CN112949682B - SAR image classification method for feature level statistical description learning - Google Patents
SAR image classification method for feature level statistical description learning Download PDFInfo
- Publication number
- CN112949682B CN112949682B CN202110112799.3A CN202110112799A CN112949682B CN 112949682 B CN112949682 B CN 112949682B CN 202110112799 A CN202110112799 A CN 202110112799A CN 112949682 B CN112949682 B CN 112949682B
- Authority
- CN
- China
- Prior art keywords
- sar image
- feature
- vector
- layer
- statistical
- 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.)
- Expired - Fee Related
Links
- 238000000034 method Methods 0.000 title claims abstract description 40
- 239000013598 vector Substances 0.000 claims abstract description 51
- 238000006243 chemical reaction Methods 0.000 claims abstract description 5
- 230000006870 function Effects 0.000 claims description 11
- 150000001875 compounds Chemical class 0.000 claims description 10
- 239000011159 matrix material Substances 0.000 claims description 9
- 238000013507 mapping Methods 0.000 claims description 8
- 238000011176 pooling Methods 0.000 claims description 5
- 239000002131 composite material Substances 0.000 claims 1
- 238000007619 statistical method Methods 0.000 abstract description 12
- 230000008569 process Effects 0.000 abstract description 8
- 239000000284 extract Substances 0.000 abstract description 6
- 238000013527 convolutional neural network Methods 0.000 description 18
- 238000000605 extraction Methods 0.000 description 7
- 230000001427 coherent effect Effects 0.000 description 5
- 230000007246 mechanism Effects 0.000 description 5
- 238000005457 optimization Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 4
- 238000002592 echocardiography Methods 0.000 description 4
- 238000003384 imaging method Methods 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 3
- 238000009795 derivation Methods 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 230000007547 defect Effects 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- 230000004913 activation Effects 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000010191 image analysis Methods 0.000 description 1
- 238000013508 migration Methods 0.000 description 1
- 230000005012 migration Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 238000013179 statistical model Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 238000013519 translation 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
- G06F18/24—Classification techniques
-
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
Abstract
The invention discloses a SAR image classification method for feature level statistical description learning, which comprises the following steps: inputting the target SAR image into an SAR image classification network; extracting characteristic elements with middle-layer semantics of the target SAR image by the convolutional layer; the feature statistics layer extracts a statistic element vector of the target SAR image based on feature elements with middle-layer semantics; the nonlinear and linear conversion layer generates a feature level statistical description vector of the target SAR image based on the statistical primitive vector; and the Softmax layer generates a classification result of the target SAR image based on the feature level statistical description vector. Compared with the traditional CNN method, the SAR image classification method based on the feature level statistics is not only dedicated to structural feature learning of the SAR image, but also particularly considers feature level statistics characteristics of the SAR image in the feature learning process, integrates feature learning and statistical analysis, and can effectively solve the problem of insufficient generalization capability when the CNN method is used for SAR image classification.
Description
Technical Field
The invention relates to the technical field of image processing, in particular to an SAR image classification method for feature level statistical description learning.
Background
Synthetic Aperture Radar (SAR) can acquire rich information of a ground object target all day long, but a coherent imaging mechanism of the SAR enables an SAR image to present an inherent random image mode, so that statistical analysis of the SAR image is important for extracting information such as geography, biology and the like. Under a coherent imaging mechanism, each pixel point of the SAR image is formed by coherent superposition of a plurality of scattering center echoes in a corresponding resolution unit. Due to the randomness of the distribution of the positions of the scattering centers in the resolution unit, the echoes of the scattering centers have random phases, so that the echoes of a plurality of scattering centers also have randomness after being coherently superposed and synthesized. The coherent imaging mechanism enables the SAR image to show a random fluctuation phenomenon at a pixel level, and the SAR echo signal has an extremely low signal-to-noise ratio. Furthermore, the backscattered echoes of the target under the coherent imaging mechanism are very sensitive to the target properties such as shape, orientation, dielectric properties and humidity, and the backscattering of the terrestrial target generally tends to be highly complex and highly random. In addition, since the ground object itself usually has diversity, the spatial distribution of its image elements is also random. Therefore, statistical analysis is crucial for information extraction of SAR images. At present, a processing method based on statistical analysis is widely applied to SAR image analysis. Statistical modeling is an effective method for statistical analysis of low-resolution SAR images, such as K-distribution and Gamma mixed distribution models. The above statistical model is a pixel-level based statistical analysis method, which is used for statistical modeling of the SAR echo signal amplitude (or intensity). For high resolution SAR images, many texture-based analysis methods have emerged as structures and geometric features in the images become more prominent. For example: and extracting a second-order statistic analysis method for SAR image description based on the gray level co-occurrence matrix. And similarly, performing statistical analysis on the low-layer or middle-layer characteristic elements based on a scale invariant feature extraction (SAR-SIFT) and wavelet element and sparse coding method, and taking a statistical histogram of the characteristic elements as the statistical description of the SAR image. The characteristic level statistical analysis method aims at performing statistical analysis on SAR images from different characteristic level layers.
In recent years, a convolutional neural network method based on autonomous learning high-level semantic feature description is gradually applied to an interpretation task of SAR images. In order to solve the problem of insufficient robustness of a CNN (Convolition Neural Network, CNN) model to speckle noise, researchers provide a CNN method based on speckle noise invariance constraint, and extract robust CNN feature description for SAR target identification. And (3) aiming at the over-fitting problem of the CNN model under the condition of limited training samples, researchers propose a full convolution network based on a Dropout layer to be used for the classification task of the SAR image. Similarly, researchers propose a deep memory CNN method based on information recording and parameter migration technology, which is used for relieving the over-fitting problem in the SAR target recognition task under a small sample. In addition, the related research starts to consider the unique special properties of the SAR data, and researches the CNN method which embodies the inherent properties of the SAR data. The analysis method based on the CNN autonomous feature learning is an important milestone for the development of the SAR image processing field and gradually becomes an important direction for research. However, the CNN feature extraction method does not consider the feature level statistical characteristics of the SAR image, and generally has a problem of insufficient generalization capability.
Therefore, there is a need for a new technical solution for solving the problem of insufficient generalization capability when classifying SAR images by using the CNN method.
Disclosure of Invention
Aiming at the defects of the prior art, the problems to be solved by the invention are as follows: the generalization capability of the CNN method for SAR image classification is insufficient.
In order to solve the technical problems, the invention adopts the following technical scheme:
a SAR image classification method for feature level statistical description learning comprises the following steps:
s1, inputting the target SAR image into an SAR image classification network;
s2, extracting characteristic elements with middle-layer semantics of the target SAR image by the convolution layer in the SAR image classification network;
s3, extracting a statistical element vector of the target SAR image based on a characteristic element with middle-layer semantics by a characteristic statistical layer in the SAR image classification network;
s4, generating a feature level statistical description vector of the target SAR image based on the statistical primitive vector by a nonlinear and linear conversion layer in the SAR image classification network;
and S5, generating a classification result of the target SAR image based on the characteristic-level statistical description vector by a Softmax layer in the SAR image classification network.
Preferably, in step S2, the feature primitives z with middle-level semantics of the target SAR image are extracted based on the following formula:
in the formula (I), the compound is shown in the specification,
representing the complex function containing the ReLU and the max pooling, w and b representing the weight vector and the bias term of the convolutional layer, respectively, and x representing the input of the convolutional layer.
Preferably, in step S3, the statistical primitive vector S of the target SAR image is extracted based on the following formula:
s＝[μ,σ2]T
in the formula (I), the compound is shown in the specification,
and
respectively representing first-order and second-order statistical element vectors of corresponding characteristic channels;
and
respectively represent z'jCorresponding mean and variance, j 1,2, … C, feature maps of feature primitives with mid-level semantics are denoted as
H. W and C respectively represent the height, width and number of channels of Z, Zj＝[z1,z2,...,zm]TFor the vectorized representation of the jth characteristic channel,
preferably, in step S4, the feature level statistical description vector h of the target SAR image is generated based on the following formula:
h＝ReLU(Ws+B)
in the formula (I), the compound is shown in the specification,
and
respectively representing a weight matrix and an offset vector for high-dimensional feature mapping, wherein P is more than 2C.
Preferably, the SAR image classification network is optimized based on the following formula:
in the formula, wkAnd bkWeight matrix and bias vector, y, representing the k-th convolutional layernDenotes the nth sample XnThe label of (a) is used,
denotes the nth sample XnN denotes the total number of samples,
represents a cross-entropy loss function of the entropy of the sample,
Ynindicating label ynVector based on One-Hot coding, anRepresenting the output vector of the Softmax layer.
The SAR image shows randomness of feature level, but the CNN feature extraction does not consider the feature level statistical characteristics of the SAR image. In order to overcome the defects of the prior art, the invention provides the SAR image classification method for feature level statistical description learning. Firstly, extracting feature primitives with middle-layer semantics based on the multilayer convolution layer to describe the structure information of the input image, wherein the middle-layer feature primitives describe image modes with different middle-layer semantics in the SAR image. Then, the feature statistical layer performs statistical analysis on the feature map of the middle-layer feature primitive, and extracts the statistical characteristics of the middle-layer features including first-order and high-order statistical primitive descriptions. And finally, learning the characteristic-level statistical description of the SAR image by the nonlinear and linear conversion layers based on the statistical primitives under the constraint of minimized classification errors, and obtaining a classification result through a Softmax layer. Compared with the traditional CNN method, the SAR image classification method based on the feature level statistics is not only dedicated to structural feature learning of the SAR image, but also particularly considers feature level statistics characteristics of the SAR image in the feature learning process, integrates feature learning and statistical analysis, and can effectively solve the problem of insufficient generalization capability when the CNN method is used for SAR image classification.
Drawings
For purposes of promoting a better understanding of the objects, aspects and advantages of the invention, reference will now be made in detail to the present invention as illustrated in the accompanying drawings, in which:
FIG. 1 is a schematic illustration of a feature level statistical description learned SAR image classification method disclosed in the present invention.
FIG. 2 is a schematic diagram of training optimization of the SAR image classification network disclosed in the present invention.
Fig. 3 is a schematic diagram of a convolutional layer feature learning process of the SAR image classification network disclosed by the present invention.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings.
As shown in fig. 1, the invention discloses a method for classifying an SAR image through feature level statistical description learning, which comprises the following steps:
a SAR image classification method for feature level statistical description learning comprises the following steps:
s1, inputting the target SAR image into an SAR image classification network;
s2, extracting characteristic elements with middle-layer semantics of the target SAR image by the convolution layer in the SAR image classification network;
s3, extracting a statistical element vector of the target SAR image based on a characteristic element with middle-layer semantics by a characteristic statistical layer in the SAR image classification network;
s4, generating a feature level statistical description vector of the target SAR image based on the statistical primitive vector by a nonlinear and linear conversion layer in the SAR image classification network;
and S5, generating a classification result of the target SAR image based on the characteristic-level statistical description vector by a Softmax layer in the SAR image classification network.
Compared with the traditional CNN method, the SAR image classification method based on the feature level statistics is not only dedicated to structural feature learning of the SAR image, but also particularly considers feature level statistics characteristics of the SAR image in the feature learning process, integrates feature learning and statistical analysis, and can effectively solve the problem of insufficient generalization capability when the CNN method is used for SAR image classification.
In step S2, in a specific implementation, the feature primitive z with middle-level semantics of the target SAR image is extracted based on the following formula:
in the formula (I), the compound is shown in the specification,
representing the complex function containing the ReLU and the max pooling, w and b representing the weight vector and the bias term of the convolutional layer, respectively, and x representing the input of the convolutional layer.
Under the non-strict definition, the learning process of the middle layer characteristic primitive based on the convolutional layer can be interpreted as the matching filtering in the broad sense. Specifically, the output z of the convolutional layer can be formulated as:
z＝wTx+b
wherein x represents the convolutional layer input, (. C)TRepresenting the transposed transform, w and b represent the weight vector and bias term of the convolutional layer, respectively. The offset term in the neglect equation is then wTx＝<w,x>Is an inner product operation in euclidean space that can be interpreted essentially as a broad matched filter. Then, the physical meaning of formula-based convolutional layer feature extraction can be interpreted as: by changing the weights w to match the implicit pattern in the data x. FIG. 3 illustrates a feature learning process for convolutional layersIt is intended that three weight vectors w for matching the hidden patterns in the input x are given1、w2And w3. It can be seen that the weight vector w2Is superior to w1And w3Due to w2Angle theta with input x2Less than weight vector w1And w3Angle theta with x1And theta3. In the invention, the problem of nonlinear mapping and characteristic translation invariance is considered in the characteristic extraction based on the convolution layer, so that a modified linear unit activation mapping (ReLU) and a maximum pooling process are added in the convolution layer.
With the increase of the number of the convolution layers, the SAR image classification network extracts SAR image feature primitives with gradually enhanced feature semantics. Similar to the standard CNN model, the convolutional layer of the SAR image classification network also employs a local connection and weight sharing mechanism. That is to say, the convolution layer output characteristic diagram of the SAR image classification network adopts a local connection mode with the previous layer output through the weight matrix, and each element in the output characteristic diagram shares the same weight. Therefore, each convolution output channel feature map of the SAR image classification network only describes one specific image pattern in the input image, i.e., the SAR image classification network extracts feature primitives for the input SAR image description. Moreover, as the number of convolution layers increases, the downsampling of the pooling process increases the receptive field of the current layer feature extraction, i.e. the feature semantics gradually increase. A global description above low level semantics and not an image may be considered a middle level semantic feature. Therefore, the convolution layer is added, and under the condition of controlling non-global image description, the SAR image classification network can extract the SAR image feature primitive with middle-layer semantics.
In step S3, in a specific implementation, the statistical primitive vector S of the target SAR image is extracted based on the following formula:
s＝[μ,σ2]T
in the formula (I), the compound is shown in the specification,
and
respectively representing first-order and second-order statistical element vectors of corresponding characteristic channels;
and
respectively represent z'jCorresponding mean and variance, j 1,2, … C, feature maps of feature primitives with mid-level semantics are denoted as
H. W and C respectively represent the height, width and number of channels of Z, Zj＝[z1,z2,…,zm]TFor the vectorized representation of the jth characteristic channel,
in step S4, in a specific implementation, the feature level statistical description vector h of the target SAR image is generated based on the following formula:
h＝ReLU(Ws+B)
in the formula (I), the compound is shown in the specification,
and
respectively representing a weight matrix and an offset vector for high-dimensional feature mapping, wherein P is more than 2C.
After the statistical primitive vector s of the middle-layer features is obtained, learning ethics are learned based on statistical description, and feature-level statistical description of the SAR image is learned based on s under the constraint of minimized classification errors. Specifically, the method autonomously maps s to a high-dimensional space through nonlinear and linear transformation, and autonomously constructs a function of a feature-level statistical primitive under the constraint of minimized classification errors, so as to learn the feature-level statistical description of the SAR image.
In the invention, in order to ensure the high-dimensional mapping of the feature level statistical description vector h, the dimension of the high-dimensional feature space needs to satisfy P & gt 2C. And changing the weight matrix and the bias vector of the high-dimensional mapping to obtain the feature level statistical description vector h under different mapping relations. Since h is a function of the statistical primitive, the vector h actually constitutes a feature-level statistical distribution parameter space. Then, the description of the SAR image based on the vector h embodies the statistical properties of the SAR image in the feature space.
As shown in fig. 2, in specific implementation, the SAR image classification network is optimized based on the following formula:
in the formula, wkAnd bkWeight matrix and bias vector, y, representing the k-th convolutional layernDenotes the nth sample XnThe label of (a) is used,
denotes the nth sample XnN denotes the total number of samples,
represents a cross-entropy loss function of the entropy of the sample,
Ynindicating label ynVector based on One-Hot coding, anRepresenting the output vector of the Softmax layer.
The optimization problem described above can be solved by a stochastic gradient descent algorithm. In the stochastic gradient descent algorithm, solving the gradient of the loss function with respect to the optimization parameters is a key step of the optimization algorithm. The invention constructs a feature statistics layer from which the gradient of the loss function with respect to the input Z of the layer needs to be deduced in order to back-propagate the classification error. The error propagation for this feature statistics layer is derived below. Recording +zAs a function of loss
Gradient with respect to Z, i.e.
Partial differential of
The derivation of (c) takes into account two aspects:
(1) when Ws + B is less than or equal to 0, i.e. the characteristic vector h shown in the formula is 0, partial differential
(2) When Ws + B>Partial differential at 0
Comprises the following steps:
in the formula (I), the compound is shown in the specification,
representing loss function
Relative to the input partial differential of the Softmax layer, the partial differential
Can be found by a back propagation algorithm. For the
The following two cases are considered for derivation of (1):
(1) when q ≠ j (q ≠ 1, 2.., C),
(2) when the value of theta is equal to j,
comprises the following steps:
from the derivation of the above formula, the gradient can be determined
Then, the optimization of the SAR image classification network can be solved by a stochastic gradient descent algorithm.
Finally, it is noted that the above-mentioned embodiments illustrate rather than limit the invention, and that, while the invention has been described with reference to preferred embodiments thereof, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims.
Claims (3)
1. A SAR image classification method for feature level statistical description learning is characterized by comprising the following steps:
s1, inputting the target SAR image into an SAR image classification network;
s2, extracting characteristic elements with middle-layer semantics of the target SAR image by the convolution layer in the SAR image classification network;
s3, extracting a statistical element vector of the target SAR image based on a characteristic element with middle-layer semantics by a characteristic statistical layer in the SAR image classification network; in step S3, a statistical primitive vector S of the target SAR image is extracted based on the following formula:
s＝[μ,σ2]T
in the formula (I), the compound is shown in the specification,
and
respectively representing corresponding characteristic channelsFirst and second order statistical primitive vectors of (1); (.)TRepresenting a transposed transform;
and
respectively represent z'jCorresponding mean and variance, j 1,2, … C, feature maps of feature primitives with mid-level semantics are denoted as
H. W and C respectively represent the height, width and number of channels of Z, Zj＝[z1,z2,…,zm]TFor the vectorized representation of the jth characteristic channel,
s4, generating a feature level statistical description vector of the target SAR image based on the statistical primitive vector by a nonlinear and linear conversion layer in the SAR image classification network; in step S4, a feature level statistical description vector h of the target SAR image is generated based on the following formula:
h＝ReLU(Ws+B)
in the formula (I), the compound is shown in the specification,
and
respectively representing a weight matrix and a bias vector for high-dimensional feature mapping, wherein P is more than 2C;
and S5, generating a classification result of the target SAR image based on the characteristic-level statistical description vector by a Softmax layer in the SAR image classification network.
2. The method for classifying SAR images for feature-level statistical description learning as claimed in claim 1, wherein in step S2, the feature primitives z with middle-level semantics of the target SAR image are extracted based on the following formula:
in the formula (I), the compound is shown in the specification,
representing a composite function containing a ReLU and a maximum pooling, w and b respectively representing a weight vector and a bias term of the convolutional layer, and x representing the input of the convolutional layer; (.)TRepresenting a transposed transform.
3. The method of feature level statistical description learned SAR image classification of claim 1, wherein the SAR image classification network is optimized based on the following formula:
in the formula, wkAnd bkWeight matrix and bias vector, y, representing the k-th convolutional layernDenotes the nth sample XnThe label of (a) is used,
denotes the nth sample XnN denotes the total number of samples,
represents a cross-entropy loss function of the entropy of the sample,
Ynindicating label ynVector based on One-Hot coding, anRepresenting the output vector of the Softmax layer.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110112799.3A CN112949682B (en) | 2021-01-27 | 2021-01-27 | SAR image classification method for feature level statistical description learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110112799.3A CN112949682B (en) | 2021-01-27 | 2021-01-27 | SAR image classification method for feature level statistical description learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112949682A CN112949682A (en) | 2021-06-11 |
| CN112949682B true CN112949682B (en) | 2022-05-20 |
Family
ID=76238080
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110112799.3A Expired - Fee Related CN112949682B (en) | 2021-01-27 | 2021-01-27 | SAR image classification method for feature level statistical description learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112949682B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114241233B (en) * | 2021-11-30 | 2023-04-28 | 电子科技大学 | Nonlinear class group sparse representation real and false target one-dimensional range profile identification method |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102682306A (en) * | 2012-05-02 | 2012-09-19 | 武汉大学 | Wavelet pyramid polarization texture primitive feature extracting method for synthetic aperture radar (SAR) images |
| CN106408030A (en) * | 2016-09-28 | 2017-02-15 | 武汉大学 | SAR image classification method based on middle lamella semantic attribute and convolution neural network |
| CN107506726A (en) * | 2017-08-22 | 2017-12-22 | 武汉大学 | SAR image sorting technique based on quadratic form primitive multitiered network |
| CN109284786A (en) * | 2018-10-10 | 2019-01-29 | 西安电子科技大学 | The SAR image terrain classification method of confrontation network is generated based on distribution and structure matching |
| CN110111352A (en) * | 2019-03-18 | 2019-08-09 | 北京理工雷科电子信息技术有限公司 | One kind detecting false-alarm elimination method based on feature cascade SAR remote sensing images waters |
| CN110969199A (en) * | 2019-11-25 | 2020-04-07 | 贝壳技术有限公司 | Image classification method and device and storage medium |
| CN111385462A (en) * | 2018-12-28 | 2020-07-07 | 上海寒武纪信息科技有限公司 | Signal processing device, signal processing method and related product |
| CN111596978A (en) * | 2019-03-03 | 2020-08-28 | 山东英才学院 | Web page display method, module and system for lithofacies classification by artificial intelligence |
-
2021
- 2021-01-27 CN CN202110112799.3A patent/CN112949682B/en not_active Expired - Fee Related
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102682306A (en) * | 2012-05-02 | 2012-09-19 | 武汉大学 | Wavelet pyramid polarization texture primitive feature extracting method for synthetic aperture radar (SAR) images |
| CN106408030A (en) * | 2016-09-28 | 2017-02-15 | 武汉大学 | SAR image classification method based on middle lamella semantic attribute and convolution neural network |
| CN107506726A (en) * | 2017-08-22 | 2017-12-22 | 武汉大学 | SAR image sorting technique based on quadratic form primitive multitiered network |
| CN109284786A (en) * | 2018-10-10 | 2019-01-29 | 西安电子科技大学 | The SAR image terrain classification method of confrontation network is generated based on distribution and structure matching |
| CN111385462A (en) * | 2018-12-28 | 2020-07-07 | 上海寒武纪信息科技有限公司 | Signal processing device, signal processing method and related product |
| CN111596978A (en) * | 2019-03-03 | 2020-08-28 | 山东英才学院 | Web page display method, module and system for lithofacies classification by artificial intelligence |
| CN110111352A (en) * | 2019-03-18 | 2019-08-09 | 北京理工雷科电子信息技术有限公司 | One kind detecting false-alarm elimination method based on feature cascade SAR remote sensing images waters |
| CN110969199A (en) * | 2019-11-25 | 2020-04-07 | 贝壳技术有限公司 | Image classification method and device and storage medium |
Non-Patent Citations (2)
| Title |
|---|
| "Ship Classification in SAR Images Using a New Hybrid CNN–MLP Classifier";Foroogh Sharifzadeh;《Journal of the Indian Society of Remote Sensing》;20181029;第47卷;第551-562页 * |
| "全卷积网络结合改进的条件随机场-循环神经网络用于SAR图像场景分类";汤浩，何楚;《计算机应用》;20161210;第36卷(第12期);第3436-3441页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112949682A (en) | 2021-06-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110135267B (en) | Large-scene SAR image fine target detection method | |
| CN109086700B (en) | Radar one-dimensional range profile target identification method based on deep convolutional neural network | |
| Zhong et al. | Spectral–spatial residual network for hyperspectral image classification: A 3-D deep learning framework | |
| Deng et al. | SAR automatic target recognition based on Euclidean distance restricted autoencoder | |
| Chen et al. | CycleGAN-STF: Spatiotemporal fusion via CycleGAN-based image generation | |
| CN107145830B (en) | Hyperspectral image classification method based on spatial information enhancing and deepness belief network | |
| CN108009559B (en) | Hyperspectral data classification method based on space-spectrum combined information | |
| Lu et al. | 3-D channel and spatial attention based multiscale spatial–spectral residual network for hyperspectral image classification | |
| CN109375186B (en) | Radar target identification method based on depth residual error multi-scale one-dimensional convolution neural network | |
| Gao et al. | Hyperspectral image classification with pre-activation residual attention network | |
| CN113240040B (en) | Polarized SAR image classification method based on channel attention depth network | |
| CN108446582A (en) | Hyperspectral image classification method based on textural characteristics and affine propagation clustering algorithm | |
| CN111860124B (en) | Remote sensing image classification method based on space spectrum capsule generation countermeasure network | |
| CN107133648B (en) | One-dimensional range profile identification method based on adaptive multi-scale fusion sparse preserving projection | |
| CN112949682B (en) | SAR image classification method for feature level statistical description learning | |
| CN108460326B (en) | Hyperspectral image semi-supervised classification method based on sparse expression graph | |
| CN113139512A (en) | Depth network hyperspectral image classification method based on residual error and attention | |
| Valsalan et al. | Hyperspectral image classification model using squeeze and excitation network with deep learning | |
| Ge et al. | Adaptive hash attention and lower triangular network for hyperspectral image classification | |
| CN112215267A (en) | Hyperspectral image-oriented depth space spectrum subspace clustering method | |
| Qin et al. | Weakly supervised adversarial learning via latent space for hyperspectral target detection | |
| CN109145993B (en) | SAR image classification method based on multi-feature and non-negative automatic encoder | |
| Zhao et al. | Hyperspectral anomaly detection via memory‐augmented autoencoders | |
| CN116863347A (en) | High-efficiency and high-precision remote sensing image semantic segmentation method and application | |
| CN112131962B (en) | SAR image recognition method based on electromagnetic scattering characteristics and depth network characteristics |
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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20220520 |