maximum pooling layer. The red-black morphology wavelet combines the inseparability of the red-black morphology wavelet with the nonlinear characteristics of the morphology wavelet, and has the advantage of retaining geometric and structural features in any direction in the low-resolution feature map. The invention introduces a lifting scheme of wavelet transformation into the field of the neural network by creatively providing a lifting scheme pooling layer, and overcomes the defects that the pooling layer of the neural network provided by the prior art can lose part of structure and geometric characteristics in a characteristic diagram and can not weaken noise. The invention can effectively improve the recognition capability of the neural network and has obvious improvement on the classification task of the reference image.
Drawings
FIG. 1 is a schematic illustration of the pooling layer of the lifting scheme of the embodiment of the invention.
FIG. 2 is a diagram illustrating the architecture of a convolutional neural network for a boosting scheme in accordance with an embodiment of the present invention.
Detailed Description
The technical scheme of the invention is described in detail below with reference to the accompanying drawings and examples.
The red-black morphological wavelet-based lifting scheme pooling method provided by the invention can well reserve geometric and structural characteristics in the characteristic diagram. The red-black morphology wavelet combines the advantages of non-linearity and inseparability and is used as a supplement to the largest pooling layer to preserve the characteristics of being weakened by the largest pooling layer. The invention fuses red-black morphology wavelets with maximum pooling, and provides a novel lifting scheme pooling layer. The method is used for constructing a convolutional neural network of a lifting scheme, and can keep the structure and geometric characteristics as much as possible while compressing the size of the feature map, so that the image classification accuracy of the whole network is improved.
The lifting scheme pooling layer provided by combining red-black morphology wavelets and maximum pooling can keep geometric and two-dimensional structure information to a large extent when the feature map is subjected to dimension reduction on the neural network. Therefore, the features still have discriminant after downsampling, so that intra-class consistency and inter-class variability can be enhanced, and classification accuracy is improved. Moreover, the invention provides a convolutional neural network image classification method based on a lifting scheme of red and black morphology wavelets, and the method can be used for reducing the dimension and simultaneously better keeping the image characteristics, so that the method has obvious lifting on the classification effect of the images.
The invention provides a red-black morphological wavelet-based lifting scheme pooling method, the principle is shown in fig. 1, and the network structure for image classification in a corresponding embodiment is shown in fig. 2. The image classification method based on the red-black morphological wavelet pooling network provided by the embodiment comprises the following 3 steps:
step 1 prepares an image dataset to be classified, and the implementation mode is as follows:
before execution, M image data sets D to be classified are prepared, and the data sets D are divided into two non-overlapping sub-data sets D train ,D test Respectively used for training and testing; all dataset images are of size n x n pixels.
And 2, fusing the red-black morphological wavelet with the maximum pooling, and constructing a lifting scheme pooling layer for extracting the characteristics of the images to be classified and performing downsampling. The implementation mode is as follows:
in the embodiment of the invention, a common convolution layer (which is specifically realized as the prior art and is not repeated in the invention) is used for extracting the characteristics, and then the characteristics are dimension-reduced by a lifting scheme pooling layer based on red-black morphology wavelets. The lifting scheme pooling layer is shown in fig. 1.
The lifting scheme pooling layer comprises two branches, namely a red-black morphological wavelet branch and a maximum pooling branch. The red-black morphology wavelet branch comprises the following two processing procedures:
(1) Lifting process in vertical and horizontal directions:
let us assume sub-data set D train ,D test Any image is taken as an input feature map x, the input feature map x is divided into two subsets, namely a Red subset (Red subset) and a Black subset (Black subset) respectively by taking the coordinates (i, j) as a rule that certain conditions are met, and splitting is realized. The rule of division is as follows.
Red＝{x ij |x ij ∈x,i％2＝j％2} (1)
Black＝{x ij |x ij ∈x,i％2≠j％2} (2)
Wherein x is ij Is the element in the feature map x at coordinates (i, j).
Then further processing using morphological erosion operations:
1) Firstly, taking the minimum value of 4 adjacent red subset samples to predict the corresponding element in the black subset, and replacing the original black subset element by the predicted result in situ.
2) Then, the predicted result is subjected to corrosion treatment to update the red subset, and the updated result replaces the original red subset in situ and is used as the output of the step.
These two steps are shown in equations (3) and (4).
x ij ＝x ij -(x i-1,j ^x i,j-1 ^x i,j+1 ^x i+1,j ),x ij ∈Black (3)
x ij ＝x ij +(0^x i-1,j ^x i,j-1 ^x i,j+1 ^x i+1,j ),x ij ∈Red (4)
(2) Lifting process in diagonal direction:
the result obtained in step (1) is used as an input of the step, and the processing is continued. Since the operations are all performed in situ, the output set of step (1) is at the location of the red subset. For this the output set is divided into Blue subsets (Blue subsets) and Green subsets (Green subsets) according to coordinates (i, j), i.e. a further splitting is achieved. The specific division rules are as follows.
Blue＝{x ij |x ij ∈x,i％2＝0,j％2＝0} (5)
Green＝{x ij |x ij ∈x,i％2≠0,j％2≠0} (6)
Then further processing using morphological erosion operations:
1) Firstly, taking the minimum value of 4 adjacent blue subset samples to predict the corresponding elements in the green subset, and replacing the original green subset elements by the predicted results in situ.
2) Then, the predicted result is subjected to corrosion treatment to update the blue subset, and the updated result replaces the original blue subset in situ and is used as the output of the step.
With erosion operations, the prediction and update steps are equations (9) and (10), respectively.
x ij ＝x ij -(x i-1,j-1 ^x i-1,j+1 ^x i+1,j-1 ^x i+1,j+1 ),x ij ∈Green (7)
x ij ＝x ij +(0^x i-1,j-1 ^x i-1,j+1 ^x i+1,j-1 ^x i+1,j+1 ),x ij ∈Blue (8)
The updated blue subset is the output of the whole red-black morphology wavelet branch, denoted by r.
The final output y of the pooling layer of the lifting scheme is obtained by fusing the output r of the red-black morphological wavelet and the output m of the maximum pooling branch, and is shown in the following formula.
y＝α·r+(1-α)·m (9)
Where the scalar α is a weight parameter between 0 and 1. In the embodiment, alpha takes a preferred value of 0.3.
The max-pooled branching section may take local maxima, for example, for the input image, output maxima within a local range of size 2 x 2. Such an operation is performed on the entire image. So if the input image size is h x w, the output image size will be 0.5h x 0.5w, h, w representing the high and wide values. The implementation mode of the maximum pooling is the prior art, and the invention uses the maximum pooling as a branch and is not repeated.
And 3, constructing and training a lifting scheme convolutional neural network, and realizing classification by a Softmax classifier.
As shown in fig. 2, the local binary threshold network structure adopted by the present invention includes 1 input layer, 2 convolution layers, 2 downsampling layers, 2 full connection layers and 1 output layer, and it can be seen that the local binary threshold network of the present invention has 8 layers in total, and the parameter settings of each layer are respectively as follows:
input layer: the input data is a texture image of n×n pixels.
Layer C1: the layer is a convolution layer, the convolution kernel size is 5 multiplied by 5, and the convolution depth is 6;
layer L1: the layer is a lifting scheme pooling layer, wherein the size of a maximum pooling branch window is 2 multiplied by 2, and the sliding step length is 2;
c2 layer: the layer is a convolution layer, the convolution kernel size is 5 multiplied by 5, and the convolution depth is 16;
l2 layer: the layer is also a lifting scheme pooling layer, wherein the maximum pooling branch window size is 2 multiplied by 2, and the sliding step length is 2;
f1 layer: the layer is a full-connection layer, and the output dimension is 120;
f2 layer: the layer is a full-connection layer, and the output dimension is 84;
output layer: is composed of 10 Euclidean radial basis functions. The 10 Euclidean radial basis functions of the output layer are obtained by using a softmax function, and represent the probability that a certain image belongs to 10 classes respectively, and the class with the highest probability is the classification result of the image.
After the network structure is set, the network is trained through the back propagation of error sensitive items and a random gradient algorithm (the specific implementation is the prior art, and the invention is not repeated), unknown parameters in the network are learned, namely, the parameters of each module are learned through repeated iteration of the network, and the characteristic expression with the image information is obtained fully.
D based on the result obtained in step 1 train Is a training set for training network parameters. This dataset serves as an input to the network, optimizing the training of network parameters by back-propagation and random gradient algorithms. After training well, D test The method is used for testing, network parameters are fixed at the moment, the output of the network is compared with the real label, and the classification accuracy is obtained and used for evaluating the performance of the network.
When the network performance meets the requirement, inputting any image to be classified into the trained network, and obtaining a corresponding classification result.
The invention is compared with the common convolutional neural network, the classification accuracy of the method is improved by 2.029% on the SVHN data set, and the classification accuracy is improved by 0.580% on the MNIST data set. In the comparison experiment, the only difference between the network of the method and the common convolutional neural network is that the former uses a lifting scheme pooling layer and the latter uses a maximum pooling layer at the corresponding part, and all other settings are the same.
In specific implementation, the above flow can be automatically operated by adopting a computer software technology, and the system device of the operation method is also in the protection scope of the invention.
The specific embodiments described herein are offered by way of example only to illustrate the spirit of the invention. Those skilled in the art may make various modifications or additions to the described embodiments or substitutions thereof without departing from the spirit of the invention or exceeding the scope of the invention as defined in the accompanying claims.
Claims (7)
1. An image classification method based on a red-black morphological wavelet pooling network is characterized by comprising the following steps of: comprises the steps of preparing an image data set to be classified; fusing red-black morphology wavelets with the maximum pooling, constructing a lifting scheme pooling layer, and extracting the characteristics of the images to be classified and performing downsampling; constructing and training a lifting scheme convolutional neural network based on an image data set to be classified, and realizing classification through a Softmax classifier, wherein the lifting scheme convolutional neural network comprises a lifting scheme pooling layer;
the lifting scheme pooling layer comprises two branches, namely a red-black morphology wavelet branch and a maximum pooling branch, and the final output y of the lifting scheme pooling layer is obtained by fusing the output r of the red-black morphology wavelet and the output m of the maximum pooling branch; the red-black morphology wavelet branch comprises a vertical and horizontal lifting process and a diagonal lifting process,
the vertical and horizontal lifting process comprises dividing an input characteristic diagram x into a Red subset and a Black subset according to the following rule to realize splitting, then predicting and updating based on morphological erosion operation to obtain an output set of the vertical and horizontal lifting process,
Red＝{x ij |x ij ∈x,i％2＝j％2}
Black＝{x ij |x ij ∈x,i％2≠j％2}
wherein x is ij Is the element in the feature map x at coordinates (i, j);
the diagonal lifting process comprises dividing an output set obtained in the vertical and horizontal lifting processes into a Blue subset and a Green subset according to the following rule, realizing splitting, and then predicting and updating based on morphological corrosion operation, wherein the updated Blue subset is the output of the whole red and black morphological wavelet branch and is denoted by r;
Blue＝{x ij |x ij ∈x,i％2＝0,j％2＝0}
Green＝{x ij |x ij ∈x,i％2≠0,j％2≠0}
wherein x is ij Is the element in the coordinates (i, j) in the output set obtained by the vertical and horizontal lifting process of the input characteristic diagram x.
2. The image classification method based on the red-black morphology wavelet pooling network according to claim 1, wherein: the final output of the pooling layer of the lifting scheme is obtained by fusing the output r of the red-black morphological wavelet and the output m of the maximum pooling branch, and the implementation mode is as follows,
y＝α·r+(1-α)·m
wherein the scalar alpha is a weight parameter and takes a value between 0 and 1.
3. The image classification method based on the red-black morphology wavelet pooling network according to claim 1, wherein: in the lifting process in the vertical and horizontal directions, prediction and updating based on morphological erosion operation are realized as follows,
1) Firstly, taking the minimum value of 4 adjacent red subset samples to predict the corresponding element in the black subset, and replacing the original black subset element with the predicted result in situ;
2) Then, the predicted result is eroded to update the red subset, and the updated result replaces the original red subset in situ and is used as the output of the lifting process in the vertical and horizontal directions.
4. The image classification method based on the red-black morphology wavelet pooling network according to claim 3, wherein: in the lifting process in the diagonal direction, prediction and updating based on morphological corrosion operation are realized as follows,
1) Firstly, taking the minimum value of 4 adjacent blue subset samples to predict the corresponding elements in the green subset, and replacing the original green subset elements with the predicted results in situ;
2) Then, the predicted result is subjected to corrosion treatment to update the blue subset, and the updated result replaces the original blue subset in situ and is used as the output of the step.
5. The red-black morphology wavelet pooling network-based image classification method according to claim 1 or 2 or 3 or 4, wherein: the lifting scheme convolutional neural network is realized by adopting a local binary threshold network structure, and sequentially comprises an input layer, a convolutional layer C1, a lifting scheme pooling layer L1, a convolutional layer C2, a lifting scheme pooling layer L2, a full-connection layer F1, a full-connection layer F2 and an output layer.
6. The image classification method based on the red-black morphology wavelet pooling network according to claim 5, wherein: the output layer is realized by adopting an Euclidean radial basis function and is obtained through a softmax function.
7. An image classification system based on red-black morphology wavelet pooling network is characterized in that: an image classification method based on red-black morphology wavelet pooling network according to any one of claims 1-6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010339964.4A CN111461259B (en) | 2020-04-26 | 2020-04-26 | Image classification method and system based on red-black morphological wavelet pooling network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010339964.4A CN111461259B (en) | 2020-04-26 | 2020-04-26 | Image classification method and system based on red-black morphological wavelet pooling network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111461259A CN111461259A (en) | 2020-07-28 |
| CN111461259B true CN111461259B (en) | 2023-11-28 |
Family
ID=71681309
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010339964.4A Active CN111461259B (en) | 2020-04-26 | 2020-04-26 | Image classification method and system based on red-black morphological wavelet pooling network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111461259B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115661006B (en) * | 2022-12-29 | 2023-03-10 | 湖南国天电子科技有限公司 | Seabed landform image denoising method |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101847203A (en) * | 2009-03-25 | 2010-09-29 | 何玉青 | Human face recognition method based on red and black small wave conversation |
| CN105096256A (en) * | 2015-08-31 | 2015-11-25 | 深圳市博铭维智能科技有限公司 | Special robot moving platform and super-resolution reconstruction method of collected image thereof |
| CN110163286A (en) * | 2019-05-24 | 2019-08-23 | 常熟理工学院 | Hybrid pooling-based domain adaptive image classification method |
| CN110796167A (en) * | 2019-09-25 | 2020-02-14 | 武汉大学 | Image classification method based on deep neural network of lifting scheme |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106897714B (en) * | 2017-03-23 | 2020-01-14 | 北京大学深圳研究生院 | Video motion detection method based on convolutional neural network |
-
2020
- 2020-04-26 CN CN202010339964.4A patent/CN111461259B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101847203A (en) * | 2009-03-25 | 2010-09-29 | 何玉青 | Human face recognition method based on red and black small wave conversation |
| CN105096256A (en) * | 2015-08-31 | 2015-11-25 | 深圳市博铭维智能科技有限公司 | Special robot moving platform and super-resolution reconstruction method of collected image thereof |
| CN110163286A (en) * | 2019-05-24 | 2019-08-23 | 常熟理工学院 | Hybrid pooling-based domain adaptive image classification method |
| CN110796167A (en) * | 2019-09-25 | 2020-02-14 | 武汉大学 | Image classification method based on deep neural network of lifting scheme |
Non-Patent Citations (2)
| Title |
|---|
| Cruz, T.N等.Detection and classification of lesions in mammographies using neural networks and morphological wavelets.IEEE Latin America Transactions.2018,第926-932页. * |
| 余珮嘉 ; 张靖 ; 谢晓尧 ; .基于自适应池化的行人检测方法.河北科技大学学报.2020,(第06期),第533-539页. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111461259A (en) | 2020-07-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108133188B (en) | Behavior identification method based on motion history image and convolutional neural network | |
| Kannojia et al. | Effects of varying resolution on performance of CNN based image classification: An experimental study | |
| US20190228268A1 (en) | Method and system for cell image segmentation using multi-stage convolutional neural networks | |
| CN109215034B (en) | Weak supervision image semantic segmentation method based on spatial pyramid covering pooling | |
| CN111753828B (en) | Natural scene horizontal character detection method based on deep convolutional neural network | |
| CN111209858B (en) | Real-time license plate detection method based on deep convolutional neural network | |
| CN110827330B (en) | Time sequence integrated multispectral remote sensing image change detection method and system | |
| CN111126226B (en) | Radiation source individual identification method based on small sample learning and feature enhancement | |
| CN111986126B (en) | Multi-target detection method based on improved VGG16 network | |
| CN111667019B (en) | Hyperspectral image classification method based on deformable separation convolution | |
| Rangkuti et al. | A novel reliable approach for image batik classification that invariant with scale and rotation using MU2ECS-LBP algorithm | |
| CN110751195A (en) | Fine-grained image classification method based on improved YOLOv3 | |
| Sun et al. | Can shape structure features improve model robustness under diverse adversarial settings? | |
| Sulehria et al. | Vehicle number plate recognition using mathematical morphology and neural networks | |
| CN114897782B (en) | Gastric cancer pathological section image segmentation prediction method based on generation type countermeasure network | |
| CN113139618B (en) | Robustness-enhanced classification method and device based on integrated defense | |
| CN114882278A (en) | Tire pattern classification method and device based on attention mechanism and transfer learning | |
| Tourtounis et al. | Salt-n-pepper noise filtering using cellular automata | |
| CN111461259B (en) | Image classification method and system based on red-black morphological wavelet pooling network | |
| CN111368865A (en) | Method and device for detecting remote sensing image oil storage tank, readable storage medium and equipment | |
| CN110349119B (en) | Pavement disease detection method and device based on edge detection neural network | |
| CN114663658B (en) | Small sample AOI surface defect detection method with cross-domain migration capability | |
| CN114241192A (en) | Infrared intelligent diagnosis system and method for substation equipment | |
| Zhang | An image recognition algorithm based on self-encoding and convolutional neural network fusion | |
| CN117911437A (en) | Buckwheat grain adhesion segmentation method for improving YOLOv x |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |