CN113657541B - Domain self-adaptive target recognition method based on depth knowledge integration - Google Patents

Domain self-adaptive target recognition method based on depth knowledge integration Download PDF

Info

Publication number
CN113657541B
CN113657541B CN202110987414.8A CN202110987414A CN113657541B CN 113657541 B CN113657541 B CN 113657541B CN 202110987414 A CN202110987414 A CN 202110987414A CN 113657541 B CN113657541 B CN 113657541B
Authority
CN
China
Prior art keywords
feature
mapping matrix
features
target
matrix
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
Application number
CN202110987414.8A
Other languages
Chinese (zh)
Other versions
CN113657541A (en
Inventor
郭贤生
张玉坤
段林甫
陆浩然
袁杨鹏
黄健
李林
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Yangtze River Delta Research Institute of UESTC Huzhou
Original Assignee
Yangtze River Delta Research Institute of UESTC Huzhou
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Yangtze River Delta Research Institute of UESTC Huzhou filed Critical Yangtze River Delta Research Institute of UESTC Huzhou
Priority to CN202110987414.8A priority Critical patent/CN113657541B/en
Publication of CN113657541A publication Critical patent/CN113657541A/en
Application granted granted Critical
Publication of CN113657541B publication Critical patent/CN113657541B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • 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
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
    • Y02T10/00—Road transport of goods or passengers
    • Y02T10/10—Internal combustion engine [ICE] based vehicles
    • Y02T10/40—Engine management systems

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Molecular Biology (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Health & Medical Sciences (AREA)
  • Image Analysis (AREA)

Abstract

The invention belongs to the technical field of target recognition, and particularly relates to a field-adaptive target recognition method based on depth knowledge integration. The invention realizes the deep knowledge integration of the feature level and the decision level. The common mapping matrix and the specific mapping matrix are designed at the feature level to realize knowledge integration, so that the robustness of the target recognition performance is improved; the public mapping matrix fully digs public knowledge of heterogeneous features, and the special mapping matrix reserves special knowledge of different features. And the importance degree of different features is quantified by designing the feature weights at the decision level, and meanwhile, the feature weights are updated by utilizing the target domain samples through online learning, so that the data distribution difference of different fields is overcome, and the field self-adaptive target identification is realized. Therefore, the field self-adaptive target recognition method based on depth knowledge integration is an intelligent field self-adaptive target recognition method.

Description

Domain self-adaptive target recognition method based on depth knowledge integration
Technical Field
The invention belongs to the technical field of target recognition, and particularly relates to a field-adaptive target recognition method based on depth knowledge integration.
Background
The automatic target recognition technology can recognize and classify targets according to sensor data, and plays an important role in military and civil fields such as battlefield sensing reconnaissance, terrain exploration, automatic driving and the like. With the continuous development of technology, various advantageous target recognition methods are sequentially proposed. How to combine the advantages of different methods to improve the performance of object recognition has become one of the hot spots of object recognition technology research.
The literature Q.Yu, H.Hu, X.Geng, Y.Jiang and J.an, "High-Performance SAR Automatic Target Recognition Under Limited Data Condition Based on a Deep Feature Fusion Network," in IEEE Access, vol.7, pp.165646-165658,2019 "proposes a feature level knowledge integration method, which splices features extracted from different convolutional layers of a neural network together as features to be input into a final classifier to integrate knowledge of features of different scales. However, the method does not explore the relationship between different features, ignores the common knowledge and the specific knowledge between different features. The literature "J.Zhang, M.Xing and Y.Xie," FEC: A Feature Fusion Framework for SAR Target Recognition Based on Electromagnetic Scattering Features and Deep CNN Features, "in IEEE Transactions on Geoscience and Remote Sensing, vol.59, no.3, pp.2174-2187, march 2021" proposes a decision level knowledge integration method, concatenates the results of different method decision levels, and trains a new classifier to combine the knowledge of the different methods, but the importance of the method in knowledge integration ignoring the different features is different. Meanwhile, due to environmental changes or changes of the observation target angles, distribution differences exist between the source domain and the target domain data, but most of current knowledge integration methods do not consider the problem of domain self-adaption, the model learned in the source domain cannot always cope with the changes generated in the target domain, and the performance of the model in the target domain is reduced. Therefore, research on the field-adaptive target recognition method based on depth knowledge integration is expected to further improve the target recognition performance.
Disclosure of Invention
The invention aims to overcome the defects and provide a field self-adaptive target recognition method based on depth knowledge integration. The invention realizes the field self-adaptive target recognition through the deep knowledge integration of the feature level and the decision level and the online learning. In the feature level, the invention designs the public mapping matrix and the special mapping matrix to realize knowledge integration, thereby improving the robustness of the target recognition performance; the public mapping matrix fully digs public knowledge of heterogeneous features, and the special mapping matrix reserves special knowledge of different features. In the decision stage, the invention designs the feature weights to quantify the importance degrees of different features, and simultaneously utilizes the target domain samples to update the feature weights through online learning, thereby overcoming the data distribution difference of different domains and realizing the domain self-adaptive target identification. Therefore, the field self-adaptive target recognition method based on depth knowledge integration is an intelligent field self-adaptive target recognition method.
The technical scheme of the invention is as follows: a field self-adaptive target recognition method based on depth knowledge integration, as shown in figure 1, comprises the following steps:
s1, respectively acquiring original image samples in a source domain and a target domain and preprocessing the original image samples;
s2, training a plurality of feature extractors by using a source domain sample, and extracting a plurality of types of features;
s3, establishing an objective function of the feature level knowledge integration model by utilizing the different types of features obtained in the step S2, and solving, wherein the method specifically comprises the following steps:
s31, designing a transformation matrix { Θ } i Public mapping matrix A 0 Unique mapping matrix { A i Mapping different types of features to a unified tag space, and measuring the difference between the features and a real tag to obtain an objective function of feature level knowledge integration;
s32, optimizing and solving the objective function obtained in the step S31 by utilizing a three-step iteration method to obtain a transformation matrix { Θ } i Public mapping matrix A 0 Unique mapping matrix { A i }；
S4, establishing a decision-level knowledge integration online model by using a target domain sample and solving, wherein the method specifically comprises the following steps:
s41, utilizing the multiple feature extractors obtained in the step S2 and the transformation matrix { Θ } obtained in the step S3 i Public mapping matrix A 0 Unique mapping matrix { A i Processing the target domain mark sample, designing feature weights, and establishing a decision level knowledge integrated target function;
s42, solving the objective function obtained in the step S41 to obtain a characteristic weight;
s5, utilizing the multiple feature extractors obtained in the step S2, and obtaining a transformation matrix { Θ ] in the step S3 i Public mapping matrix A 0 Unique mapping matrix { A i And (3) processing the sample to be detected and the characteristic weight obtained in the step S4 to obtain a recognition result.
The beneficial effects of the invention are as follows: the method fully excavates public knowledge and special knowledge of heterogeneous features at a feature level, quantifies importance degrees of different features at a decision level, updates feature weights by utilizing target domain data based on online learning, realizes deep knowledge integration of the feature level and the decision level, overcomes data distribution differences in different fields, improves robustness of a target recognition technology, and realizes field-adaptive target recognition. Therefore, the field self-adaptive target recognition method based on depth knowledge integration is an intelligent field self-adaptive target recognition method.
Drawings
FIG. 1 is a schematic diagram of a frame of the present invention;
FIG. 2 is a flowchart of an algorithm of the present invention;
FIG. 3 is a graph comparing recognition accuracy of the background method and the method of the present invention.
Detailed Description
The following describes the technical scheme of the present invention in detail with reference to the drawings and the embodiments:
as shown in fig. 2, the present invention includes:
and step 1, respectively acquiring original image samples in a source domain and a target domain and preprocessing the original image samples.
And acquiring original images of each target at different pitching angles in static state by using a radar, and observing the target at different azimuth angles at each fixed pitching angle. And marking the acquired images as a source domain and a target domain according to the difference of the pitch angles.
In the source domainA large number of marked samples are collected and marked as Z = { Z after cutting pretreatment j j=1, 2, …, n }, where n is the total number of samples of the source domain training set. Labeling all source domain samples as matrix +.>Where L is the total number of categories of samples. Matrix->The list of each of the corresponding samples indicates one-hot tags [0, …,0,1,0, …,0] T The first position being 1 and the remaining elements being 0 represent the tag number being L, L e {1,2, …, L }.
In the form of online in the target domainA small number of labelled samples are collected, i.e. one sample at a time, in succession. It was cut and pre-treated and then marked +.>The corresponding one-hot tag is noted as wherein />The total number of samples is marked for the target domain.
Step 2, training N kinds of feature extractors f by utilizing the source domain samples obtained in the step 1 i (. Cndot.) i.e {1,2, …, N }, the different types of source domain samples are characterized by
wherein ,the representation utilizes a feature extractor f i (. Cndot.) features extracted from Source Domain sample Z, d i Representing the dimension of the class i feature.
Step 3, establishing an objective function of the feature level knowledge integration model according to the features of different types obtained in the step 2, and solving, wherein the specific steps of the step 3 are as follows:
step 3-1. Design a transformation matrix { Θ } i Public mapping matrix A 0 Unique mapping matrix { A i Mapping different types of features to a unified tag space, measuring the difference between the features and a real tag, establishing a feature level knowledge integration model, and recording a total objective function as follows:
wherein ,representing a loss of classification,/->Representing regularization terms. />Representing the mapping of features of different dimensions to specified dimensions,/->M is a designated dimension; />Further mapping different types of features to a unified tag space, wherein +.>And measuring the difference between the features and the real labels in the label space by using the Frobenius norm to obtain a classification loss term. For the transformation matrix { Θ i Public mapping matrix A 0 Unique mapping matrix { A i Frobenius norm constraint is carried out on the distribution to obtain a regularization term, wherein alpha, beta and gamma respectively represent regularization coefficients and are used for controlling the degree of regularization of parameters. A can be found by minimizing the objective function 0 ,{A i },{Θ i }：
And 3-2, carrying out optimization solution on the formula (3) obtained in the step 3-1 by using a three-step iteration method, wherein the specific steps are as follows:
step 3-2-1: fixed transformation matrix { Θ i Sum of specific mapping matrix { A } i For a public mapping matrix A 0 Optimizing, and converting an objective function into:
this is an unconstrained optimization problem, and can be solved as a closed-form solution:
wherein I is an M-order unit array.
Step 3-2-2: fixed transformation matrix { Θ i Sum-common mapping matrix a 0 For the peculiar mapping matrix { A i Optimizing, and converting an objective function into:
splitting equation (6) into N independent unconstrained least squares problems, each of which can be expressed as:
solving to obtain optimized specific mapping matrix { A } i }：
Step 3-2-3: finally, the public mapping matrix A is fixed 0 And a unique mapping matrix { A } i For the transformation matrix { Θ } i Optimizing, and the objective function is:
the same as the solving process of the step 3-2-2, the optimized transformation matrix { Θ } is obtained i }：
wherein ,I1 Is d i Order unit array, I 2 Is an M-order unit array.
Iterating steps 3-2-1, 3-2-2 and 3-2-3 until convergence, obtaining a transformation matrix { Θ } i Public mapping matrix A 0 Unique mapping matrix { A i When the degree of change of the objective function formula (2) is smaller than a given threshold δ, the algorithm is considered to converge:
where t is the number of iterations.
Step 4. The transformation matrix { Θ ] obtained according to step 3 i Public mapping matrix A 0 Unique mapping matrix { A i Establishing a decision-level knowledge integration online model by using a target domain sample and solving, wherein the specific steps of the step 4 are as follows:
step 4-1. Marking the target Domain sample obtained in step 1 with the multiple feature extractors obtained in step 2Extracting multiple types of features->Where i.epsilon. {1,2, …, N },>transformation matrix { Θ according to step 3 i Public mapping matrix A 0 Unique mapping matrix { A i The corresponding transformation characteristics after the characteristics of different types of the calculated target domain samples are transformed into the label space ∈>
In order to explore the importance degree of different features in a target domain, a decision-level knowledge integration online model objective function is established:
wherein ,ξi Representing class i transform featuresImportance of (I)>The distance between the transformed feature and the real tag is described; log (xi) i ) Is a constraint term to avoid the obtained xi i Very close to 0; lambda is the balance parameter.
And 4-2, carrying out optimization solution on the decision-making level knowledge integration online model established in the step 4-1, wherein the specific steps are as follows:
step 4-2-1. The objective function is transformed to minimize the logarithmic function problem.
Recording deviceGiven feature weight ζ i Is provided with->Obeys normal distribution->There is a likelihood function:
maximizing likelihood function equation (14) is equivalent to minimizing its negative logarithmic function:
it can be seen that minimizing equation (15) is equivalent to solving equation (13).
And 4-2-2, further converting the objective function into a maximum posterior probability estimation problem and solving the characteristic weights.
Setting xi i Obeys the Γ distribution ζ i ～Γ(γ 1 ,γ 2), wherein ,γ1 and γ2 Is a parameter. There is a probability density function:
calculation of xi i The posterior probability distribution of (2) is:
thus, xi i Is subject to Γ distributionGiven->And marking samples by the target domains, and obtaining feature weights by using maximum posterior probability estimation:
and 4-2-3, updating the feature weight obtained by solving in an online mode.
Order theThe representation is based on +.>Cumulative count when feature weights are updated for each sample, +.>And->Representing the corresponding accumulated error, +.>And-> wherein ,/> and />Respectively corresponding initial values. Thus, the feature weights can be updated online:
step 5. Utilizing the feature extractor f obtained in step 2 i (. About.) different types of features are extracted from the target domain sample to be measured and recorded asi.epsilon. {1,2, …, N }. The transformation matrix { Θ ] obtained according to step 3 i Public mapping matrix A 0 Unique mapping matrix { A i Step 4 feature weight ζ after online update i Will->Depth knowledge integration through feature level and decision level maps to tag space:
taking the index corresponding to the maximum value in c to obtain the predicted tag number of the target to be detectedObtaining a target identification result:
examples
The model is used for carrying out experiments on ten targets in the MSTAR data set acquired and identified by moving and static targets in the United states, a sensor for acquiring the data set is a high-resolution bunched synthetic aperture radar, the radar works in an X-band, a HH polarization mode is adopted, and the resolution is 0.3m multiplied by 0.3m. The acquired data is subjected to pre-processing, and slice images with the pixel size of 128×128 and containing various targets are extracted from the acquired data. The data are mostly SAR slice images of stationary vehicles, comprising ten classes of targets BMP2, T72, BTR70, 2S1, BRDM2, BTR60, D7, T62, ZIL131, ZSU234, and T72. Sample data observed at a pitch angle of 17 degrees is taken as a source domain sample, sample data observed at a pitch angle of 15 degrees is taken as a target domain sample, the target domain sample is recorded as standard operation conditions, 10 types of targets are contained in total, and the specific sample numbers are shown in table 1. Sample data observed at a pitch angle of 17 degrees is taken as a source domain sample, sample data observed at a pitch angle of 30 degrees is taken as a target domain sample, the target domain sample is recorded as an extended operation condition, 3 types of targets are contained in total, and the specific sample numbers are shown in table 2.
Table 1 number of specific samples under standard operating conditions
Table 2 specific sample numbers under extended operating conditions
To remove the effect of background clutter, the sample image size is cut to 64×64 centered. In the case, a convolutional neural network and a sparse representation method are adopted as two feature extractors, so that N=2, and the extracted feature dimensions are d respectively 1 ＝128，d 2 =256, the remaining parameters are set to m=25, α=10, respectively 3 ，β＝γ＝10 2 ，λ＝10，γ 1 ＝10，γ 2 =5, the convergence threshold is set to δ=10 -3 . All the source domain samples are used as source domain marking samples, 10 samples are randomly extracted from the target domain samples to serve as target domain marking samples to participate in training, and the rest samples are used as samples to be tested.
Experiments under different operation conditions are designed to verify the superiority of the proposed algorithm, and the performances of the background technology method and the method of the invention under standard operation conditions and extended operation conditions are respectively compared. As shown in fig. 3, under the standard operation condition, the target domain sample is observed at a pitch angle of 15 degrees, the source domain is observed at a pitch angle of 17 degrees, the data distribution difference of the target domain sample and the source domain sample is not large, the method and the background art method have good recognition performance, and the recognition rate is over 95 percent, wherein the recognition rate of the method is the highest and reaches 98.6 percent. Under the condition of expansion operation, the target domain sample is observed at a pitch angle of 30 degrees, the difference between the target domain sample and the source domain sample is larger, the identification performance of the background technology method is greatly reduced, the method still can keep the identification accuracy of 98.4 percent, and the advantage is obvious. In conclusion, experimental results prove that the method effectively combines the advantages of different types of characteristics, realizes field self-adaption, improves the accuracy of target identification, and enhances the generalization capability of the model.

Claims (1)

1. The field self-adaptive target recognition method based on depth knowledge integration is characterized by comprising the following steps of:
s1, respectively acquiring original image samples in a source domain and a target domain and preprocessing the original image samples; the specific method comprises the following steps:
acquiring original images of each target at different pitching angles in static state by a radar, observing the target at different azimuth angles at each fixed pitching angle, and marking the acquired images as a source domain and a target domain according to the difference of the pitching angles;
in the source domainThe marked sample is obtained after cutting pretreatment and is marked as Z= { Z j I j = 1,2, …, n }, where n is the total number of samples of the source domain training set; labeling all source domain samples as matrix +.>Wherein L is the total category number of the sample, matrix +.>The list of each of the corresponding samples indicates one-hot tags [0, …,0,1,0, …,0] T The first position being 1 and the remaining elements being 0 represent the tag number being L, L ε {1,2, …, L };
in the form of online in the target domainCollecting labeled samples, i.e. sequentially, one sample at a time, performing cutting pretreatment, and marking as +.>The corresponding one-hot tag is designated +.> wherein />Marking the total number of samples for the target domain;
s2, training a plurality of feature extractors by using a source domain sample, and extracting a plurality of types of features; the specific method comprises the following steps:
training N feature extractors f using source domain samples obtained in step S1 i (. Cndot.) i.e {1,2, …, N }, the different types of source domain samples are characterized by
wherein ,the representation utilizes a feature extractor f i (. Cndot.) features extracted from Source Domain sample Z, d i Dimension representing class i features;
s3, establishing an objective function of the feature level knowledge integration model by utilizing the features of different types obtained in the step S2, and solving the objective function, wherein the specific method comprises the following steps of:
s31, designing a transformation matrix { Θ } i Public mapping matrix A 0 Unique mapping matrix { A i Mapping different types of features to a unified tag space, measuring the difference between the features and a real tag, and establishing a feature level knowledge integration model to obtain an objective function of feature level knowledge integration, wherein the objective function is as follows:
wherein ,representing a loss of classification,/->Representing regularized item, ++>Representing the mapping of features of different dimensions to specified dimensions,/->M is a designated dimension; />Mapping different types of features to a unified tag space, wherein +.>The difference between the features in the tag space and the real tags is measured by using the Frobenius norm, and a classification loss term is obtained; for the transformation matrix { Θ i Public mapping matrix A 0 Unique mapping matrix { A i Frobenius norm constraint on the distribution is performed to obtain regularization term, wherein alpha, beta and gamma each represent regularization coefficient for controlling degree of regularization of parameters by minimizing objective functionThe number can be used to obtain A 0 ,{A i },{Θ i }：
Step S32, carrying out optimization solution on the formula (3) obtained in the step S31 by utilizing a three-step iteration method, wherein the specific steps are as follows:
s321, fixed transformation matrix { Θ ] i Sum of specific mapping matrix { A } i For a public mapping matrix A 0 Optimizing, and converting an objective function into:
this is an unconstrained optimization problem, solving a closed solution as:
wherein I is an M-order unit array;
step S322: fixed transformation matrix { Θ i Sum-common mapping matrix a 0 For the peculiar mapping matrix { A i Optimizing, and converting an objective function into:
splitting equation (6) into N independent unconstrained least squares problems, each of which can be expressed as:
solving to obtain optimized specific mapping matrix { A } i }：
Step S323: finally, the public mapping matrix A is fixed 0 And a unique mapping matrix { A } i For the transformation matrix { Θ } i Optimizing, and the objective function is:
the same as the solving process of step S322, the optimized transformation matrix { Θ is obtained i }：
wherein ,I1 Is d i Order unit array, I 2 Is an M-order unit array;
iterating steps S321, S322 and S323 until convergence to obtain a transformation matrix { Θ } i Public mapping matrix A 0 Unique mapping matrix { A i When the degree of change of the objective function formula (2) is smaller than a given threshold δ, the algorithm is considered to converge:
wherein t is the number of iterations;
s4, establishing a decision-level knowledge integration online model by using a target domain sample and solving, wherein the method specifically comprises the following steps:
step S41, marking the target domain obtained in step S1 with the plurality of feature extractors obtained in step S2Extracting multiple types of features->Where i.epsilon. {1,2, …, N },>transformation matrix { Θ according to step S3 i Public mapping matrix A 0 Unique mapping matrix { A i The corresponding transformation characteristics after the characteristics of different types of the calculated target domain samples are transformed into the label space ∈>
In order to obtain the importance degree of different features in a target domain, establishing a decision-level knowledge integration online model target function:
wherein ,ξi Representing class i transform featuresImportance of (I)>The distance between the transformed feature and the real tag is described; log (xi) i ) Is a constraint term to avoid the obtained xi i Very close to 0; lambda is the equilibrium parameter;
step S42, carrying out optimization solution on the decision-making level knowledge integration online model established in the step S41, wherein the specific steps are as follows:
step S421, converting the objective function into a minimized logarithmic function problem:
recording deviceGiven feature weight ζ i Is provided with->Obeys normal distribution->There is a likelihood function:
maximizing likelihood function equation (14) is equivalent to minimizing its negative logarithmic function:
minimizing equation (15) is equivalent to solving equation (13);
step S422, the objective function is further converted into a maximum posterior probability estimation problem and feature weights are solved:
setting xi i Obeys the Γ distribution ζ i ～Γ(γ 1 ,γ 2), wherein ,γ1 and γ2 Is a parameter; there is a probability density function:
calculation of xi i The posterior probability distribution of (2) is:
thus, xi i Is subject to Γ distributionGiven->And marking samples by the target domains, and obtaining feature weights by using maximum posterior probability estimation:
step S423, updating the feature weight obtained by solving in an online mode:
order theThe representation is based on +.>Cumulative count when feature weights are updated for each sample, +.>And-> Representing the corresponding accumulated error, +.>And-> wherein ,/> and />Respectively updating the feature weights on line for the corresponding initial values:
step S5, utilizing the feature extractor f obtained in step S2 i (. About.) different types of features are extracted from the target domain sample to be measured and recorded asThe transformation matrix { Θ ] obtained according to step S3 i Public mapping matrix A 0 Unique mapping matrix { A i Step S4, feature weight xi after online update i Will->Depth knowledge integration through feature level and decision level maps to tag space:
taking the index corresponding to the maximum value in c to obtain the predicted tag number of the target to be detectedObtaining a target identification result:
and completing self-adaptive target recognition.
CN202110987414.8A 2021-08-26 2021-08-26 Domain self-adaptive target recognition method based on depth knowledge integration Active CN113657541B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110987414.8A CN113657541B (en) 2021-08-26 2021-08-26 Domain self-adaptive target recognition method based on depth knowledge integration

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110987414.8A CN113657541B (en) 2021-08-26 2021-08-26 Domain self-adaptive target recognition method based on depth knowledge integration

Publications (2)

Publication Number Publication Date
CN113657541A CN113657541A (en) 2021-11-16
CN113657541B true CN113657541B (en) 2023-10-10

Family

ID=78492902

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110987414.8A Active CN113657541B (en) 2021-08-26 2021-08-26 Domain self-adaptive target recognition method based on depth knowledge integration

Country Status (1)

Country Link
CN (1) CN113657541B (en)

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110210545A (en) * 2019-05-27 2019-09-06 河海大学 Infrared remote sensing water body classifier construction method based on transfer learning
CN110598636A (en) * 2019-09-09 2019-12-20 哈尔滨工业大学 Ship target identification method based on feature migration
CN111723823A (en) * 2020-06-24 2020-09-29 河南科技学院 Underwater target detection method based on third-party transfer learning

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180024968A1 (en) * 2016-07-22 2018-01-25 Xerox Corporation System and method for domain adaptation using marginalized stacked denoising autoencoders with domain prediction regularization

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110210545A (en) * 2019-05-27 2019-09-06 河海大学 Infrared remote sensing water body classifier construction method based on transfer learning
CN110598636A (en) * 2019-09-09 2019-12-20 哈尔滨工业大学 Ship target identification method based on feature migration
CN111723823A (en) * 2020-06-24 2020-09-29 河南科技学院 Underwater target detection method based on third-party transfer learning

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
陶洋 ; 胡昊 ; 鲍灵浪 ; .基于重建分类网络特征增强的盲域自适应分类器.信息通信.2020,(第06期),全文. *

Also Published As

Publication number Publication date
CN113657541A (en) 2021-11-16

Similar Documents

Publication Publication Date Title
CN110443143B (en) Multi-branch convolutional neural network fused remote sensing image scene classification method
CN110516095B (en) Semantic migration-based weak supervision deep hash social image retrieval method and system
Han et al. Combining 3D‐CNN and Squeeze‐and‐Excitation Networks for Remote Sensing Sea Ice Image Classification
CN110210545B (en) Infrared remote sensing water body classifier construction method based on transfer learning
CN112395987B (en) SAR image target detection method based on unsupervised domain adaptive CNN
CN112818822B (en) Automatic identification method for damaged area of aerospace composite material
CN110991257B (en) Polarized SAR oil spill detection method based on feature fusion and SVM
CN110728694B (en) Long-time visual target tracking method based on continuous learning
CN108428220A (en) Satellite sequence remote sensing image sea island reef region automatic geometric correction method
CN110414616B (en) Remote sensing image dictionary learning and classifying method utilizing spatial relationship
CN112347930B (en) High-resolution image scene classification method based on self-learning semi-supervised deep neural network
CN112084871B (en) High-resolution remote sensing target boundary extraction method based on weak supervised learning
CN111259733A (en) Point cloud image-based ship identification method and device
CN117152503A (en) Remote sensing image cross-domain small sample classification method based on false tag uncertainty perception
CN113344045A (en) Method for improving SAR ship classification precision by combining HOG characteristics
CN115019103A (en) Small sample target detection method based on coordinate attention group optimization
CN117152606A (en) Confidence dynamic learning-based remote sensing image cross-domain small sample classification method
CN115908924A (en) Multi-classifier-based small sample hyperspectral image semantic segmentation method and system
Li et al. Airplane detection using convolutional neural networks in a coarse-to-fine manner
CN113723572B (en) Ship target identification method, computer system, program product and storage medium
CN109558803B (en) SAR target identification method based on convolutional neural network and NP criterion
Yang et al. Attention-based dynamic alignment and dynamic distribution adaptation for remote sensing cross-domain scene classification
Liu et al. A novel deep framework for change detection of multi-source heterogeneous images
Rafi et al. Attention-based domain adaptation for hyperspectral image classification
CN113657541B (en) Domain self-adaptive target recognition method based on depth knowledge integration

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