CN115620038A - Common knowledge constrained remote sensing sample migration method - Google Patents
Common knowledge constrained remote sensing sample migration method Download PDFInfo
- Publication number
- CN115620038A CN115620038A CN202211151014.4A CN202211151014A CN115620038A CN 115620038 A CN115620038 A CN 115620038A CN 202211151014 A CN202211151014 A CN 202211151014A CN 115620038 A CN115620038 A CN 115620038A
- Authority
- CN
- China
- Prior art keywords
- sample
- feature
- sample data
- common
- data
- 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.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/762—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using clustering, e.g. of similar faces in social 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- 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 commonality knowledge constrained remote sensing sample migration method, which relates to the field of remote sensing image processing and comprises the following steps: acquiring the characteristics and the common characteristic space of the source domain sample and the target domain sample by using a characteristic extraction model; determining a commonality feature and a non-commonality feature using a feature clusterer; inputting the commonality characteristics and the source domain non-commonality characteristics added with random noise into a generator to generate a pseudo sample, inputting the pseudo sample into a discriminator, discriminating the pseudo sample according to the target domain sample data, and iteratively training an optimized generator to obtain a trained generator; and inputting the source domain sample data into a trained generator to generate a migration sample. The method realizes the sample migration from the source domain to the target domain under the constraint of the common characteristic, and avoids the occurrence of negative migration.
Description
Technical Field
The invention relates to the field of remote sensing image processing, in particular to a common knowledge constrained remote sensing sample migration method.
Background
In recent years, the rapid development of remote sensing technology has promoted the wide application of remote sensing technology in various fields. The real-time ground monitoring of the plurality of satellites provides massive multi-remote sensing image data support for the development of the whole remote sensing field, and lays the foundation for the rapid development of the remote sensing technology. The effective utilization of mass remote sensing image data is one of the important directions for the development of the remote sensing field.
The artificial intelligence technology has great development prospect as one of the most popular high and new technologies at present. In fact, the implementability of the artificial intelligence technology depends on the support of mass data and big data to a great extent, so that the artificial intelligence technology can also effectively utilize the mass data to realize various functions.
The artificial intelligence technology is used in the field of remote sensing, and the utilization rate of mass remote sensing image data can be greatly improved. However, most of the artificial intelligence applications in the remote sensing direction adopt a supervised learning or semi-supervised learning mode, so that massive remote sensing image data cannot be directly used, and the labeled remote sensing sample is required. The marked remote sensing sample is difficult to obtain, and besides the existing manual sample marking method which is high in precision, high in labor cost and low in efficiency, the method for marking the sample by using a machine learning method is researched, but the standard of large-scale engineering implementation level is not achieved. Therefore, how to maximize the effective utilization of the existing remote sensing labeling sample is one of the current research directions.
The transfer learning method in the remote sensing sample can effectively improve the utilization rate of the existing sample. The purpose of the transfer learning is to use sufficient labeled samples in a source domain for a target domain with few samples or no samples, and the sample characteristics between the source domain and the target domain have partial correlation or no correlation. An existing migration method, such as traadaboost, selects source domain sample data with high correlation for migration learning, and improves migration learning performance. However, for this method of optimizing source domain samples, the following problems are mainly encountered: even for the preferred sample, not all features in the sample are beneficial to the transfer learning, and some features with low correlation may negatively affect the transfer learning, even resulting in negative transfer, and it is necessary to ensure that the source domain and the target domain are sufficiently correlated.
Disclosure of Invention
The invention provides a remote sensing sample migration method based on common knowledge constraint, which can solve the problems in the prior art, and can obtain a generator under the common knowledge constraint by directly inputting common characteristics of source domain sample data and target domain sample data into a well-constructed generator and training the generator based on a target function while not influencing the distribution of the source domain sample data. And inputting the source domain sample data into the generator under the common knowledge constraint to generate the sample data of the fitting target domain, thereby realizing the sample migration under the common knowledge constraint and effectively avoiding the negative migration. Meanwhile, the whole sample migration process is end-to-end, and automatic adjustment of the model can be realized.
In order to achieve the technical purpose, the technical scheme of the invention is as follows:
a remote sensing sample migration method based on common knowledge constraint comprises the following steps:
s1, inputting source domain sample data and target domain sample data into a feature extraction model to obtain feature data and a public feature space of the source domain sample data and the target domain sample data;
s2, inputting the feature data of the source domain sample data and the target domain sample data into a feature clustering device, determining a common feature space and a non-common feature space, and extracting common features and source domain non-common features, wherein the common feature space and the non-common feature space are subspaces of the common feature space;
s3, inputting the common characteristic and the source domain non-common characteristic added with random noise into a generator to generate a pseudo sample;
s4, inputting the pseudo sample and target domain sample data into the discriminator, discriminating the pseudo sample according to the target domain sample data, and optimizing a generator according to a discrimination result and a target function;
s5, iterating the training processes from S3 to S4 until the target function converges;
and S6, inputting the source domain sample data into a trained generator to generate a migration sample.
Optionally, in step S2, inputting the feature data of the source domain sample data and the target domain sample data into a feature clustering device, and determining a common feature space and a non-common feature space, including:
clustering the source domain sample data and the target domain sample data on the public characteristic space to obtain k groups of mixed sample data x i I =1,.. K, wherein each set of mixed sample data includes a feature correlation therein;
mixing the mixed sample data x i A plurality of feature subspaces F mapped to the common feature space j In step (c), a sample-feature set (x) is obtained i ,F j )；
Analysis (x) i ,F j ) Distribution of (a):
if the distributions have correlation, the feature subspace F is divided into two parts j Dividing into a common characteristic space; if the distribution has no correlation, the feature subspace F is used j Partitioning into a non-common feature space.
Optionally, the discrimination manner that the distribution has correlation is as follows:
computing the same feature subspace F using a probability distribution distance metric algorithm j The fitting degree of the characteristic distribution of each group of mixed samples and other groups of mixed samples;
obtaining the same characteristic subspace F according to the fitting degree j Mixed sample feature of each group (x) i ,F j ) The overall degree of fit of (c);
and counting the number M of the mixed samples with the overall fitting degree larger than a first preset threshold, and judging that the feature subspace is distributed with correlation when the M is larger than or equal to a second preset threshold.
Optionally, the fitting degree is calculated according to a probability distribution distance measurement algorithm.
Optionally, the feature extraction model is:
adopting a machine learning model of a feature extraction operator;
or a model constructed by a convolutional neural network;
or a combined model of the machine learning model and a model constructed by a convolutional neural network.
Optionally, the feature extraction model is an encoder part of a convolutional auto-encoder constructed by a convolutional neural network;
accordingly, the generator is a decoder portion of the convolutional auto-encoder.
Optionally, the feature extraction model is symmetrical to the structure of the generator.
Optionally, the generator and the discriminator form a generation countermeasure network, the objective function is an objective function of the generation countermeasure network, and the objective function is:
wherein G is the generator, D is the discriminator, E is the expectation function, x is the pseudo-sample data generated by the generator, p data Is the probability that x is from the true data distribution, p g Is the probability of x from the generator output sample.
The invention provides a common knowledge constrained remote sensing sample migration method, which comprises the steps of extracting features of sample data of a source domain and a target domain to obtain feature data corresponding to the sample data, performing cluster analysis on the feature data to obtain common features and non-common features of the sample of the source domain and the sample of the target domain, inputting the common features and the noisy non-common features of the source domain and the sample of the target domain into a generator to generate a pseudo sample under the constraint of the common features, inputting the pseudo sample and the sample of the target domain into a discriminator, iteratively training the optimized generator according to a discrimination result and a target function to obtain a generator capable of migrating the features of the source domain to the target domain, and finally inputting the sample of the source domain into the generator to directly obtain a migration sample fitting the target domain. The invention has the beneficial effects that:
(1) With the technical support of the invention, the source domain sample data can be input into the generator under the constraint of the common knowledge to generate the sample data of the fitting target domain while the distribution of the source domain sample data is not influenced, so that the sample migration under the constraint of the common knowledge is realized, and the negative migration is effectively avoided.
(2) The sample migration framework constructed by the invention supports full-automatic training and adjustment of the model, and realizes an end-to-end sample migration process.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without creative efforts.
FIG. 1 is a schematic flow chart diagram of an embodiment of a method for remote sensing sample migration constrained by common knowledge according to the present invention;
FIG. 2 is a schematic diagram of data transmission of a sample migration model in an embodiment of the common knowledge constrained remote sensing sample migration method of the present invention;
fig. 3 is a schematic diagram of sample migration of a source domain sample by using a trained self-encoder in an embodiment of the common knowledge constraint remote sensing sample migration method of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived from the embodiments of the present invention by a person skilled in the art, are within the scope of the present invention.
Referring to fig. 1, fig. 1 is a schematic flow chart of an embodiment of a common knowledge constraint remote sensing sample migration method according to the present invention, and compared with a conventional migration learning mode, the method realizes generation of a sample fitting a target domain by inputting a source domain sample, realizes sample migration under common knowledge constraint, and effectively avoids negative migration, and the method includes the following steps:
s1, inputting source domain sample data and target domain sample data into a feature extraction model to obtain feature data and a public feature space of the source domain sample data and the target domain sample data.
It should be noted that the embodiment of the present invention is directed to a migration task of a remote sensing sample, and belongs to the category of isomorphic migration learning, and sample data of a source domain and a target domain are in the same feature space, that is, the common feature space.
Optionally, the feature extraction model is: and adopting a machine learning model of a characteristic extraction operator, or a convolution neural network model, or a combination model of the models.
In this embodiment, an encoder portion of a convolutional self-encoder constructed using a convolutional neural network is used as a feature extraction model, and a self-encoder is used as a powerful feature detector, so that learned input data can be efficiently expressed as features through unsupervised learning. The encoder of the convolution self-encoder adopts a three-layer convolution neural network structure, the number of convolution kernels of a first layer is 16, the size of the convolution kernels is 3 multiplied by 3, and the step length is 1; the number of the second layer of convolution kernels is 8, the size of the convolution kernels is 3 multiplied by 3, and the step length is 1; the number of convolution kernels in the third layer is 8, the size of the convolution kernels is 3 multiplied by 3, and the step size is 1. And connecting a 2 x 2 maximum pooling layer behind each convolution layer for dimension reduction and feature compression.
S2, inputting the feature data of the source domain sample data and the target domain sample data into a feature clustering device, determining a common feature space and a non-common feature space, and extracting common features and source domain non-common features, wherein the common feature space and the non-common feature space are subspaces of the common feature space.
It should be noted that the sample data of the source domain and the target domain are both in a common feature space, but the specific distribution situation is different, and the feature data may be divided into common features and non-common features according to the data distribution, where a subspace in which the common features are located is a common feature space, and a subspace in which the non-common features are located is a non-common feature space.
In this embodiment, on the common feature space, source domain sample data and target domain sample data are divided into k groups through collaborative clustering, and k groups of mixed sample data x are obtained i I = 1. There is a characteristic correlation within each set of mixed sample data. And normalizing the sample characteristic value, so that the subsequent analysis and input are facilitated to generate the countermeasure network. Mixing sample data x i Mapping to multiple feature subspaces F j In step (c), a sample-feature set (x) is obtained i ,F j ). Analysis (x) 1 ,F j ),(x 2 ,F j ),...,(x k ,F j ) Distribution of (2). If the distribution has correlation, dividing the feature subspace Fj into a common feature space; if the distribution has no correlation, the feature subspace Fj is divided into a non-common feature space.
The discrimination mode that the distribution has the correlation is as follows:
computing the same feature subspace F j The fitting degree of the characteristic distribution of each group of mixed samples and other groups of mixed samples;
obtaining the same characteristic subspace F according to the fitting degree j Each group of mixed sample features (x) i ,F j ) The overall degree of fit of;
and counting the number M of the mixed samples with the overall fitting degree larger than a first preset threshold, and judging that the distribution of the feature subspace has correlation when the M is larger than or equal to a second preset threshold.
Specifically, for F 1 The steps of performing correlation analysis on the feature distribution of the mixed sample are as follows: computing the same feature subspace F using a probability distribution distance metric algorithm j All the groups are mixedThe fitting degree of the characteristic distribution of the sample and other groups of mixed samples is obtained, and the subspace F with the same characteristic is obtained according to the fitting degree j Mixed sample feature of each group (x) i ,F j ) The number M of the mixed samples with the overall fitting degree larger than a first preset threshold value is counted, and when the M is larger than or equal to a second preset threshold value, the characteristic subspace is judged to have correlation for distribution. The fitting degree can be represented by KL divergence, the smaller the KL divergence is, the higher the fitting degree is, and the overall fitting degree is a set of KL divergence. The second predetermined threshold may be 80% of the total amount, and may also be other parameters, which are not limited in this embodiment, such as (x) i ,F 1 ) Calculating KL divergence of every two groups of mixed sample data to obtain
A KL divergence value, and
comparing the KL divergence values with a preset threshold, and if the overall fitting degree corresponding to the KL divergence values exceeding 80% is greater than a first preset threshold, judging that the characteristic subspace F is 1 The distributions have a correlation. If the distributions have correlation, the feature subspace F is identified 1 Dividing into common feature space, if distribution has no correlation, then dividing into feature subspace F 1 Dividing into non-common feature space, and dividing all feature subspaces F according to the method j Into a common feature space or a non-common feature space.
It can be understood that the source domain and target domain sample data have relevant features and irrelevant features in the same feature space, and the collaborative clustering method is used for realizing the simultaneous clustering of the sample data and the feature data in the same feature space, so that the relationship between the source domain and target domain sample data and the features can be intuitively embodied. The invention aims to distinguish relevant features and irrelevant features, and divide the relevant features and the irrelevant features into a common feature space and a non-common feature space respectively. The distribution of the normalized feature value extracted from the sample on the same feature subspace shows the characteristic of the sample on the feature subspace, and on the same common feature subspace, the source domain sample data and the target domain sample data should have distribution consistency, and the different groups of data should also have similar distribution, so that the feature subspace can be divided into a common feature space and a non-common feature space through the correlation analysis between the feature distributions of different groups of data in the same feature subspace.
With the technical support of the invention, the sample migration under the constraint of common knowledge can be realized without influencing the sample data distribution of the source domain, and the negative migration is effectively avoided.
And S3, inputting the commonality characteristics and the source domain non-commonality characteristics added with random noise into a generator to generate a pseudo sample.
And S4, inputting the pseudo sample and the target domain sample data into the discriminator, discriminating the pseudo sample according to the target domain sample data, and optimizing a generator according to a discrimination result and a target function.
And S5, repeating the training process from the step S3 to the step S4 until the objective function converges.
It should be noted that the generator used in the present invention is structurally symmetrical to the feature extraction model for feature extraction, so as to be able to directly input the extracted common features into the generator to generate samples with minimal loss, and the whole migration learning model and data transmission of an embodiment of the present invention are shown in fig. 2. The common characteristic is directly input to the generator to generate the sample, so that the constraint on the common characteristic can be effectively realized, and the generated sample is ensured to be fitted with the target domain sample as much as possible on the premise that the common characteristic is not changed.
In this embodiment, a decoder portion of a convolutional self-encoder is used as a generator, a decoder structure symmetrical to the encoder structure is adopted, the decoder adopts a three-layer deconvolution neural network structure, the number of convolution kernels in the first layer is 8, the size of the convolution kernels is 3 × 3, and the step length is 1; the number of the second layer of convolution kernels is 8, the size of the convolution kernels is 3 multiplied by 3, and the step length is 1; the number of convolution kernels in the third layer is 16, the size of the convolution kernels is 3 x 3, and the step size is 1. Each deconvolution layer is followed by a 2 x 2 upsampling layer to restore the image size.
In the embodiment, the same self-encoder is used, so that the reusability of the model is improved, and the cost for constructing the model is reduced. The generator and the discriminator form a non-traditional generation countermeasure network with the input as the characteristic, and the objective function of the generation countermeasure network is
Wherein G is the generator, D is the discriminator, E is the expectation function, x is the pseudo-sample data generated by the generator, p data Is the probability that x is from the true data distribution, p g Is the probability of x from the generator output sample. And normalizing the commonality characteristic and the source domain non-commonality characteristic added with the random noise, and inputting the normalized commonality characteristic and the source domain non-commonality characteristic into a decoder part of a convolution self-encoder to generate a pseudo sample. Inputting the pseudo sample and the target domain sample data into a discriminator, discriminating the pseudo sample according to the target domain sample data, and optimizing the whole generated confrontation network according to a discrimination result and a target function. And repeating the iteration of the training process until the objective function is converged.
And S6, inputting the source domain sample data into a trained generator to generate a migration sample.
It can be understood that the trained generator can generate sample data of a fitting target domain by receiving sample data of a source domain, so as to implement sample migration, a flow for implementing sample migration in this embodiment is shown in fig. 3, and a source domain sample can generate a migration sample of the fitting target domain after passing through a self-encoder.
The sample migration framework constructed by the invention supports full-automatic training and adjustment of the model, and realizes an end-to-end sample migration process. And automatically generating a sample fitting the target domain after the source domain sample is input.
It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising a … …" does not exclude the presence of another identical element in a process, method, article, or apparatus that comprises the element.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.
Claims (8)
1. A remote sensing sample migration method constrained by common knowledge is characterized by comprising the following steps:
s1, inputting source domain sample data and target domain sample data into a feature extraction model to obtain feature data and a public feature space of the source domain sample data and the target domain sample data;
s2, inputting the feature data of the source domain sample data and the target domain sample data into a feature clustering device, determining a common feature space and a non-common feature space, and extracting common features and source domain non-common features, wherein the common feature space and the non-common feature space are subspaces of the common feature space;
s3, inputting the common characteristic and the source domain non-common characteristic added with random noise into a generator to generate a pseudo sample;
s4, inputting the pseudo sample and target domain sample data into the discriminator, discriminating the pseudo sample according to the target domain sample data, and optimizing a generator according to a discrimination result and a target function;
s5, iterating the training processes from S3 to S4 until the target function converges;
and S6, inputting the source domain sample data into a trained generator to generate a migration sample.
2. The method for migrating remote sensing samples based on commonality knowledge constraint according to claim 1, wherein in step S2, the feature data of the source domain sample data and the target domain sample data is input into a feature clustering device, and a commonality feature space and a non-commonality feature space are determined, including:
clustering the source domain sample data and the target domain sample data on the public characteristic space to obtain k groups of mixed sample data x i I =1,.. K, wherein each set of mixed sample data includes a feature correlation therein;
mixing the mixed sample data x i A plurality of feature subspaces F mapped to the common feature space j In step (c), a sample-feature set (x) is obtained i ,F j )；
Analysis (x) i ,F j ) Distribution of (a):
if the distributions have correlation, the feature subspace F is divided into two parts j Dividing into a common characteristic space; if the distribution has no correlation, the feature subspace F is used j Partitioning into a non-common feature space.
3. The commonality knowledge constrained remote sensing sample migration method according to claim 2, wherein the discrimination manner that the distribution has correlation is:
computing the same feature subspace F j The fitting degree of the characteristic distribution of each group of mixed samples and other groups of mixed samples;
obtaining the same characteristic subspace F according to the fitting degree j Mixed sample feature of each group (x) i ,F j ) The overall degree of fit of;
and counting the number M of the mixed samples with the overall fitting degree larger than a first preset threshold, and judging that the distribution of the feature subspace has correlation when the M is larger than or equal to a second preset threshold.
4. The commonality knowledge-constrained remote sensing sample migration method of claim 3, wherein the degree of fit is calculated according to a probability distribution distance metric algorithm.
5. The commonality knowledge-constrained remote sensing sample migration method of claim 1, wherein the feature extraction model is:
adopting a machine learning model of a feature extraction operator;
or a model constructed by a convolutional neural network;
or a combined model of the machine learning model and a model constructed by a convolutional neural network.
6. The commonality knowledge-constrained remote sensing sample migration method of claim 5, wherein the feature extraction model is an encoder part of a convolutional auto-encoder constructed for a convolutional neural network;
accordingly, the generator is a decoder portion of the convolutional auto-encoder.
7. The commonality knowledge-constrained remote sensing sample migration method of claim 6, wherein the feature extraction model is symmetrical to the generator structure.
8. The commonality knowledge-constrained remote sensing sample migration method of claim 1, wherein the generator and the arbiter form a generative confrontation network, the objective function is an objective function of the generative confrontation network, the objective function is:
wherein G is the generator, D is the discriminator, E is the expectation function, x is the pseudo-sample data generated by the generator, p data Is the probability that x is from the true data distribution, p g Is the probability of x from the generator output sample.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211151014.4A CN115620038A (en) | 2022-09-21 | 2022-09-21 | Common knowledge constrained remote sensing sample migration method |
| PCT/CN2023/119287 WO2024061141A1 (en) | 2022-09-21 | 2023-09-18 | Method for remote-sensing sample transfer under common knowledge constraints |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211151014.4A CN115620038A (en) | 2022-09-21 | 2022-09-21 | Common knowledge constrained remote sensing sample migration method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115620038A true CN115620038A (en) | 2023-01-17 |
Family
ID=84857811
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211151014.4A Pending CN115620038A (en) | 2022-09-21 | 2022-09-21 | Common knowledge constrained remote sensing sample migration method |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN115620038A (en) |
| WO (1) | WO2024061141A1 (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2024061141A1 (en) * | 2022-09-21 | 2024-03-28 | 北京数慧时空信息技术有限公司 | Method for remote-sensing sample transfer under common knowledge constraints |
Family Cites Families (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109961096B (en) * | 2019-03-19 | 2021-01-05 | 大连理工大学 | Multimode hyperspectral image migration classification method |
| US20220004428A1 (en) * | 2020-07-02 | 2022-01-06 | International Business Machines Corporation | Artificial intelligence optimized cloud migration |
| CN113486827B (en) * | 2021-07-13 | 2023-12-08 | 上海中科辰新卫星技术有限公司 | Multi-source remote sensing image migration learning method based on domain countermeasure and self supervision |
| CN113936227A (en) * | 2021-12-17 | 2022-01-14 | 北京数慧时空信息技术有限公司 | Remote sensing image sample migration method |
| CN114529769B (en) * | 2022-02-21 | 2022-11-01 | 哈尔滨工业大学 | Separable multi-mode joint representation method for large-scene remote sensing image classification |
| CN115620038A (en) * | 2022-09-21 | 2023-01-17 | 北京数慧时空信息技术有限公司 | Common knowledge constrained remote sensing sample migration method |
-
2022
- 2022-09-21 CN CN202211151014.4A patent/CN115620038A/en active Pending
-
2023
- 2023-09-18 WO PCT/CN2023/119287 patent/WO2024061141A1/en unknown
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2024061141A1 (en) * | 2022-09-21 | 2024-03-28 | 北京数慧时空信息技术有限公司 | Method for remote-sensing sample transfer under common knowledge constraints |
Also Published As
| Publication number | Publication date |
|---|---|
| WO2024061141A1 (en) | 2024-03-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US9367602B2 (en) | Probabilistic cluster assignment | |
| CN111488917A (en) | Garbage image fine-grained classification method based on incremental learning | |
| Baghanam et al. | Conjunction of wavelet-entropy and SOM clustering for multi-GCM statistical downscaling | |
| CN116307212B (en) | Air quality prediction method and system | |
| CN115620038A (en) | Common knowledge constrained remote sensing sample migration method | |
| CN113378899A (en) | Abnormal account identification method, device, equipment and storage medium | |
| Halpin | An EM algorithm for Hawkes process | |
| CN116071352A (en) | Method for generating surface defect image of electric power safety tool | |
| CN114841296A (en) | Device clustering method, terminal device and storage medium | |
| CN117117859B (en) | Photovoltaic power generation power prediction method and system based on neural network | |
| CN103824285B (en) | Image segmentation method based on bat optimal fuzzy clustering | |
| CN115115090A (en) | Wind power short-term prediction method based on improved LSTM-CNN | |
| CN113094448B (en) | Analysis method and analysis device for residence empty state and electronic equipment | |
| CN115577797A (en) | Local noise perception-based federated learning optimization method and system | |
| CN115292361A (en) | Method and system for screening distributed energy abnormal data | |
| CN114511747A (en) | Unbalanced load data type identification method based on VAE preprocessing and RP-2DCNN | |
| CN116188834B (en) | Full-slice image classification method and device based on self-adaptive training model | |
| CN112926723A (en) | Automatic network growth method based on Split LBI algorithm | |
| CN110609832A (en) | Non-repeated sampling method for streaming data | |
| CN116383743A (en) | Photovoltaic operation abnormality detection method and system based on machine learning | |
| CN117688362A (en) | Photovoltaic power interval prediction method and device based on multivariate data feature enhancement | |
| Luo et al. | Learning Re-sampling Methods with Parameter Attribution for Image Super-resolution | |
| CN114726800A (en) | Rule type application flow classification method and system based on model interpretation | |
| CN116796622A (en) | Wind power output scene generation method based on conditional importance weighted self-encoder | |
| CN116169681A (en) | Power grid morphology analysis method, system, electronic equipment and medium |
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 |