CN116935121B - Dual-drive feature learning method for cross-region spectral image ground object classification - Google Patents
Dual-drive feature learning method for cross-region spectral image ground object classification Download PDFInfo
- Publication number
- CN116935121B CN116935121B CN202310899005.1A CN202310899005A CN116935121B CN 116935121 B CN116935121 B CN 116935121B CN 202310899005 A CN202310899005 A CN 202310899005A CN 116935121 B CN116935121 B CN 116935121B
- Authority
- CN
- China
- Prior art keywords
- domain
- target
- feature
- projection
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 30
- 230000003595 spectral effect Effects 0.000 title claims abstract description 22
- 230000006870 function Effects 0.000 claims abstract description 68
- 238000001228 spectrum Methods 0.000 claims abstract description 12
- 238000005457 optimization Methods 0.000 claims abstract description 6
- 230000008602 contraction Effects 0.000 claims description 4
- 239000011159 matrix material Substances 0.000 claims description 3
- 238000000513 principal component analysis Methods 0.000 claims description 3
- 230000009977 dual effect Effects 0.000 claims 1
- 238000006073 displacement reaction Methods 0.000 description 2
- 238000002372 labelling Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000002310 reflectometry Methods 0.000 description 1
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
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Multimedia (AREA)
- Databases & Information Systems (AREA)
- Medical Informatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a dual-drive feature learning method for cross-region spectral image feature classification, which belongs to the technical field of spectral image feature classification, and comprises the steps of constructing a domain feature alignment item based on statistical distribution drive and a domain feature alignment item based on model drive, wherein projection feature subspaces of common drive source domain and target domain data are mutually aligned, so that the hyperspectral cross-region alignment capability is optimized, and the model robustness is improved; and a discrimination information constraint item is constructed, so that the domain discrimination of the projection characteristic subspace is improved. Finally, solving an objective function by adopting an alternate direction multiplier algorithm through an alternate optimization strategy to respectively obtain two aligned subspaces of a source domain and a target domain; training a classifier by utilizing projection characteristics and category labels of the source domain dataset in subspaces of the source domain dataset; and sending the projection characteristics of the target domain data in the subspace thereof to a trained classifier so as to obtain the classification labels of the target domain data set. The implementation result on the disclosed cross-region spectrum dataset shows that compared with the existing method, the method has higher classification precision and more stable performance.
Description
Technical Field
The invention belongs to the technical field of spectrum image feature classification, and particularly relates to a cross-region spectrum image feature classification-oriented dual-drive feature learning method
Background
In recent years, a plurality of classification methods based on feature learning are used for classifying features of remote sensing images, especially in the field of classifying features of spectral images. However, most ground object classification methods require a large and accurate sample labeling, and the sample labeling process is often time consuming and requires a large amount of human resources. To solve this problem, a straightforward idea is to use a sample of labeled feature class labels in the source domain to classify similar target domains of newly acquired unlabeled feature class labels. Such tasks are referred to as cross-domain terrain classification tasks. However, in a practical task, since the source domain data and the target domain data are taken from different regions, the difference of the geographical conditions, the earth coverage type and the environmental factors of the two domains can cause the difference of the conditional distribution and the spectral reflectivity of the same ground object in the source domain and the target domain. In order to solve the challenge that the ideal classification effect is often not obtained by directly using the source domain to classify the target domain, namely the so-called cross-region spectrum image feature classification, a dual-drive feature learning method for the cross-region spectrum image feature classification is provided.
Disclosure of Invention
In order to solve the above challenges, the present invention provides a dual-drive feature learning method for cross-regional spectrum image feature classification, so as to construct a domain feature alignment item based on statistical distribution driving by using edge distribution constraint and conditional distribution constraint; in addition, constructing domain feature alignment items based on model driving, and searching unchanged subspaces by forcing mutual linear representation between target domain data and source domain data so as to realize further alignment of the source domain and the target domain; further, in order to improve the domain discriminant of the projection feature subspace, data reconstruction is introduced into the objective function as a discriminant information constraint term. And finally, solving an objective function by adopting an alternate direction multiplier algorithm through an alternate optimization strategy.
In an embodiment of the present invention, a dual-drive feature learning method for cross-region spectral image feature classification is provided, including: loading a source domain data set and a target domain data set, wherein the source domain data is a spectrum data sample with a ground object type label acquired at a certain region position, and the target domain data is a spectrum data sample to be classified without the ground object type label acquired at another different region position; defining an objective function of a dual-drive feature learning model related to cross-region spectral image feature classification, wherein the objective function comprises feature alignment items in two drive modes, namely a domain feature alignment item based on statistical distribution drive and a domain feature alignment item based on model drive, and the two alignment items jointly drive a projection feature subspace of source domain and target domain data to be mutually aligned; in order to improve the field discriminant of the projection feature subspace, introducing the data reconstruction into an objective function as a discriminant information constraint item, and re-planning the objective function; solving an objective function through an alternating optimization strategy by utilizing an alternating direction multiplier algorithm to respectively obtain two aligned subspaces of a source domain and a target domain; training a classifier by utilizing projection characteristics and category labels of the source domain dataset in subspaces of the source domain dataset; and sending the projection characteristics of the target domain data in the subspace thereof to a trained classifier so as to obtain the classification labels of the target domain data set.
Further, the defining step of the objective function of the dual-drive feature learning model for cross-regional spectrum image feature classification is as follows:
In order to constrain the association of the projection subspaces of the source domain and the target domain, a projection subspace constraint term is introduced, and the formulated objective function is as follows:
Wherein lambda 1 is a balance parameter, W s∈d×k and W t∈d×k are a source domain subspace and a target domain subspace respectively, d represents the dimensions of source domain data and target domain data, and k is a feature vector obtained by principal component analysis
Further, to construct domain feature alignment terms based on statistical distribution driving, edge distribution constraints are introduced as follows:
Where n s and n t are the number of source samples and target samples respectively, And/>Representing the ith spectral data sample of the source domain and the jth spectral data sample of the target domain, respectively.
Further, to construct domain feature alignment items based on statistical distribution driving, a conditional distribution constraint is introduced as follows:
where c is the number of sample classes, And/>The number of class i samples in the source domain and target domain data sets, respectively.
Combining the edge distribution constraint and the conditional distribution constraint and the projection subspace constraint to obtain an expression of a domain feature alignment term based on statistical distribution driving, wherein the expression is as follows:
further, an expression based on the domain feature alignment term driven by the model is constructed as follows:
Wherein lambda 3～λ6 is the balance parameter, Reconstruction coefficient matrix of source domain and target domain respectively,/>And/>Representing sets of spectral data for the source domain and the target domain, respectively.
The expression for combining the domain feature alignment item based on statistical distribution driving and the domain feature alignment item based on model driving to obtain the objective function is as follows:
further, a discriminant information constraint item is introduced into the objective function, and is redeveloped into the following form:
Wherein the method comprises the steps of Representing two orthogonal reconstruction matrices.
Further, the step of solving the values of the respective variables while minimizing the objective function value includes: iteratively solving a minimization problem for each variable under the condition that other variables are unchanged by using an alternate direction multiplier algorithm; fixing other variables, deleting function items irrelevant to P s, obtaining a target function formula of the variable P s, and solving through a singular value contraction operator; fixing other variables, deleting function items irrelevant to P t, obtaining a target function formula of the variable P t, and solving through a singular value contraction operator; fixing other variables, deleting function items irrelevant to Z s to obtain a target function formula of the variable Z s, and solving the derivative of the forced target function formula to be zero to obtain a closed form; fixing other variables, deleting function items irrelevant to Z t to obtain a target function formula of the variable Z t, and solving the derivative of the forced target function formula to be zero to obtain a closed form; fixing other variables, deleting function items irrelevant to W s to obtain a target function formula of the variable W s, and solving the derivative of the forced target function formula to be zero to obtain a closed form; and fixing other variables, deleting function items irrelevant to W t, obtaining a target function formula of the variable W t, and solving the derivative of the forced target function formula to be zero to obtain a closed form.
According to the dual-drive feature learning method for cross-region spectral image feature classification, which is provided by the embodiment of the invention, the domain feature alignment items based on statistical distribution drive and the domain feature alignment items based on model drive are introduced into a learning frame to jointly drive projection feature subspaces of source domain and target domain data to be mutually aligned. On the basis, in order to improve the domain discriminant of the projection feature subspace, a data reconstruction term is introduced into the objective function. Finally, solving an objective function through an alternating direction multiplier algorithm by an alternating optimization strategy; compared with the existing method, the method has higher classification precision and more stable performance.
The specific advantages are as follows:
1. The embodiment of the invention adopts a new double-driving feature learning model, and combines the domain feature alignment based on statistical distribution driving and the domain feature alignment based on model driving into a unified framework; in the new model, domain characteristic alignment items driven based on statistical distribution align two domains from two angles of edge distribution displacement and conditional distribution displacement, which is beneficial to improving the expression capacity of the model and reducing the risk of overfitting; the model-driven domain feature alignment item does not need label information of a sample in the learning process, and mutual representation learning is performed by utilizing the structure of data. This makes it of significant application value in cases where the processed data is unlabeled or otherwise marked in a limited manner.
2. According to the embodiment of the invention, the data reconstruction item is introduced into the framework, so that the domain discriminant of the projection characteristic subspace of the source domain and the target domain is effectively improved.
3. The embodiment of the invention adopts an alternate direction multiplier algorithm, solves an objective function through an alternate optimization strategy, and deduces the solution of each variable in the algorithm.
Drawings
In order to more clearly illustrate the exemplary embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the exemplary embodiments of the present invention or the prior art will be briefly described below.
The foregoing and other objects, features and advantages of exemplary embodiments of the invention will be better understood by reference to the following detailed description when read in conjunction with the accompanying drawings. Several embodiments of the present invention are illustrated by way of example, and not by way of limitation, in the figures of the accompanying drawings and in which:
FIG. 1 is a flow chart of a dual-drive feature learning method for cross-region spectral image feature classification provided by the invention;
Detailed Description
The principles and spirit of the present invention will be described below with reference to several exemplary embodiments. It should be understood that these embodiments are presented by way of example only and are not intended to limit the scope of the invention. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.
The invention provides a dual-drive feature learning method for cross-region spectral image ground feature classification, which comprises the following steps:
as shown in fig. 1, after the process flow starts, step S1 is first executed.
In step S1, a dataset is loaded, including a source domain dataset with a feature class label manually labeled and a target domain dataset to be classified without a feature class label.
Next, in step S2, domain feature alignment items driven based on the statistical distribution are defined.
As an example, the expression for the domain feature alignment term driven by statistical distribution defined in step S2 is as follows:
Where λ 1、λ2 is a balance parameter, W s∈d×k and W t∈d×k represent projection subspaces of the source domain and the target domain, respectively, d represents dimensions of the source domain data and the target domain data, and k is a feature vector obtained by principal component analysis. For edge distribution constraint, n s and n t are the number of source and target samples, respectively,/>And/>Representing the ith spectral data sample of the source domain and the jth spectral data sample of the target domain, respectively.Is a conditional distribution constraint, where c is the number of sample categories,/>And/>The number of class i samples in the source domain and target domain data sets, respectively. /(I)Associated constraint terms of subspaces are projected for the source and target domains. By means of the joint alignment constraint model, the projection feature subspaces of the source domain and the target domain data can be forced to have similar distribution characteristics from the perspective of statistical distribution driving, and the purpose of improving the classification accuracy of the target domain data in the classifier trained by the source domain data is achieved.
Then, in step S3, items are aligned based on the model-driven domain features.
As an example, the model-driven-based domain feature alignment term defined in step S3 takes the expression:
Wherein lambda 3～λ6 is the balance parameter, Reconstruction coefficient matrix of source domain and target domain respectively,/>And/>Representing sets of spectral data for the source domain and the target domain, respectively.
Then, in step S4, the domain feature alignment item driven based on the statistical distribution and the domain feature alignment item driven based on the model are combined and defined as an objective function.
As an example, the objective function defined in step S4 is expressed as follows:
Next, in step S5, a discriminant information constraint item is introduced into the objective function defined in step S4, and the objective function is re-formulated.
As an example, the objective function re-formulated in step S5 is represented as follows:
as an example, step S6 may include, for example, a sub-flow including steps S601 to S606, which will be described below.
In step S601, the minimization is solved iteratively for each variable using the alternate direction multiplier algorithm ADMM (Alternating Direction Method of Multipliers), the other variables (where "other variables" refer to all variables except P s) are fixed, and the function term unrelated to P s is deleted, resulting in the following formula:
equation (5) can be solved by Singular Value Decomposition (SVD) to The solution of P s is P s＝UVT.
Similarly, in step S602, other variables (herein, "other variables" refer to all variables except P t) are fixed by the alternate direction multiplier algorithm ADMM, and the function term unrelated to P t is deleted, resulting in the following formula:
SVD can also solve equation (6) so that The solution of P t is P t＝UVT.
In step S603, the alternate direction multiplier algorithm ADMM is also used to fix other variables (here, "other variables" means all variables except Z s), and delete function items unrelated to Z s as follows:
then, in step S603, let the result of deriving Z s by equation (7) be 0, the following result is obtained:
Similarly, in step S604, other variables (herein, "other variables" refer to all variables except Z t) are fixed by the alternate direction multiplier algorithm ADMM, and function terms unrelated to Z t are deleted, resulting in the following formula:
Then, in step S604, let the result of deriving Z t by equation (9) be 0, the following result is obtained:
In step S605, other variables (here, "other variables" means all variables except W s) are fixed by the alternate direction multiplier algorithm ADMM, and function terms unrelated to W s are deleted, resulting in the following formula:
Then, in step S605, let the result of deriving W s by equation (11) be 0, the following result is obtained:
In step S606, other variables (herein, "other variables" means all variables except W t) are fixed by the alternate direction multiplier algorithm ADMM, and function terms unrelated to W t are deleted, resulting in the following formula:
then, in step S606, let the result of deriving W t by equation (13) be 0, the following result is obtained:
by performing step S6, two aligned subspaces of the source domain and the target domain are obtained, respectively.
Finally, in step S7, training a classifier by utilizing the projection characteristics and class labels of the source domain data set in the subspace of the source domain data set; and sending the projection characteristics of the target domain data in the subspace thereof to a trained classifier so as to obtain the classification labels of the target domain data set.
Detailed description results
This embodiment uses HYRANK datasets. Details of the dataset are described below:
HYRANK dataset was developed within the framework of the international society of photogrammetry and the scientific initiative for RS (isps). The HYRANK dataset contains two different regions of scenes Dioni and Loukia, consisting of 250 x 1376 pixels and 249 x 945 pixels, respectively. The number and name of the experimental class and the number of samples of class samples in two different geographical scenarios are shown in table 1. To verify the superiority of this embodiment (Ours), comparing this embodiment with several existing methods, including KNN, TCA, etc., will compare the accuracy of these methods to the classification of the disclosed dataset, specific pairs of data are shown in table 2.
Table 1 number of source and target samples of HYRANK dataset
TABLE 2 HYRANK dataset Classification results
Continuous table 2
By comparing the data in the table, it can be clearly seen that the method achieves good performance and remarkably improves classification performance.
The embodiment provides a dual-drive feature learning method for cross-region spectral image ground feature classification. A dual-drive feature learning model based on a domain feature alignment item driven by statistical distribution and a domain feature alignment item driven by a model is established, and a numerical solution method based on an alternate direction multiplier method is designed for the model to solve an objective function. The results on the published test data set demonstrate the superiority of this embodiment.
It is to be understood that the above-described embodiments of the present invention are merely illustrative of or explanation of the principles of the present invention and are in no way limiting of the invention. Accordingly, the invention is intended to cover various modifications and equivalent arrangements included within the spirit and scope of the appended claims.
Claims (2)
1. The double-drive feature learning method for cross-region spectral image ground object classification is characterized by comprising the following steps:
Dividing a spectrum image data set into a source domain data set and a target domain data set, wherein the source domain data is a spectrum data sample with a ground object type label, which is acquired at a certain region position, and the target domain data is a spectrum data sample to be classified, which is acquired at another different region position and has no ground object type label; defining an objective function of a projection feature subspace learning model, wherein the objective function comprises two feature pairs in a driving form, namely a domain feature alignment item based on statistical distribution driving and a domain feature alignment item based on model driving, and the two alignment items jointly drive the projection feature subspaces of source domain and target domain data to be aligned with each other;
the expression of the domain feature alignment term based on statistical distribution driving in the objective function is as follows:
Wherein lambda 1、λ2 is the balance parameter, And/>Respectively representing a projection subspace of a source domain and a projection subspace of a target domain, d represents dimensions of source domain data and target domain data, and k is a feature vector obtained through principal component analysis; for the edge distribution constraint, n s and n t are the number of source and target samples, respectively,/> And/>Representing the ith spectral data sample of the source domain and the jth spectral data sample of the target domain respectively,Is a conditional distribution constraint, where c is the number of sample categories,/>And/>The number of class I samples in the source domain and target domain data sets respectively; /(I)The method is characterized in that the projection feature subspaces of the source domain and the target domain are forced to have similar distribution characteristics from the perspective of statistical distribution driving through a joint alignment constraint model for the associated constraint items of the projection subspaces of the source domain and the target domain, so that the aim of improving the classification accuracy of the target domain data in a classifier trained by the source domain data is fulfilled;
The expression of the domain feature alignment term based on model driving in the objective function is as follows:
Wherein lambda 3～λ6 is the balance parameter, Reconstruction coefficient matrix of source domain and target domain respectively,/>And/>Spectral data sets respectively representing a source domain and a target domain, and through the domain feature alignment item driven by the model, the features of the source domain data and the target domain data in a projection subspace can be interactively linearly reconstructed, and the data projection features are further forced to have similar distribution characteristics from the angle of linear representation model driving;
In order to improve the domain discriminant of the projection feature subspace, the data reconstruction is introduced into an objective function as a discriminant information constraint term, and the objective function is re-formulated into the following form:
Wherein therein is Representing two orthogonal reconstruction matrixes, wherein the discriminant constraint item can effectively keep discriminant information of the respective fields of source domain and target domain data after projection;
Finally, solving a redevelopment objective function through an alternate optimization strategy by utilizing an alternate direction multiplier algorithm to respectively obtain two alignment subspaces of a source domain and a target domain; training a classifier by utilizing projection characteristics and category labels of the source domain dataset in subspaces of the source domain dataset; and sending the projection characteristics of the target domain data in the subspace thereof to a trained classifier so as to obtain the classification labels of the target domain data set.
2. The dual drive feature learning method of claim 1, wherein the solving the redefined objective function using an alternate direction multiplier algorithm solves the minimization problem iteratively for each variable by:
Fixing other variables, deleting function items irrelevant to P s, obtaining a target function formula of the variable P s, and solving through a singular value contraction operator;
Fixing other variables, deleting function items irrelevant to P t, obtaining a target function formula of the variable P t, and solving through a singular value contraction operator;
Fixing other variables, deleting function items irrelevant to Z s to obtain a target function formula of the variable Z s, and solving the derivative of the forced target function formula to be zero to obtain a closed form;
Fixing other variables, deleting function items irrelevant to Z t to obtain a target function formula of the variable Z t, and solving the derivative of the forced target function formula to be zero to obtain a closed form;
Fixing other variables, deleting function items irrelevant to W s to obtain a target function formula of the variable W s, and solving the derivative of the forced target function formula to be zero to obtain a closed form;
and fixing other variables, deleting function items irrelevant to W t, obtaining a target function formula of the variable W t, solving the derivative of the forced target function formula to be zero, obtaining a closed form, and solving by updating the variable item by item iteration.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310899005.1A CN116935121B (en) | 2023-07-20 | 2023-07-20 | Dual-drive feature learning method for cross-region spectral image ground object classification |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310899005.1A CN116935121B (en) | 2023-07-20 | 2023-07-20 | Dual-drive feature learning method for cross-region spectral image ground object classification |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116935121A CN116935121A (en) | 2023-10-24 |
| CN116935121B true CN116935121B (en) | 2024-04-19 |
Family
ID=88383946
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310899005.1A Active CN116935121B (en) | 2023-07-20 | 2023-07-20 | Dual-drive feature learning method for cross-region spectral image ground object classification |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116935121B (en) |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109389174A (en) * | 2018-10-23 | 2019-02-26 | 四川大学 | A kind of crowd massing Sensitive Image Detection Method |
| CN110619367A (en) * | 2019-09-20 | 2019-12-27 | 哈尔滨理工大学 | Joint low-rank constraint cross-view-angle discrimination subspace learning method and device |
| CN112052888A (en) * | 2020-08-26 | 2020-12-08 | 西安电子科技大学 | Domain adaptive mode identification method based on coupled projection and embedded subspace |
| CN112836736A (en) * | 2021-01-28 | 2021-05-25 | 哈尔滨理工大学 | Hyperspectral image semi-supervised classification method based on depth self-encoder composition |
| CN113240030A (en) * | 2021-05-24 | 2021-08-10 | 哈尔滨理工大学 | Domain self-adaptive subspace learning method based on interactive representation |
| CN113627084A (en) * | 2021-08-06 | 2021-11-09 | 西南大学 | Electronic nose signal drift compensation subspace alignment method based on extreme learning machine |
| CN115034369A (en) * | 2022-06-14 | 2022-09-09 | 天津国能盘山发电有限责任公司 | Fault diagnosis method and device, storage medium and electronic equipment |
| US11487273B1 (en) * | 2021-04-30 | 2022-11-01 | Dalian University Of Technology | Distributed industrial energy operation optimization platform automatically constructing intelligent models and algorithms |
| CN116451179A (en) * | 2023-04-18 | 2023-07-18 | 常州工业职业技术学院 | Domain adaptive dictionary pair learning method based on projection reconstruction |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9684951B2 (en) * | 2014-03-31 | 2017-06-20 | Los Alamos National Security, Llc | Efficient convolutional sparse coding |
| EP3618287B1 (en) * | 2018-08-29 | 2023-09-27 | Université de Genève | Signal sampling with joint training of learnable priors for sampling operator and decoder |
-
2023
- 2023-07-20 CN CN202310899005.1A patent/CN116935121B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109389174A (en) * | 2018-10-23 | 2019-02-26 | 四川大学 | A kind of crowd massing Sensitive Image Detection Method |
| CN110619367A (en) * | 2019-09-20 | 2019-12-27 | 哈尔滨理工大学 | Joint low-rank constraint cross-view-angle discrimination subspace learning method and device |
| CN112052888A (en) * | 2020-08-26 | 2020-12-08 | 西安电子科技大学 | Domain adaptive mode identification method based on coupled projection and embedded subspace |
| CN112836736A (en) * | 2021-01-28 | 2021-05-25 | 哈尔滨理工大学 | Hyperspectral image semi-supervised classification method based on depth self-encoder composition |
| US11487273B1 (en) * | 2021-04-30 | 2022-11-01 | Dalian University Of Technology | Distributed industrial energy operation optimization platform automatically constructing intelligent models and algorithms |
| CN113240030A (en) * | 2021-05-24 | 2021-08-10 | 哈尔滨理工大学 | Domain self-adaptive subspace learning method based on interactive representation |
| CN113627084A (en) * | 2021-08-06 | 2021-11-09 | 西南大学 | Electronic nose signal drift compensation subspace alignment method based on extreme learning machine |
| CN115034369A (en) * | 2022-06-14 | 2022-09-09 | 天津国能盘山发电有限责任公司 | Fault diagnosis method and device, storage medium and electronic equipment |
| CN116451179A (en) * | 2023-04-18 | 2023-07-18 | 常州工业职业技术学院 | Domain adaptive dictionary pair learning method based on projection reconstruction |
Non-Patent Citations (2)
| Title |
|---|
| Cross-Scene Hyperspectral Image Classification With Discriminative Cooperative Alignment;Yuxiang Zhang 等;《IEEE Transactions on Geoscience and Remote Sensing》;20210113;第59卷(第11期);9646-9600 * |
| 多核低冗余表示学习的稳健多视图子空间聚类方法;李骜 等;《通信学报》;20211130;第42卷(第11期);193-204 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116935121A (en) | 2023-10-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Amid et al. | TriMap: Large-scale dimensionality reduction using triplets | |
| US11875267B2 (en) | Systems and methods for unifying statistical models for different data modalities | |
| Greenberg et al. | Automatic posterior transformation for likelihood-free inference | |
| Noroozi et al. | Representation learning by learning to count | |
| Gui et al. | Joint learning of visual and spatial features for edit propagation from a single image | |
| Du et al. | Spatial and spectral unmixing using the beta compositional model | |
| CN106777318B (en) | Matrix decomposition cross-modal Hash retrieval method based on collaborative training | |
| US7512273B2 (en) | Digital ink labeling | |
| US9070047B2 (en) | Decision tree fields to map dataset content to a set of parameters | |
| US9449395B2 (en) | Methods and systems for image matting and foreground estimation based on hierarchical graphs | |
| CN112232374B (en) | Irrelevant label filtering method based on depth feature clustering and semantic measurement | |
| Díaz et al. | Micro‐structural tissue analysis for automatic histopathological image annotation | |
| CN117992805B (en) | Zero sample cross-modal retrieval method and system based on tensor product graph fusion diffusion | |
| Simran et al. | Content based image retrieval using deep learning convolutional neural network | |
| Sui et al. | Robust tracking via locally structured representation | |
| CN115019103A (en) | Small sample target detection method based on coordinate attention group optimization | |
| Siddiqa et al. | Spectral segmentation based dimension reduction for hyperspectral image classification | |
| CN115457311A (en) | Hyperspectral remote sensing image band selection method based on self-expression transfer learning | |
| Ma et al. | Multifeature-based discriminative label consistent K-SVD for hyperspectral image classification | |
| CN111259176B (en) | Cross-modal Hash retrieval method based on matrix decomposition and integrated with supervision information | |
| CN116935121B (en) | Dual-drive feature learning method for cross-region spectral image ground object classification | |
| Oh et al. | Acceleration of simple linear iterative clustering using early candidate cluster exclusion | |
| Biswas et al. | Attendance Tracking with Face Recognition Through Hidden Markov Models | |
| Wang et al. | Conscience online learning: an efficient approach for robust kernel-based clustering | |
| CN114299342B (en) | Unknown mark classification method in multi-mark picture classification based on deep learning |
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 |