CN113516231A - A DSN-based Deep Adversarial Transfer Network for Recognition of Daily Behavior Transfer - Google Patents
A DSN-based Deep Adversarial Transfer Network for Recognition of Daily Behavior Transfer Download PDFInfo
- Publication number
- CN113516231A CN113516231A CN202110913768.8A CN202110913768A CN113516231A CN 113516231 A CN113516231 A CN 113516231A CN 202110913768 A CN202110913768 A CN 202110913768A CN 113516231 A CN113516231 A CN 113516231A
- Authority
- CN
- China
- Prior art keywords
- daily behavior
- domain
- daily
- target
- domains
- 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.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; 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 OR CALCULATING; 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 OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; 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
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Molecular Biology (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Probability & Statistics with Applications (AREA)
- Image Analysis (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
本发明提供一种基于DSN深度对抗迁移网络的日常行为迁移识别方法，包括：获取若干个候选源域及目标域；分别将候选源域与目标域的日常行为标签及传感器映射到同一空间；在该同一空间内获取与日常行为标签对应的日常行为特征向量，从而由候选源域中筛选出目标域的相似源域；采用领域自适应方法拉近各相似源域与目标域的特征向量分布；分别将每个相似源域与目标域的特征向量组合后作为DSN网络的输入，从而训练获得与相似源域数量相当的基分类器；对各个基分类器对目标域内特征向量的分类结果进行集成学习，从而获得目标域的日常行为识别结果。本发明采用深度对抗网络迁移DSN方法，可自动化地提取更具表现力的特征，效果好于手动提取特征。
The present invention provides a method for recognizing daily behavior transfer based on DSN deep confrontation transfer network, comprising: acquiring several candidate source domains and target domains; mapping daily behavior labels and sensors of the candidate source domains and target domains to the same space respectively; The daily behavior feature vector corresponding to the daily behavior label is obtained in the same space, so that the similar source domains of the target domain are screened out from the candidate source domains; the domain adaptive method is used to narrow the feature vector distribution of each similar source domain and the target domain; Combine the feature vectors of each similar source domain and the target domain as the input of the DSN network, so as to train and obtain a base classifier with a similar number of source domains; integrate the classification results of each base classifier on the feature vectors in the target domain learning, so as to obtain the daily behavior recognition results of the target domain. The invention adopts the deep confrontation network migration DSN method, which can automatically extract more expressive features, and the effect is better than manually extracting features.
Description
Technical Field
The invention relates to the technical field of smart home, in particular to a daily behavior migration recognition method based on a DSN deep antagonistic migration network.
Background
The behavior recognition problem based on the heterogeneous smart home environment mainly comprises the mapping problem of the sensor and the daily behavior, the daily behavior recognition problem and the problem that a target domain is lack of marking data. The existing technology such as XLaarn is a method for identifying behaviors in a heterogeneous environment by using an ensemble learning technology method under a knowledge-driven background, and mainly comprises the following steps: an ontology is first constructed to perform feature and daily behavior space remapping. The unlabeled daily behavior is then labeled with the clusters, and ensemble learning inputs are generated therefrom. And finally, processing the data which are not labeled in the cluster by utilizing the integrated learning to identify daily behaviors.
The above method has the following disadvantages:
first, the method requires manual feature extraction and no more expressive features can be obtained.
Secondly, the method does not select proper features for migration, nor selects similar data sets for migration, which may cause negative migration.
Disclosure of Invention
In view of the defects of the prior art, the invention provides a daily behavior migration recognition method based on a DSN (Domain Separation Networks) deep antagonistic migration network. Compared with the existing heterogeneous intelligent home behavior identification method, the method can utilize the DSN deep generation countermeasure network to automatically extract the expressive force characteristics, and simultaneously, the integrated learning and field self-adaption method is added to achieve the purposes of identifying the unique daily behavior of the target domain by the similar multi-source domain transfer learning and improving the accuracy of daily behavior identification.
The technical means adopted by the invention are as follows:
a daily behavior migration recognition method based on a DSN deep antagonistic migration network comprises the following steps:
acquiring a plurality of candidate source domains and target domains, wherein the daily behavior labels in the candidate source domains are known, and the daily behavior labels in the target domains are partially known or completely unknown;
respectively mapping daily behavior labels and sensors of the candidate source domain and the target domain to the same space;
acquiring daily behavior feature vectors corresponding to the daily behavior labels in the same space, and processing the daily behavior feature vectors based on a distance measurement method, so as to screen out similar source domains of the target domain from the candidate source domains;
adopting a domain self-adaptive method to approximate the feature vector distribution of each similar source domain and target domain;
combining the feature vectors of each similar source domain and the target domain respectively to be used as the input of the DSN, thereby training and obtaining the base classifiers with the quantity equivalent to that of the similar source domains;
and performing ensemble learning on the classification result of the feature vector in the target domain by each base classifier so as to obtain the daily behavior recognition result of the target domain.
Further, mapping the daily behavior labels of the candidate source domain and the target domain to the same space includes:
extracting all known daily behavior labels of the candidate source domain and the target domain;
putting an original daily behavior label into a Word2vec model for training to obtain a daily behavior label digital feature vector which corresponds to the daily behavior label and has semantic content;
and dividing the daily behavior labels corresponding to the two daily behavior label digital feature vectors exceeding a certain threshold into the same daily behavior by utilizing the cosine similarity and the distance between the daily behavior label digital feature vectors, thereby obtaining a daily behavior template integrated by similar daily behaviors and completing the mapping of the daily behaviors.
Further, mapping the sensors of the candidate source domain and the target domain to the same space includes:
acquiring configuration vectors of all sensors of a candidate source domain and a candidate target domain, wherein the configuration vectors comprise positions, occurrence frequency and types of each daily behavior;
inputting the sensor configuration vector into a Word2vec model for training to obtain a sensor data vector with digital characteristics corresponding to the type of the sensor;
and clustering the sensor data vectors, and taking the sensors corresponding to the sensor data vectors in the same cluster as the same sensor based on the clustering result to complete sensor mapping.
Further, the method based on distance measurement processes each daily behavior feature vector, so as to screen out a similar source domain of the target domain from the candidate source domain, including:
acquiring daily behavior feature vectors of each candidate source domain and each candidate target domain in a mapping space;
and respectively calculating the distance characteristics of the daily behavior characteristic vector of each candidate source domain and the daily behavior characteristic vector of the target source domain, and screening out the candidate source domains with the distance characteristics meeting the preset requirements as similar source domains.
Compared with the prior art, the invention has the following advantages:
1. the invention adopts a DSN method for deeply resisting network migration, can automatically extract more expressive characteristics, and has better effect than manually extracting the characteristics. The unique network structure of the DSN keeps the uniqueness of the characteristics, the similarity of the characteristic sharing part is improved by utilizing the generation countermeasure thought, meanwhile, the DSN network can also keep the unique part of the characteristics of each domain, and the negative migration is effectively avoided. And the extracted features and the integrated learning method are utilized, so that the purpose of identifying the unique daily behaviors of the target domain or mutually identifying the daily behaviors in multiple domains by utilizing multiple source domains can be achieved.
2. The method creates innovations for the sensor and the daily behavior mapping method. And mapping is carried out through a word2vec semantic model and a clustering method. The daily behaviors can be better identified by utilizing semantic model mapping.
3. The method utilizes the distance measurement to select the similar data sets, utilizes a field self-adaptive method to improve the similarity between the sample characteristics, and can improve the effect of transfer learning compared with the existing method.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the description of the embodiments or the prior art will be briefly introduced below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to these drawings without creative efforts.
FIG. 1 is a flow chart of multi-source domain daily behavior recognition using DSN as a base classifier according to the present invention.
FIG. 2 is a flowchart illustrating the mapping of multi-domain daily behavior tags according to the present invention.
FIG. 3 is a flow chart of multi-domain sensor mapping in the present invention.
Fig. 4 is a plan view of an intelligent home apartment where a data set HH102 published by Washington State University casas (center for Advanced students in Adaptive systems) is located.
Fig. 5 is a plan view of an intelligent home apartment where a data set HH103 published by Washington State University casas (center for Advanced students in Adaptive systems) is located.
Fig. 6 is a plan view of an intelligent home apartment where a data set HH104 published by Washington State University casas (center for Advanced students in Adaptive systems) is located.
Fig. 7 is a plan view of an intelligent home apartment where a data set HH105 published by Washington State University casas (center for Advanced students in Adaptive systems) is located.
Fig. 8 is a plan view of an intelligent home apartment where a data set HH106 published by Washington State University casas (center for Advanced students in Adaptive systems) is located.
Detailed Description
In order to make the technical solutions of the present invention better understood, 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 by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
It should be noted that the terms "first," "second," and the like in the description and claims of the present invention and in the drawings described above are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used is interchangeable under appropriate circumstances such that the embodiments of the invention described herein are capable of operation in sequences other than those illustrated or described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.
As shown in fig. 1 to 3, the present invention provides heterogeneous smart home behavior recognition based on a DSN deep-countermeasure migration network, including the following steps:
s1, obtaining a plurality of candidate source domains and target domains, wherein the daily behavior labels in the candidate source domains are known, and the daily behavior labels in the target domains are partially known or totally unknown.
And S2, mapping the daily behavior labels and the sensors of the candidate source domain and the target domain to the same space respectively.
Specifically, mapping the daily behavior labels of the candidate source domain and the target domain to the same space, as shown in fig. 2, includes:
s211, extracting all known daily behavior labels of the candidate source domain and the target domain;
s212, putting the original daily behavior label into a Word2vec model for training to obtain a daily behavior label digital feature vector which corresponds to the daily behavior label and has semantics;
and S213, dividing the daily behavior labels corresponding to the two daily behavior label digital feature vectors exceeding a certain threshold into the same daily behavior by using the cosine similarity and the distance between the daily behavior label digital feature vectors, thereby obtaining a daily behavior template integrated by similar daily behaviors and completing daily behavior mapping. Wherein, the threshold value is a value with better result which is selected in advance according to the experimental result and is used as the standard measurement. The cosine similarity formula is as follows:
wherein x and y are two different daily behavior label digital feature vectors respectively.
Further, mapping the sensors of the candidate source domain and the target domain to the same space, as shown in fig. 3, includes:
s221, obtaining configuration vectors of all sensors of the candidate source domain and the candidate target domain, wherein the configuration vectors comprise positions, occurrence frequency and types of each daily behavior;
s222, inputting the sensor configuration vector into a Word2vec model for training to obtain a sensor data vector with digital characteristics corresponding to the sensor type;
and S223, clustering the sensor data vectors, and taking the sensors corresponding to the sensor data vectors in the same cluster as the same sensor based on the clustering result to complete sensor mapping. Preferably, the K-Means method and the DBSCAN method are used for clustering, and the method with the better clustering effect is selected. In a preferred embodiment of the invention, a K-Means method is adopted, which is also named as a K-Means algorithm, wherein K represents that the clusters are K clusters, and Means represents that the mean value of the data values in each cluster is taken as the center of the cluster. The algorithm idea is roughly as follows: firstly, randomly selecting K vectors from a sample set as cluster centers, calculating the distance between all the vectors and the K cluster centers, dividing each vector into the cluster where the cluster center closest to the vector is located, and calculating the new cluster center of each cluster for the new cluster until the cluster center does not move.
And S3, acquiring daily behavior feature vectors corresponding to the daily behavior labels in the same space, and processing the daily behavior feature vectors based on a distance measurement method, so as to screen out similar source domains of the target domain from the candidate source domains.
Specifically, the method shown in fig. 2 and 3 is firstly adopted to map the sensor, daily activities of the randomly selected multi-source domain and the target domain to the same space. The extracted sensor configuration vectors include: location, sensor type, frequency of triggers in each day-to-day activity. Sensor type: magnetic door sensor, light sensor, infrared motion sensor, wide area infrared motion sensor, temperature sensor. The daily behavior of each data set: rest, clean, go out, go home, sleep, cook, eat, wash dishes, go to the toilet, work. Then, feature vectors of daily behaviors are obtained, and a source domain similar to a target domain is found by using a Distance measurement method (a Rank of domain method in a Wassertein Distance method or a GFK (Geodesic Flow Kernel) method). Wherein, the feature vector of the daily behavior comprises: the starting time of the daily behavior, the ending time of the daily behavior, the duration time of the daily activity, the proportion of the mapped sensor streams of the daily behavior, and the daily behavior label. And respectively calculating the distance characteristics of the daily behavior characteristic vector of each candidate source domain and the daily behavior characteristic vector of the target source domain, and screening out the candidate source domains with the distance characteristics meeting the preset requirements as similar source domains.
And S4, adopting a domain self-adaptive method to approximate the feature vector distribution of each similar source domain and target domain.
Specifically, by mapping similar source domain and target domain sensor, daily behaviors to the same space, we obtain daily behavior feature vectors of the same dimension. And then observing the domain distribution, and utilizing a proper domain self-adaptive method to draw the feature vector distribution of the similar source domain and the target domain, so that the feature vector distribution of the source domain and the target domain is closer, and the training effect in the DSN network is better. In a preferred embodiment of the invention, the domain adaptive method is, for example, a TCA method: it is assumed that there is a feature mapping such that the edge distributions of the mapped source domain and target domain are close, wherein the feature mapping is implemented using MMD distance metrics and kernel functions. As can be seen from the probability distribution map, the probability distributions of the source domain and the target domain become more overlapping.
And S5, combining the feature vector of each similar source domain after being drawn and the feature vector of the target domain respectively to be used as the input of the DSN, thereby training and obtaining the base classifiers with the quantity equivalent to that of the similar source domains. Based on each base classifier, a primary classification result output by a plurality of base classifiers is obtained, namely, data of a target domain is marked by a label approximate to a source domain.
And S6, performing ensemble learning on the classification result of the feature vector in the target domain by each base classifier, thereby obtaining the daily behavior recognition result of the target domain.
Specifically, the results of the plurality of basis classifiers are put into ensemble Learning, and the ensemble Learning method of Stacking Learning is preferably used in this embodiment. The output of each base classifier is added with weight, the predicted output of each base classifier to the same characteristic is used as the input of the last classifier, and the target domain data is labeled again. The aim of identifying the daily behaviors (including the unique daily behaviors) of the target domain by the multi-source domain is fulfilled.
The following further describes the aspects and effects of the present invention based on specific application examples.
Example 1: daily behaviors of the target domain are identified by using daily behavior sensor event streams collected at two different smart home apartments (the two apartment data are processed by similarity, namely similar source domains are selected for migration). After daily behavior mapping and sensor mapping, feature spaces of all daily behaviors are unified. The common daily behaviors of three apartments are: eating, sleeping, cooking, resting, cleaning, going out, going home, washing dishes, working, bathing and going to the toilet. The first nine daily behaviors are common daily behaviors of the three data sets, bathing only occurs in a first source domain and a target domain, and toileting only occurs in a second source domain and the target domain. The daily behaviors of the target domain can be identified by a plurality of source domains through integrated learning, and the problem that a certain source domain does not have the identification of a certain daily behavior of the target domain is solved. That is, all daily activities of the target domain can be identified by using the two source domains (as long as one source domain in the multi-source domain has the daily activities). The data set of the two apartments can identify daily behaviors of all categories of the third apartment, and the accuracy of identification is improved by using the DSN network extraction features as a base classifier.
Selecting a data set: data sets published by Washington State University (Washington State University) CASAS (center for Advanced students in Adaptive systems). Wherein CASAS is the largest and most widely used daily behavior recognition dataset in scale so far. The daily behavior datasets-hh 102, hh103, hh104, hh105 dataset (as preselected source domain), and hh106 dataset (as target domain) were randomly selected among them for a single user. The five data sets respectively consist of sensor information collected by different volunteers in different intelligent household environment layouts. Sensor information: time of trigger, location, sensor type, sensor name, daily behavior tag. Sensor type: magnetic door sensor, light sensor, infrared motion sensor, wide area infrared motion sensor, temperature sensor. The daily behavior of each data set: rest, clean, go out, go home, sleep, cook, eat, wash dishes, go to the toilet, work. For each dataset, redundant sensor sequences without daily activity markers are deleted, leaving only labeled sensor sequences. The environment layout of each data set and the installation position of the sensor are shown in fig. 4, 5, 6, 7, and 8 in this order.
Mapping the sensors and the daily behavior labels of the five domains to the same feature space by the methods of the figure 1 and the figure 2 respectively, and obtaining the feature vector of each domain: the starting time of the daily behavior, the ending time of the daily behavior, the duration time of the daily behavior, the proportion of the mapped sensors in the sensor stream of the daily behavior, and the daily behavior label. The feature vector of this daily activity, such as sleeping, can be represented as the following numerical features (22:00,7:00,9,1/3,1/3,0,0,0,0,0,0,1/3,0,0.2,0.2, 0.6).
Through distance measurement, a plurality of source domains similar to the target domain are found: hh102, hh 103. And repeating the second step, and mapping the sensors and daily behaviors of the similar source domain and the target domain to obtain new feature vectors of each domain. And then, respectively approximating the distribution of the feature vectors of hh102, hh103 and the target domain hh106 by using a domain adaptive method. Mapped daily behavior templates (removing tense, near word interferences): eating, sleeping, cooking, resting, cleaning, going out, going home, washing dishes, working, bathing and going to the toilet. The mapped sensor code numbers are numbers 1-10. To demonstrate the functionality of the method, when processing the data sets, we consider the first nine daily activities to be common daily activities of the three data sets, bathing only occurring in hh102, hh106, and toileting only occurring in hh103, hh 106.
Now there are two source domains similar to the target domain, so using the combination one: hh102, hh106, and combination two: training the DSN by using the characteristic vectors of hh103 and hh106 to obtain two base classifiers, and respectively marking the characteristic vectors of the target domain by using the labels of the source domains. The marking result can be observed, the recognition effect on the public daily behaviors and bathing in the target domain is good in the first combination, and the recognition on the toilet daily behaviors in hh106 is not ideal. The same principle is used for combining the two recognition effects.
The result of the base classifier is put into the integrated learning model for calibration again, and the result shows that after integrated learning, the identification of the daily behaviors of the target domain is good, including the daily behaviors of bathing and toileting types (the daily behaviors of the target domain do not exist in a single source domain), and meanwhile, the identification precision of the public daily behaviors is improved.
In the above-described embodiments of the present invention, the description of the daily behaviors in each apartment has each emphasis, and the recognition effect of the daily behaviors that do not exist in a single source domain but exist in a target domain is highlighted.
In the embodiments provided in the present application, it should be understood that the disclosed technical content can be implemented in other ways. The above-described clustering, distance measurement, and method for improving similarity between the source domain and the target domain are merely exemplary, and may be converted into other methods with better effect.
Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some or all of the technical features may be equivalently replaced; and the modifications or the substitutions do not make the essence of the corresponding technical solutions depart from the scope of the technical solutions of the embodiments of the present invention.
Claims (4)
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110913768.8A CN113516231B (en) | 2021-08-10 | 2021-08-10 | A daily behavior migration recognition method based on DSN deep adversarial migration network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110913768.8A CN113516231B (en) | 2021-08-10 | 2021-08-10 | A daily behavior migration recognition method based on DSN deep adversarial migration network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113516231A true CN113516231A (en) | 2021-10-19 |
| CN113516231B CN113516231B (en) | 2024-03-29 |
Family
ID=78068939
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110913768.8A Active CN113516231B (en) | 2021-08-10 | 2021-08-10 | A daily behavior migration recognition method based on DSN deep adversarial migration network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113516231B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114357998A (en) * | 2021-12-28 | 2022-04-15 | 中国人民解放军战略支援部队信息工程大学 | A security attribute calibration method, system, device and computer storage medium |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106056043A (en) * | 2016-05-19 | 2016-10-26 | 中国科学院自动化研究所 | Animal behavior identification method and apparatus based on transfer learning |
| US9565521B1 (en) * | 2015-08-14 | 2017-02-07 | Samsung Electronics Co., Ltd. | Automatic semantic labeling based on activity recognition |
| CN108960270A (en) * | 2018-04-08 | 2018-12-07 | 中国科学院计算技术研究所 | A kind of data scaling method and system based on manifold transfer learning |
| US20190147854A1 (en) * | 2017-11-16 | 2019-05-16 | Microsoft Technology Licensing, Llc | Speech Recognition Source to Target Domain Adaptation |
| WO2020191282A2 (en) * | 2020-03-20 | 2020-09-24 | Futurewei Technologies, Inc. | System and method for multi-task lifelong learning on personal device with improved user experience |
| CN112235264A (en) * | 2020-09-28 | 2021-01-15 | 国家计算机网络与信息安全管理中心 | Network traffic identification method and device based on deep migration learning |
| CN112801718A (en) * | 2021-02-22 | 2021-05-14 | 平安科技（深圳）有限公司 | User behavior prediction method, device, equipment and medium |
| CN112800352A (en) * | 2021-02-05 | 2021-05-14 | 大连海事大学 | A Heterogeneous Migration Behavior Recognition Method Based on the Combination of Features and Instances |
| CN113157094A (en) * | 2021-04-21 | 2021-07-23 | 杭州电子科技大学 | Electroencephalogram emotion recognition method combining feature migration and graph semi-supervised label propagation |
| US20210233239A1 (en) * | 2020-01-24 | 2021-07-29 | GE Precision Healthcare LLC | Systems and methods for medical image style transfer using deep neural networks |
-
2021
- 2021-08-10 CN CN202110913768.8A patent/CN113516231B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9565521B1 (en) * | 2015-08-14 | 2017-02-07 | Samsung Electronics Co., Ltd. | Automatic semantic labeling based on activity recognition |
| CN106056043A (en) * | 2016-05-19 | 2016-10-26 | 中国科学院自动化研究所 | Animal behavior identification method and apparatus based on transfer learning |
| US20190147854A1 (en) * | 2017-11-16 | 2019-05-16 | Microsoft Technology Licensing, Llc | Speech Recognition Source to Target Domain Adaptation |
| CN108960270A (en) * | 2018-04-08 | 2018-12-07 | 中国科学院计算技术研究所 | A kind of data scaling method and system based on manifold transfer learning |
| US20210233239A1 (en) * | 2020-01-24 | 2021-07-29 | GE Precision Healthcare LLC | Systems and methods for medical image style transfer using deep neural networks |
| WO2020191282A2 (en) * | 2020-03-20 | 2020-09-24 | Futurewei Technologies, Inc. | System and method for multi-task lifelong learning on personal device with improved user experience |
| CN112235264A (en) * | 2020-09-28 | 2021-01-15 | 国家计算机网络与信息安全管理中心 | Network traffic identification method and device based on deep migration learning |
| CN112800352A (en) * | 2021-02-05 | 2021-05-14 | 大连海事大学 | A Heterogeneous Migration Behavior Recognition Method Based on the Combination of Features and Instances |
| CN112801718A (en) * | 2021-02-22 | 2021-05-14 | 平安科技（深圳）有限公司 | User behavior prediction method, device, equipment and medium |
| CN113157094A (en) * | 2021-04-21 | 2021-07-23 | 杭州电子科技大学 | Electroencephalogram emotion recognition method combining feature migration and graph semi-supervised label propagation |
Non-Patent Citations (2)
| Title |
|---|
| GARRETT WILSON ET AL: "A Survey of Unsupervised Deep Domain Adaptation", ACM TRANSACTIONS ON INTELLIGENT SYSTEMS AND TECHNOLOGY, 5 July 2020 (2020-07-05), pages 1 - 46, XP058680020, DOI: 10.1145/3400066 * |
| YUN ZHANG ET AL: "Unsupervised Domain Adaptation by Mapped Correlation Alignment", IEEE ACCESS ( VOLUME: 6), 13 August 2018 (2018-08-13), pages 44698 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114357998A (en) * | 2021-12-28 | 2022-04-15 | 中国人民解放军战略支援部队信息工程大学 | A security attribute calibration method, system, device and computer storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113516231B (en) | 2024-03-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Arrotta et al. | Dexar: Deep explainable sensor-based activity recognition in smart-home environments | |
| Sener et al. | Unsupervised semantic parsing of video collections | |
| Sharma et al. | Classification in pattern recognition: A review | |
| CN107205016B (en) | Retrieval method of Internet of things equipment | |
| Prossegger et al. | Multi-resident activity recognition using incremental decision trees | |
| CN106650202B (en) | A kind of the room area layout prediction technique and system of data-driven | |
| Zhou et al. | Multiview deep graph infomax to achieve unsupervised graph embedding | |
| Pronobis et al. | Hierarchical Multi-Modal Place Categorization. | |
| CN112380350A (en) | Text classification method and device | |
| CN113392761A (en) | Component identification method, device, equipment and storage medium | |
| Kazllarof et al. | Active learning Rotation Forest for multiclass classification | |
| US20130006906A1 (en) | Knowledge Transfer in Smart Environments | |
| CN113516231A (en) | A DSN-based Deep Adversarial Transfer Network for Recognition of Daily Behavior Transfer | |
| Hardegger et al. | Enhancing action recognition through simultaneous semantic mapping from body-worn motion sensors | |
| Nadi et al. | A profile based data segmentation for in-home activity recognition | |
| Henriques et al. | Spatial clustering using hierarchical SOM | |
| Ammar et al. | Open-set object detection: towards unified problem formulation and benchmarking | |
| Zhang et al. | GSpect: Spectral filtering for cross-scale graph classification | |
| CN110569761A (en) | A Hand-drawn Sketch Retrieval Method for Remote Sensing Images Based on Adversarial Learning | |
| CN113779248A (en) | Data classification model training method, data processing method and storage medium | |
| Kapidis et al. | Object detection-based location and activity classification from egocentric videos: A systematic analysis | |
| TWI665566B (en) | System and method for product classification | |
| CN117392714A (en) | Face beauty prediction method, equipment and media based on semi-supervised learning | |
| CN112988699B (en) | Model training method, data label generation method and device | |
| JP7177013B2 (en) | Information processing device, information processing method and information processing program |
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 |