CN109934292B - Unbalanced polarization SAR terrain classification method based on cost sensitivity assisted learning - Google Patents
Unbalanced polarization SAR terrain classification method based on cost sensitivity assisted learning Download PDFInfo
- Publication number
- CN109934292B CN109934292B CN201910198273.4A CN201910198273A CN109934292B CN 109934292 B CN109934292 B CN 109934292B CN 201910198273 A CN201910198273 A CN 201910198273A CN 109934292 B CN109934292 B CN 109934292B
- Authority
- CN
- China
- Prior art keywords
- layer
- class
- classification
- cost
- convolution
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses an unbalanced polarization SAR terrain classification method based on cost sensitivity assisted learning, which comprises the steps of firstly inputting a polarization SAR image to be classified and real terrain label information corresponding to the polarization SAR image; taking an absolute value of a polarization coherent matrix T of the to-be-classified polarized SAR image data to obtain a modulus matrix | T |; then selecting a training sample set; building a cost sensitivity auxiliary learning model; classifying the polarized SAR images after training the cost sensitivity auxiliary learning model; and finally, outputting a visual classification result graph of the whole graph. The cost sensitivity auxiliary learning model built by the invention not only has simple network hierarchy and less required label data sets, but also can solve the problem of unbalanced ground objects in the polarized SAR ground object classification, realize better classification of small objects and ensure higher classification accuracy.
Description
Technical Field
The invention belongs to the technical field of image processing, and particularly relates to an unbalanced polarization SAR terrain classification method based on cost sensitivity assisted learning.
Background
The classification of polarized SAR images has recently received attention from more and more researchers as important research content for understanding and interpreting polarized SAR images, and is widely applied to various fields, such as land cover type discrimination, ground target detection, geological exploration, vegetation type discrimination, and the like. According to the utilization modes of the marked samples and the unmarked samples in the classification method, the polarized SAR terrain classification method can be mainly divided into three types: unsupervised classification methods, supervised classification methods and semi-supervised classification methods.
For the polarized SAR image classification problem, the supervised classification method usually obtains a good classification result more easily than the unsupervised classification, but the supervised classification method usually needs sufficient labeled samples as training samples, and actually, the acquisition of the labeled samples is very difficult, and a large amount of manpower and material resources are consumed. The non-labeled data is relatively easy to acquire, the non-labeled data can reflect certain information of the data, but the polarized SAR image classification effect of the unsupervised method is not as good as that of the supervised method. Therefore, the semi-supervised learning method for performing supplementary training on a small amount of labeled samples by using a large amount of unlabelled samples attracts extensive attention of researchers, improves the classification accuracy by finding the implicit information in the unlabelled samples and combining the labeled sample information, makes up the defects of supervised learning and unsupervised learning, and becomes a research hotspot in the field of machine learning.
The existing method still has the following defects:
the full convolution network has too deep hierarchy, which results in too long network training time, and the way of selecting the labeled sample is complicated when the network is trained, and manpower and material resources are consumed.
Due to the imaging mechanism of the synthetic aperture radar, a large-area homogeneous region and small-block targets such as artificial buildings, vehicles, ships and the like exist in the polarized SAR image, and due to the unbalanced ground object type, when the pixels of the image are classified, the label information acquired by the small-block target region is small, so that the classification result of the small-area targets is still not accurate.
Disclosure of Invention
The technical problem to be solved by the invention is to provide a cost sensitivity assisted learning-based unbalanced polarization SAR terrain classification method aiming at the defects in the prior art, and the method can be used for solving the problems of reducing the depth of a network and the types of unbalanced terrain in a polarized SAR image, is beneficial to extracting small target information and achieves the aim of improving the classification accuracy of the polarized SAR image.
The invention adopts the following technical scheme:
the method for classifying the unbalanced polarization SAR terrain based on cost sensitivity assisted learning comprises the steps of firstly inputting a polarization SAR image to be classified and real terrain label information corresponding to the polarization SAR image; taking an absolute value of a polarization coherent matrix T of the to-be-classified polarized SAR image data to obtain a modulus matrix | T |; then selecting a training sample set; building a cost sensitivity auxiliary learning model; classifying the polarized SAR images after training the cost sensitivity auxiliary learning model; and finally, outputting a visual classification result graph of the whole graph.
Specifically, the polarized coherent matrix after modulus taking is used as a training sample, and 10% of pixels are randomly selected from all pixels with real labels of the polarized SAR image as label information.
Specifically, the steps of constructing the cost sensitivity auxiliary learning model are as follows:
s401, building a feature extraction backbone network, wherein the feature extraction network is divided into four layers, and the structure of the feature extraction backbone network is divided into the following layers in sequence: an input layer, a first convolution layer, a second convolution layer, a third convolution layer;
s402, building a classification network of one of the branches, which is divided into two layers, wherein the structure of the classification network is as follows: a fourth convolution layer, a classification layer;
and S403, constructing an auxiliary clustering network of one of the branches, wherein the auxiliary clustering network is divided into three layers, and the structure of the auxiliary clustering network is a category subdivision layer, a multi-category parallel full-connection layer and a clustering layer in sequence.
Further, in step S401, the contents and parameters of each layer are set as follows:
an input layer, the number of nodes of which is set to the dimension of the input correlation matrix | T |;
the number of convolution kernels is set to be 50, the size of the convolution kernels is set to be 5 x 5, and the convolution step size is 1 x 1;
the number of convolution kernels of the second convolution layer is set to be 30, the size of the convolution kernels is set to be 5 x 5, and the convolution step size is 1 x 1;
and in the third convolution layer, the number of convolution kernels is set to be 9, the size of the convolution kernels is 5 x 5, and the convolution step size is 1 x 1.
Further, in step S402, the contents and parameters of each layer are as follows:
the number of convolution kernels is set as the number of ground object types in the polarized SAR data, the size of the convolution kernels is 7 x 7, and the convolution step is 1 x 1;
and the softmax classification layer, wherein the node number of the output layer is set as the number of the ground object types in the polarized SAR data.
Further, in step S403, the content and parameters of each layer are as follows:
a category subdivision layer, which subdivides feature maps output by the feature extraction backbone network into single-class feature maps of each category according to the ground object categories in the polarized SAR image data;
multiple classes of full connection layers are connected in parallel, the full connection layers are respectively established for single-class feature maps of each class, the number of the parallel full connection layers is equal to the number of ground feature classes in the polarized SAR image, the number of nodes of each class of full connection layer is the number of pixels of the single-class feature maps of the class, and the output of each full connection layer is the clustering center of the class;
and the clustering layer is used for realizing multi-class clustering by using the following formula according to the single-class feature map and the clustering center of each class:
wherein, i is 1, …, N, SiIs a cost sensitive coefficient, i represents the ground object class in the polarized SAR image data, N is the total number of the ground object class in the polarized SAR image data, xiNumber of pixels of single-class feature map of each class which is output by class subdivision, CiIs the clustering center, M, of each class of parallel full-link outputiIs the number of pixel points per class predicted by the softmax classification layer.
Specifically, the steps of training the cost sensitivity aided learning model are as follows:
s501, setting training parameters of a cost sensitivity auxiliary learning model;
s502, inputting the training sample set selected previously and the corresponding label into a cost sensitivity auxiliary learning model, and calculating to obtain a loss function value l;
s503, updating each layer parameter of the cost sensitivity auxiliary learning model by using a loss function value l and a back propagation algorithm according to the set learning rate;
and S504, repeating the steps S502 to S503 for 10000 times, finishing the training of the cost sensitivity auxiliary learning model, and storing the updated network model parameters.
Further, in step S501, the learning rate is 0.1, the number of iterations is 10000, and the loss function is L2 regularization loss function L, which is calculated as follows:
wherein y represents the class probability of the training sample output by the cost sensitivity auxiliary learning model,
real pixel labels representing training samples.
Specifically, the correlation matrix | T | after the modulus extraction is input into a trained cost sensitivity auxiliary learning model to obtain a final prediction label of the polarized SAR image data, and meanwhile, the classification accuracy is calculated.
Specifically, according to the spatial positions of the prediction label and the polarized SAR image data, a final classification result graph is drawn as follows:
the classification result color map of the polarized SAR image data is output by using colors having RGB values of [255, 255, 255], [255, 0, 0], [128, 0, 0], [171, 138, 80], [255, 255, 0], [183, 0, 255], [191, 191, 255], [90, 11, 255], [0, 252, 255], [0, 255, 0], [255, 182, 229], [255, 128, 0], [191, 255, 191], [255, 217, 157], [0, 131, 74], [0, 0, 255] to represent the feature types having the type numbers of 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15.
Compared with the prior art, the invention has at least the following beneficial effects:
the invention discloses an unbalanced polarization SAR terrain classification method based on cost sensitivity assisted learning, wherein a constructed cost sensitivity assisted learning model can automatically obtain high-level semantic features of a polarization SAR image through a feature extraction network from original data of a polarization coherent matrix, and overcomes the problems that the network in the prior art is deep in level and needs more label data as a training sample set, so that end-to-end semi-supervised classification is realized by directly starting from the original data under the condition of a small amount of labeled samples.
Furthermore, the cost sensitive coefficient is adopted, and is updated on line through the classification result of each iteration, so that the problem of unbalanced ground object types in the polarized SAR image is solved, the small targets are classified more conveniently, and the classification performance of the small targets in the polarized SAR image is improved by fully utilizing the labeled sample information.
Furthermore, a clustering method is adopted, feature maps output by the network are extracted by using features, clustering centers of each class are searched according to the clustering features of the feature maps, the distinguishing performance of the class features of the unbalanced ground object types is improved, the classification network is assisted to better learn and classify, and the classification precision of the model is improved.
Furthermore, in the process of training the cost sensitivity assisted learning model, the L2 regularization loss function and the back propagation algorithm enable the feature extraction network to learn the feature map of the polarization SAR, the classification network obtains the cost sensitivity coefficient while realizing the feature classification, the cost sensitivity coefficient is used for the clustering network to better search the clustering center, the clustering network in turn assists the feature extraction network to better learn the features, and meanwhile the performance of the classification network is improved.
Furthermore, the learning rate of the model is set to be 0.1, so that the network can be converged quickly, the loss function reaches the minimum value, and the training speed of the network is improved; by setting the iteration times to 10000 times, the robustness of the network can be improved, and the trained model weight has a better classification effect.
Furthermore, different categories of the polarized SAR image data are visualized through RGB values, and then the model classification result graph is compared with the manual marking graph, so that the performance of the model and the classification effect of the model on different categories are visually represented, and subsequent improvement is facilitated.
In conclusion, the cost sensitivity auxiliary learning model established by the invention not only has simple network hierarchy and less required label data sets, but also can solve the problem of unbalanced ground objects in the polarized SAR ground object classification, realize better classification of small targets and ensure higher classification accuracy.
The technical solution of the present invention is further described in detail by the accompanying drawings and embodiments.
Drawings
FIG. 1 is a flow chart of an implementation of the present invention;
FIG. 2 is a network structure diagram of a cost-sensitive aided learning model constructed in the present invention;
FIG. 3 is an artificial labeling diagram of a polarized SAR image to be classified in the present invention;
fig. 4 is a diagram of the classification result of the polarized SAR image to be classified by the present invention.
Detailed Description
The invention provides a cost sensitivity assisted learning-based classification method for unbalanced polarized SAR terrain, which comprises the steps of firstly inputting polarized SAR image data; then selecting a training data set and building a cost sensitivity auxiliary learning model; training a cost sensitivity auxiliary learning model; classifying the polarized SAR images to obtain a prediction label and classification precision; and finally, drawing a final classification result graph according to the space positions of the prediction label vector and the test sample. On the basis of cost sensitive clustering, the method solves the problem of unbalanced ground object types of the polarized SAR image data, optimizes the structural hierarchy of the model, improves the classification precision of the polarized SAR data of the polarized synthetic aperture radar, and can achieve the end-to-end classification effect. The method can be applied to target classification, detection and identification of the polarized SAR image.
Please refer to fig. 1, which illustrates an unbalanced polarimetric SAR terrain classification method based on cost-sensitive assisted learning according to the present invention, comprising the following steps:
s1, inputting a polarized SAR image to be classified and real ground feature label information corresponding to the polarized SAR image;
s2, taking an absolute value of a polarization coherent matrix T of the SAR image data to be classified to obtain a modulus matrix | T |;
s3, selecting a training sample set:
taking the polarized coherent matrix after modulus as a training sample, and randomly selecting 10% of pixels from all pixels with real labels of a polarized SAR image as label information;
s4, building a cost sensitivity auxiliary learning model;
referring to fig. 2, the cost-sensitive assisted learning model is composed of a main network and two branch networks, wherein the main network is a feature extraction network, one of the branch networks is a classification network, and the other of the branch networks is an assisted clustering network. The feature extraction network is used for learning semantic features of the polarized SAR image data, the feature extraction network is assisted to extract information of the polarized SAR image data, meanwhile, the classification network learning is assisted, so that the classification result is more accurate, and the cost sensitive coefficient is calculated through the result of the classification network and the clustering center is assisted to be more accurately found through the clustering learning.
S401, constructing a feature extraction backbone network:
the feature extraction network is divided into four layers in total, and the structure of the feature extraction network is divided into the following layers in sequence: an input layer, a first convolution layer, a second convolution layer, a third convolution layer;
the contents and parameters of each layer are set as follows:
an input layer, the number of nodes of which is set to the dimension of the input matrix | T |;
the number of convolution kernels is set to be 50, the size of the convolution kernels is set to be 5 x 5, and the convolution step size is 1 x 1;
the number of convolution kernels of the second convolution layer is set to be 30, the size of the convolution kernels is set to be 5 x 5, and the convolution step size is 1 x 1;
the number of convolution kernels of the third convolution layer is set to be 9, the size of the convolution kernels is 5 x 5, and the convolution step size is 1 x 1;
s402, building a classification network of one of the branches:
the classification network is divided into two layers in total, and the structure of the classification network is as follows: a fourth convolution layer, a classification layer;
the contents and parameters of the layers are set as follows:
the number of convolution kernels is set as the number of ground object types in the polarized SAR data, the size of the convolution kernels is 7 x 7, and the convolution step is 1 x 1;
and the softmax classification layer, wherein the node number of the output layer is set as the number of the ground object types in the polarized SAR data.
S403, building an auxiliary clustering network of one of the branches:
the auxiliary clustering network is divided into three layers in total, and the structure of the auxiliary clustering network sequentially comprises a category subdivision layer, a multi-category parallel full-connection layer and a clustering layer;
the contents and parameters of the layers are set as follows:
a category subdivision layer, which subdivides feature maps output by the feature extraction backbone network into single-class feature maps of each category according to the ground object categories in the polarized SAR image data;
multiple classes of full connection layers are connected in parallel, the full connection layers are respectively established for single-class feature maps of each class, the number of the parallel full connection layers is equal to the number of ground feature classes in the polarized SAR image, the number of nodes of each class of full connection layer is the number of pixels of the single-class feature maps of the class, and the output of each full connection layer is the clustering center of the class;
and the clustering layer is used for realizing multi-class clustering by using the following formula according to the single-class feature map and the clustering center of each class:
wherein, i is 1, …, N, SiIs a cost sensitive coefficient, i represents the ground object class in the polarized SAR image data, N is the total number of the ground object class in the polarized SAR image data, xiNumber of pixels of single-class feature map of each class which is output by class subdivision, CiIs the clustering center, M, of each class of parallel full-link outputiIs the number of pixel points per class predicted by the softmax classification layer.
S5 training cost sensitivity auxiliary learning model
S501, setting training parameters of a cost sensitivity auxiliary learning model, wherein the learning rate is 0.1, the iteration times are 10000, and a loss function is L2, and a regularization loss function L is as follows:
wherein y represents the class probability of the training sample output by the cost sensitivity auxiliary learning model,
real pixel labels representing training samples;
s502, inputting the training sample set selected previously and the corresponding label into a cost sensitivity auxiliary learning model, and calculating to obtain a loss function value l;
s503, updating each layer parameter of the cost sensitivity auxiliary learning model by using a loss function value l and a back propagation algorithm according to the set 0.1 learning rate;
and S504, repeating the steps S502 to S503 for 10000 times, finishing the training of the cost sensitivity auxiliary learning model, and storing the updated network model parameters.
S6, classifying the polarized SAR image by using the trained model weight value
Inputting the acquired correlation matrix | T | into a trained cost sensitivity auxiliary learning model to obtain a final prediction label of the polarized SAR image data, and calculating the classification accuracy;
and S7, outputting a visual classification result graph of the whole graph.
And drawing a final classification result graph according to the spatial positions of the prediction label and the polarized SAR image data. Each pixel in the final classification result of the polarized SAR image data is identified with a different color, which means that the color with RGB values of [255, 255, 255], [255, 0, 0], [128, 0, 0], [171, 138, 80], [255, 255, 0], [183, 0, 255], [191, 191, 255], [90, 11, 255], [0, 252, 255], [0, 255, 0], [255, 182, 229], [255, 128, 0], [191, 255, 191], [255, 217, 157], [0, 131, 74], and [0, 0, 255] represents a terrain category with a category number of 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, and the classification result of the polarized SAR image data is output by this method.
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, 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 some, but not all, embodiments of the present invention. The components of the embodiments of the present invention generally described and illustrated in the figures herein may be arranged and designed in a wide variety of different configurations. Thus, the following detailed description of the embodiments of the present invention, presented in the figures, is not intended to limit the scope of the invention, as claimed, but is merely representative of selected embodiments of the invention. 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 effects of the present invention can be further illustrated by the following simulations:
1. simulation conditions and content
The simulation experiment of the invention is carried out under the Intel (R) core (TM) i5-7500 CPU with the main frequency of 3.4GHz, GTX1060-6GD5 with the core frequency of 1569 and 1784MHz, a hardware environment with the internal memory of 8GB and a software environment of Tensorflow.
The experimental data of the invention is farmland data in Flevoland area obtained by AIRSAR in 1989, as shown in figure 3, the image size is 750 multiplied by 1024, and the image size corresponds to 15 different ground objects; fig. 3 (a) is a real map of the polarized SAR image of the to-be-classified map in the present invention, (b) in fig. 3 is a real ground feature label map of the polarized SAR image to be classified, and (c) in fig. 3 different colors represent different ground feature classes; FIG. 4 is a graph of the classification results of a polarized SAR image to be classified using the present invention; table 1 shows the classification accuracy comparisons using CV-RNN, RV-CNN and the half invention.
2. Analysis of simulation results
TABLE 1 comparison of classification accuracies
As can be seen from the table 1, compared with the existing CV-CNN and RV-CNN semi-supervised classification methods, the method of the invention obtains higher classification accuracy, and proves the excellent effect of the method of the invention in polarimetric SAR image semi-supervised classification; in addition, in the classification of single class, the classification precision of the method is generally higher than that of CV-RNN and RV-CNN, and the problem of unbalanced ground object class in the polarized SAR image is solved.
As can be seen from FIG. 4, the method of the present invention realizes an end-to-end training method in the training process, so that the misclassification of the classification result diagram is less, and the classification result is more accurate.
The above-mentioned contents are only for illustrating the technical idea of the present invention, and the protection scope of the present invention is not limited thereby, and any modification made on the basis of the technical idea of the present invention falls within the protection scope of the claims of the present invention.
Claims (6)
1. The method for classifying the unbalanced polarization SAR terrain based on cost sensitivity assisted learning is characterized by comprising the steps of firstly inputting a polarization SAR image to be classified and real terrain label information corresponding to the polarization SAR image to be classified; taking an absolute value of a polarization coherent matrix T of the to-be-classified polarization SAR image to obtain a correlation matrix | T | after modulus taking; then selecting a training sample set; building a cost sensitivity auxiliary learning model, wherein the step of building the cost sensitivity auxiliary learning model is as follows:
s401, building a feature extraction backbone network, wherein the feature extraction backbone network is divided into four layers, and the structure of the feature extraction backbone network is divided into the following layers in sequence: the input layer, the first convolution layer, the second convolution layer and the third convolution layer are set, and the content and parameters of each layer are as follows:
an input layer, the number of nodes of which is set to the dimension of the input correlation matrix | T |;
the number of convolution kernels is set to be 50, the size of the convolution kernels is set to be 5 x 5, and the convolution step size is 1 x 1;
the number of convolution kernels of the second convolution layer is set to be 30, the size of the convolution kernels is set to be 5 x 5, and the convolution step size is 1 x 1;
the number of convolution kernels of the third convolution layer is set to be 9, the size of the convolution kernels is 5 x 5, and the convolution step size is 1 x 1;
s402, building a classification network of one of the branches, which is divided into two layers, wherein the structure of the classification network is as follows: the fourth convolution layer, softmax classification layer, the contents and parameters of each layer are as follows:
the number of convolution kernels is set to be the number of ground object types in the polarized SAR data to be classified, the size of the convolution kernels is 7 x 7, and the convolution step length is 1 x 1;
the node number is set as the number of surface feature types in the polarized SAR data to be classified;
s403, building an auxiliary clustering network of one of the branches, wherein the auxiliary clustering network is divided into three layers, the structure of the auxiliary clustering network is a category subdivision layer, a multi-category parallel full-connection layer and a clustering layer in sequence, and the content and parameters of each layer are as follows:
a category subdivision layer, which subdivides the feature map output by the feature extraction backbone network into single-class feature maps of each category according to the ground object category in the polarized SAR image data to be classified;
multiple classes of full connection layers are connected in parallel, the full connection layers are respectively established for single-class feature maps of each class, the number of the parallel full connection layers is equal to the number of ground object classes in the polarized SAR images to be classified, the number of nodes of each class of full connection layers is the number of pixels of the single-class feature maps of the class, and the output of each full connection layer is the clustering center of the class;
and the clustering layer is used for realizing multi-class clustering by using the following formula according to the single-class feature map and the clustering center of each class:
wherein, i is 1, …, N, SiIs a cost sensitive coefficient, i represents the ground object class in the polarized SAR image data, N is the total number of the ground object class in the polarized SAR image data, xiNumber of pixels of single-class feature map of each class which is output by class subdivision, CiIs the clustering center, M, of each class of parallel full-link outputiThe number of pixels of each type predicted by the softmax classification layer; after a cost sensitivity auxiliary learning model is trained, classifying the polarized SAR images to be classified; and finally, outputting a visual classification result graph of the whole graph.
2. The cost-sensitive assistant learning-based method for classifying the unbalanced polarization SAR terrain features in claim 1, wherein a polarization coherent matrix after modulus taking is used as a training sample, and 10% of pixels are randomly selected as label information from all pixels with real labels of a polarized SAR image to be classified.
3. The method for classifying the unbalanced polarization SAR terrain based on cost-sensitive aided learning of claim 1, wherein the step of training the cost-sensitive aided learning model is as follows:
s501, setting training parameters of a cost sensitivity auxiliary learning model;
s502, inputting the training sample set selected previously and the corresponding label into a cost sensitivity auxiliary learning model, and calculating to obtain a loss function value l;
s503, updating each layer parameter of the cost sensitivity auxiliary learning model by using the loss function value l and a back propagation algorithm according to the set learning rate;
and S504, repeating the steps S502 to S503 for 10000 times, finishing the training of the cost sensitivity auxiliary learning model, and storing the updated network model parameters.
4. The method for classifying the features of the unbalanced polarization SAR based on the cost-sensitive assisted learning of claim 3, wherein in step S501, the learning rate is 0.1, the iteration number is 10000, and the loss function is L2 regularization loss function L, which is calculated as follows:
wherein y represents the class probability of the training sample output by the cost sensitivity auxiliary learning model,
real pixel labels representing training samples.
5. The method for classifying the unbalanced polarization SAR terrain based on cost-sensitive aided learning of claim 1, wherein the modulo correlation matrix | T | is input into a trained cost-sensitive aided learning model to obtain a final prediction label of the polarized SAR image data to be classified, and meanwhile, the classification accuracy is calculated.
6. The cost-sensitive assistant learning-based unbalanced polarized SAR terrain classification method according to claim 1 is characterized in that according to the spatial positions of the prediction labels and the polarized SAR image data, a final classification result graph is drawn as follows:
the classification result color map of the polarized SAR image data is output by using colors having RGB values of [255, 255, 255], [255, 0, 0], [128, 0, 0], [171, 138, 80], [255, 255, 0], [183, 0, 255], [191, 191, 255], [90, 11, 255], [0, 252, 255], [0, 255, 0], [255, 182, 229], [255, 128, 0], [191, 255, 191], [255, 217, 157], [0, 131, 74], [0, 0, 255] to represent the feature types having the type numbers of 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910198273.4A CN109934292B (en) | 2019-03-15 | 2019-03-15 | Unbalanced polarization SAR terrain classification method based on cost sensitivity assisted learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910198273.4A CN109934292B (en) | 2019-03-15 | 2019-03-15 | Unbalanced polarization SAR terrain classification method based on cost sensitivity assisted learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109934292A CN109934292A (en) | 2019-06-25 |
| CN109934292B true CN109934292B (en) | 2021-04-09 |
Family
ID=66987292
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910198273.4A Active CN109934292B (en) | 2019-03-15 | 2019-03-15 | Unbalanced polarization SAR terrain classification method based on cost sensitivity assisted learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109934292B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111325256A (en) * | 2020-02-13 | 2020-06-23 | 上海眼控科技股份有限公司 | Vehicle appearance detection method and device, computer equipment and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103824084A (en) * | 2014-03-12 | 2014-05-28 | 西安电子科技大学 | Polarimetric SAR (Synthetic Aperture Radar) image classification method based on SDIT (Secretome-Derived Isotopic Tag) and SVM (Support Vector Machine) |
| CN104156728A (en) * | 2014-07-14 | 2014-11-19 | 西安电子科技大学 | Polarized SAR image classification method based on stacked code and softmax |
| CN107122753A (en) * | 2017-05-08 | 2017-09-01 | 西安电子科技大学 | SAR target discrimination methods based on integrated study |
| CN109064455A (en) * | 2018-07-18 | 2018-12-21 | 清华大学深圳研究生院 | A kind of classification method of the breast ultrasound Image Multiscale fusion based on BI-RADS |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6552677B2 (en) * | 2001-02-26 | 2003-04-22 | Time Domain Corporation | Method of envelope detection and image generation |
-
2019
- 2019-03-15 CN CN201910198273.4A patent/CN109934292B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103824084A (en) * | 2014-03-12 | 2014-05-28 | 西安电子科技大学 | Polarimetric SAR (Synthetic Aperture Radar) image classification method based on SDIT (Secretome-Derived Isotopic Tag) and SVM (Support Vector Machine) |
| CN104156728A (en) * | 2014-07-14 | 2014-11-19 | 西安电子科技大学 | Polarized SAR image classification method based on stacked code and softmax |
| CN107122753A (en) * | 2017-05-08 | 2017-09-01 | 西安电子科技大学 | SAR target discrimination methods based on integrated study |
| CN109064455A (en) * | 2018-07-18 | 2018-12-21 | 清华大学深圳研究生院 | A kind of classification method of the breast ultrasound Image Multiscale fusion based on BI-RADS |
Non-Patent Citations (1)
| Title |
|---|
| SAR目标鉴别中的非平衡数据问题研究;吕翠文;《中国优秀硕士学位论文全文数据库》;20190215(第2期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109934292A (en) | 2019-06-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Lei et al. | Intelligent fault detection of high voltage line based on the Faster R-CNN | |
| CN109919108B (en) | Remote sensing image rapid target detection method based on deep hash auxiliary network | |
| CN108038445B (en) | SAR automatic target identification method based on multi-view deep learning framework | |
| Costea et al. | Creating roadmaps in aerial images with generative adversarial networks and smoothing-based optimization | |
| CN111222434A (en) | Method for obtaining evidence of synthesized face image based on local binary pattern and deep learning | |
| CN110097101B (en) | Remote sensing image fusion and coastal zone classification method based on improved reliability factor | |
| CN112347970B (en) | Remote sensing image ground object identification method based on graph convolution neural network | |
| CN110633708A (en) | Deep network significance detection method based on global model and local optimization | |
| CN106611423B (en) | SAR image segmentation method based on ridge ripple filter and deconvolution structural model | |
| CN110598564B (en) | OpenStreetMap-based high-spatial-resolution remote sensing image transfer learning classification method | |
| CN113033520A (en) | Tree nematode disease wood identification method and system based on deep learning | |
| CN108399420A (en) | A kind of visible light naval vessel false-alarm elimination method based on depth convolutional network | |
| CN113223042A (en) | Intelligent acquisition method and equipment for remote sensing image deep learning sample | |
| Chen et al. | Employing deep learning for automatic river bridge detection from SAR images based on adaptively effective feature fusion | |
| CN113723492A (en) | Hyperspectral image semi-supervised classification method and device for improving active deep learning | |
| CN113988147A (en) | Multi-label classification method and device for remote sensing image scene based on graph network, and multi-label retrieval method and device | |
| CN116977633A (en) | Feature element segmentation model training method, feature element segmentation method and device | |
| CN107292268A (en) | The SAR image semantic segmentation method of quick ridge ripple deconvolution Structure learning model | |
| CN114579794A (en) | Multi-scale fusion landmark image retrieval method and system based on feature consistency suggestion | |
| CN109934292B (en) | Unbalanced polarization SAR terrain classification method based on cost sensitivity assisted learning | |
| Du et al. | Training SegNet for cropland classification of high resolution remote sensing images | |
| CN117011714A (en) | Hyperspectral image classification method based on pseudo tag assistance | |
| CN116910571A (en) | Open-domain adaptation method and system based on prototype comparison learning | |
| CN113627480B (en) | Polarization SAR image classification method based on reinforcement learning | |
| Lei et al. | Land cover classification for remote sensing imagery using conditional texton forest with historical land cover map |
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 |