[go: up one dir, main page]

WO2023016168A1 - 信号识别方法及装置、计算机可读存储介质 - Google Patents

信号识别方法及装置、计算机可读存储介质 Download PDF

Info

Publication number
WO2023016168A1
WO2023016168A1 PCT/CN2022/104966 CN2022104966W WO2023016168A1 WO 2023016168 A1 WO2023016168 A1 WO 2023016168A1 CN 2022104966 W CN2022104966 W CN 2022104966W WO 2023016168 A1 WO2023016168 A1 WO 2023016168A1
Authority
WO
WIPO (PCT)
Prior art keywords
signal
signal data
data
network
domain
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.)
Ceased
Application number
PCT/CN2022/104966
Other languages
English (en)
French (fr)
Inventor
陈建强
李云
曹啡
徐殿平
张海军
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.)
ZTE Corp
Original Assignee
ZTE Corp
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 ZTE Corp filed Critical ZTE Corp
Publication of WO2023016168A1 publication Critical patent/WO2023016168A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/10—Pre-processing; Data cleansing
    • 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
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • 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
    • 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]

Definitions

  • the embodiments of the present application relate to but are not limited to the technical field of wireless communication, and in particular, relate to a signal identification method and device, and a computer-readable storage medium.
  • Wireless communication technology because of its mobility, high efficiency, low-cost deployment and maintenance, no need for difficult wiring engineering and maintenance measures, highly flexible networking operations, good scalability, and simple user addition and deletion operations without It will affect the overall performance of the communication network and other advantages, and has been developed rapidly in recent years.
  • ISM International Scientific Medical
  • a frequency band that does not require a license or fee that is, the ISM frequency band, has been set up, and services such as wireless metropolitan area network and wireless local area network have been generated, making more and more Many products using this frequency band have a wider range of applications in various fields such as smart home, medical health and public infrastructure.
  • interference identification methods usually need to analyze and identify based on a large amount of signal data in the area, which consumes a lot of resources and is not simple, and does not actually consider the impact of field offset caused by different signal data collection environments , so the recognition accuracy is not high.
  • Embodiments of the present application provide a signal identification method and device, and a computer-readable storage medium.
  • the embodiment of the present application provides a signal identification method, including: obtaining wireless signal data from the target domain; determining the second identification network for the target domain signal according to the wireless signal data and the first identification network for the source domain signal Identifying networks, the first identifying network is derived from the first signal data obtained from the source domain.
  • the embodiment of the present application also provides a signal identification device, including: a data acquisition module configured to acquire wireless signal data from the target domain; a data processing module configured to A first recognition network for a domain signal determines a second recognition network for a target domain signal, the first recognition network being derived from first signal data obtained from a source domain.
  • the embodiment of the present application also provides a signal identification device, including: a memory, a processor, and a computer program stored in the memory and operable on the processor, and the processor implements the computer program when executing the computer program.
  • a signal identification device including: a memory, a processor, and a computer program stored in the memory and operable on the processor, and the processor implements the computer program when executing the computer program.
  • the embodiment of the present application further provides a computer-readable storage medium storing computer-executable instructions, and the computer-executable instructions are used to execute the signal identification method in the first aspect as described above.
  • Fig. 1 is a flowchart of a signal recognition method provided by an embodiment of the present application
  • Fig. 2 is a flow chart of acquiring wireless signal data in a signal identification method provided by an embodiment of the present application
  • Fig. 3 is a flowchart of determining wireless signal data in a signal identification method provided by an embodiment of the present application
  • Fig. 4 is a flow chart of determining the second identification network in the signal identification method provided by one embodiment of the present application.
  • Fig. 5 is a flow chart of updating the first identification network in the signal identification method provided by one embodiment of the present application.
  • Fig. 6 is a flow chart after determining wireless signal data in the signal identification method provided by an embodiment of the present application.
  • Fig. 7 is a flow chart after determining the signal data to be identified in the signal identification method provided by one embodiment of the present application.
  • Fig. 8 is a schematic diagram of execution of a signal identification method provided by an embodiment of the present application.
  • Fig. 9 is a schematic diagram of execution of the WADA algorithm provided by an embodiment of the present application.
  • Fig. 10 is a schematic diagram of a signal identification device provided by an embodiment of the present application.
  • Fig. 11 is a schematic diagram of a signal identification device provided by another embodiment of the present application.
  • the present application provides a signal identification method and device, and a computer-readable storage medium.
  • the first identification network capable of identifying the source domain signal
  • the second identification network so as to realize the domain transfer of the identification signal, is conducive to improving the accuracy of signal identification, and because the signal identification can be realized through wireless signal data, it is not necessary to analyze and identify based on a large amount of signal data in the area, thus It can save resources and reduce the difficulty of signal identification.
  • FIG. 1 is a flow chart of a signal identification method provided by an embodiment of the present application, and the signal identification method includes but not limited to step S100 and step S200 .
  • Step S100 acquiring wireless signal data from the target domain.
  • the target domain is distinguished from the source domain, wherein the source domain represents a known environment or an original environment.
  • the source domain represents a known environment or an original environment.
  • wireless signals in the source domain have been well identified; the target domain represents an unknown environment or a new environment.
  • the target domain represents an unknown environment or a new environment.
  • there are domain differences between the target domain and the source domain there are domain differences between the target domain and the source domain.
  • this may bring about differences in signal recognition, so the signal recognition of the target domain needs to be further expanded.
  • wireless signal data there may be multiple types of wireless signal data.
  • different types of wireless signal data it can be processed in a subsequent selection and adaptive manner.
  • the wireless signal data is mainly used as the The WIFI signal is described, and the corresponding data processing method is also described based on the fact that the wireless signal data is a WIFI signal, but those skilled in the art can make corresponding settings according to actual application scenarios, which is not limited in this embodiment.
  • step S100 includes but not limited to steps S110 to S130.
  • Step S110 obtaining original signal data from the target domain
  • Step S120 screening the original signal data to obtain second signal data carrying effective signals
  • Step S130 determining wireless signal data from the second signal data.
  • the invalid second signal data can be eliminated by screening the original signal data, so as to further obtain valid wireless signal data according to the second signal data, so as to ensure that the wireless signal data has a good practical effect, so as to improve the subsequent Process the reliability of wireless signal data, thereby improving the accuracy of signal recognition.
  • a wireless access point can be set to obtain the original signal data, and at least one terminal connected to the AP can be set for analysis, that is, the obtained original signal data can be obtained by the AP.
  • the signal data is transmitted to the terminal, and then the terminal is used to analyze and screen the original signal data so as to extract effective second signal data as signal samples.
  • the corresponding data carrying noise signals in the original signal data can be filtered out to obtain all data that does not contain noise signals.
  • the signal other than the noise signal is the effective signal.
  • the effective signal can have many types, including the signal that transmits the information required by the user, or is used to make the receiving device generate a preset signal after receiving the signal. Action signals, etc.
  • all the data obtained by screening are the second signal data, but the "effective signal" here is not a restrictive concept, and can be distinguished according to different actual application scenarios, not limited to the elimination of noise signals , those skilled in the art can set the distinguishing form by themselves, which is not limited in this embodiment.
  • multiple terminals can be set to match the AP, and each terminal can be called an access terminal, a user equipment (User Equipment, UE), a user unit, a user station, a mobile station, a mobile station, or a remote station , remote terminal, mobile device, user terminal, wireless communication device, user agent, or user device.
  • UE User Equipment
  • each terminal may be a cellular phone, a cordless phone, a Session Initiation Protocol (SIP) phone, a Wireless Local Loop (WLL) station, a Personal Digital Assistant (PDA), a Handheld devices with wireless communication functions, computing devices or other processing devices connected to wireless modems, vehicle-mounted devices, wearable devices, 5G networks or terminal devices in future 5G or higher networks, etc., are not specifically limited in this embodiment.
  • SIP Session Initiation Protocol
  • WLL Wireless Local Loop
  • PDA Personal Digital Assistant
  • step S130 includes but not limited to step S131.
  • Step S131 determining from the second signal data that the second signal data having a wireless signal protocol stack indication bit is wireless signal data.
  • the wireless signal protocol stack defines the characteristics of communication hardware and software for coordinating work at different levels for wireless signals
  • the wireless signal protocol stack can be determined It has been set, indicating that the wireless signal exists correspondingly, so that it can be determined that the second signal data with the indication bit of the wireless signal protocol stack is wireless signal data. Therefore, the wireless signal data can be easily and reliably screened in this way, and the signal classification can be improved.
  • the wireless signal protocol stack indicator bit is the WIFI signal protocol stack indicator bit, and so on, which will not be repeated here.
  • Step S200 determining a second identification network for a signal in a target domain according to wireless signal data and a first identification network for a signal in a source domain, where the first identification network is obtained from the first signal data obtained from the source domain.
  • the source domain is a known environment, and the relevant signal can be accurately identified according to the first identification network, because the first identification network is obtained from the first signal data in the source domain, so Taking the deep convolutional neural network as an example, the qualified first recognition network is obtained by training the first signal data, and the target domain is a new environment, which is affected by the domain migration of the signal, so the first recognition network alone cannot accurately Identify related information, in order to solve the above difficulties, on the basis of the first identification network that can identify the source domain signal, cooperate with the wireless signal data obtained from the target domain to determine the second identification network that can identify the target domain signal, so as to realize the identification Signal domain migration, that is, extending the signal recognition domain from the original target domain to the current target domain, can not only identify the wireless signals in the original source domain, but also identify the wireless signals in the current target domain, thus greatly improving the signal quality.
  • the area range of identification can improve the identification accuracy for wireless signals, and because the signal identification can be realized through wireless signal data, it is not necessary to analyze and identify based on a large amount of signal data in the area, and also There is no need to add new network architecture or parameters for auxiliary identification, which can save relevant resources and further reduce the difficulty of signal identification.
  • the first signal data there may be multiple types of the first signal data, for example, it may involve signals such as Bluetooth (Bluetooth) signals, cordless phone (Cordless Phone) signals, and game controller (Game Controller) commonly used in the ISM frequency band. signal, microwave oven (Microwave Oven) signal, wireless fidelity (Wireless Fidelity, WIFI) signal, wireless analog video monitor (Video Monitor) signal and ZigBee signal etc., in other words, the first signal data in the present embodiment can be above One or more of each example, which is not limited in this embodiment.
  • signals such as Bluetooth (Bluetooth) signals, cordless phone (Cordless Phone) signals, and game controller (Game Controller) commonly used in the ISM frequency band. signal, microwave oven (Microwave Oven) signal, wireless fidelity (Wireless Fidelity, WIFI) signal, wireless analog video monitor (Video Monitor) signal and ZigBee signal etc.
  • the first signal data in the present embodiment can be above One or more of each example, which is not limited in this embodiment.
  • first signal data, the first recognition network and the second recognition network can be selected or set up according to the actual application scenario.
  • the following is mainly based on the background of deep convolutional neural network for purposes of illustration, but not limitation.
  • step S200 includes but not limited to step S210 and step S220.
  • Step S210 input the wireless signal data and the first signal data obtained from the source domain into the feature extraction network, so that the feature extraction network extracts and outputs the first sample feature corresponding to the source domain signal and the first sample feature corresponding to the target domain signal Two-sample features;
  • Step S220 updating the first identification network for the source domain signal according to the first sample feature and the second sample feature, and determining the updated first identification network as the second identification network for the target domain signal.
  • the feature extraction network can analyze the wireless signal data and the first signal data obtained from the source domain, it can output the first sample features corresponding to the source domain signal and the second sample corresponding to the target domain signal Features, that is, the signal states of the source domain and the target domain can be known separately, so that the processed first recognition network can be updated according to the signal states of the source domain and the target domain, which is equivalent to the first sample feature.
  • the influence of the second sample feature is further considered to obtain an updated first recognition network, that is, to determine the second recognition network so that the second recognition network can recognize signals in the target domain.
  • step S220 “update the first recognition network for the source domain signal according to the first sample feature and the second sample feature" in step S220 includes but not limited to step S221 and step S222.
  • Step S221 in the case that the first sample feature and the second sample feature are mutually mapped between the source domain and the target domain, determine the first loss function according to the mapping condition, and the mapping condition is used to characterize the first sample before and after mutual mapping
  • the data categories corresponding to the feature and the second sample feature are unchanged;
  • Step S222 updating the first recognition network for the source domain signal according to the first loss function and the preset second loss function.
  • first sample features and second sample features there can be multiple first sample features and second sample features and the numbers can be consistent, and each first sample feature and each second sample feature can correspond to each other, which is beneficial to the first sample
  • the features and the second sample features are mutually mapped to each other.
  • the mapping condition characterizes cycle consistency (Cycle Consistency, CC), indicating that in the case where the first sample feature and the second sample feature are mutually mapped between the source domain and the target domain, regardless of the first sample
  • the feature is mapped to the second sample feature several times, or the second sample feature is mapped to the first sample feature several times, and the category of the first sample feature or the second sample feature before and after the mapping will not change, for example,
  • the category of the first sample feature is the category corresponding to the Bluetooth signal, then the category of the first sample feature after mapping is still the category corresponding to the Bluetooth signal, or, the category of the first sample feature obeys a uniform distribution, then after mapping
  • the transformed data category also obeys the uniform distribution, and the first loss function is determined by mapping conditions, and the first recognition network is updated based on the first loss function and the preset second loss function, so that the first sample features and the second sample
  • the variation of features after data processing will not be too large, which can reduce the error of data processing and improve the data correlation between
  • the above-mentioned first loss function can be minimized by simply associating the target domain Sample data implementation, that is, it may only mine a small part of the common features between the two domains, but cannot characterize the more general similarity features, so the introduction of the preset second loss function can better characterize The characteristics of domain migration can achieve the effect of reducing errors, which is conducive to further improving the accuracy of signal recognition.
  • “mutual mapping” refers to the round-trip mapping between the first sample feature and the second sample feature between the source domain and the target domain, or called circular mapping.
  • a first sample feature can be continuously A second sample feature is mapped from the source domain to the target domain, and then the second sample feature is mapped from the target domain back to another first sample feature in the source domain, which is equivalent to the first sample feature and the second sample feature
  • the two-sample feature completes a basic mutual mapping process between the source domain and the target domain.
  • the above mutual mapping process can also be performed multiple times, but in any case it can be called mutual mapping.
  • a second sample feature can be Continuously map from the target domain to a certain first sample feature in the source domain, and then the first sample feature is mapped from the source domain back to another second sample feature in the target domain.
  • the principle is the same as that described above. The principle of the mapping process is similar and will not be repeated here.
  • the second loss function may include but not limited to at least one of the following types:
  • a second label loss function associated with the wireless signal data is A second label loss function associated with the wireless signal data.
  • the integrity loss function can make up for the possible lack of relevance of the first loss function.
  • the integrity loss function consists of the uniform distribution of sample data on the target domain and accessing the target from any sample data on the source domain.
  • the cross-entropy definition between the probability of the sample data on the domain, when there is a difference in the data category distribution of the source domain and the target domain, the weight for the integrity loss function can be appropriately reduced.
  • the first label loss function and the second label loss function characterize the intra-domain correlation, as a supplement to the inter-domain correlation function, which can further improve the loss adjustment for domain transfer.
  • the framework used in the embodiments of the present application for signal recognition is simple and reliable, easy to implement and process, and is applicable to data collection scenarios in different environments, using the limited wireless signal data and the first signal data through the same feature extraction Network, to achieve the purpose of constructing the feature vectors of two domain samples, and at the same time construct the predicted label cycle consistency of source domain sample features and target domain sample features through inter-domain conversion, combined with integrity regularization requirements and traditional supervised loss as a total
  • the loss function updates the first recognition network to obtain the second recognition network.
  • the second recognition network can not only inherit the invariance of the source domain to the domain offset, but also achieve high recognition of wireless signals in the target domain, improving Domain generalization in the source domain and signal test performance in the target domain.
  • step S132 is also included but not limited to.
  • Step S132 determining from the second signal data that the second signal data except the wireless signal data is the signal data to be identified in the target domain.
  • the wireless signal data is mainly used for data processing so that the second identification network can reliably identify the signal, it is not necessary to re-identify the determined wireless signal data , and because the second signal data is unknown data obtained from the target domain, the second signal data other than the wireless signal data can be used as the signal data to be identified, so as to identify the signal data to be identified, thereby Verify the effect of domain transfer from the source domain to the target domain.
  • wireless signal data may also be identified multiple times, which is not limited in this embodiment.
  • step S133 is also included but not limited to.
  • Step S133 inputting the signal data to be recognized to the second recognition network, so that the second recognition network outputs a signal recognition result for the signal data to be recognized.
  • the second signal data other than the wireless signal data is used as the signal data to be identified to prevent the wireless signal data from affecting the identification, and the signal data to be identified is identified through the second identification network, which can be well detected.
  • the construction effect of the second identification network further verifies the domain transfer effect from the source domain to the target domain.
  • FIG. 8 a schematic diagram of an execution flow for realizing signal identification is given.
  • the original signal data is obtained based on the first AP; then, the first terminal extracts the effective part of the original signal data to obtain the second signal data; then, the second signal data is detected, that is, for the second signal data Extract the WIFI signal protocol stack indicator bit to distinguish the WIFI signal sample from the rest of the interference signal samples in the target domain;
  • the adaptive correction algorithm (hereinafter referred to as "WADA algorithm") realizes the adaptive correction of the first recognition network in the target domain environment, and obtains the second recognition network.
  • the WADA algorithm is Domain Adaptation (Domain Adaptation) in transfer learning.
  • the average accuracy rate of the corrected recognition network for various interference signals in the target domain is more than 90%, while the second recognition network can maintain the accuracy of various interference signals in the source domain.
  • the average accuracy rate also remains above 90%, which can meet the accuracy requirements of signal recognition.
  • FIG. 9 a schematic diagram of the principle of the WADA algorithm is given.
  • the network structure corresponding to the WADA algorithm includes a feature extraction network convolutional neural architecture and a category prediction network fully connected architecture.
  • the basic principles are as follows:
  • association mapping is performed for the sample features of the source domain and the target domain:
  • H(T, P aba ) means cross entropy
  • class() indicates the category of the sample.
  • the overall loss function also includes the labeled cross-entropy loss L cla1 of the source domain and the labeled loss L cla2 of the WIFI data of the target domain.
  • ⁇ 1 , ⁇ 2 ⁇ [0,1] are weight parameters.
  • the first recognition network for the source domain becomes the second recognition network that can adapt to the current target domain environmental data through the correction of WIFI signal samples, that is, the category prediction network is fully connected as shown in Figure 9 Architecture, where the fully connected architecture of the category prediction network is a convolutional network model, including several sequentially connected convolutional layers, batch normalization layers, and pooling layers, and then passes through several fully connected layers after global average pooling , and finally output the category prediction results for the current target domain environment data.
  • the fully connected architecture of the category prediction network is a convolutional network model, including several sequentially connected convolutional layers, batch normalization layers, and pooling layers, and then passes through several fully connected layers after global average pooling , and finally output the category prediction results for the current target domain environment data.
  • an embodiment of the present application also provides a signal identification device, which includes:
  • the data acquisition module 100 is configured to acquire wireless signal data from the target domain
  • the data processing module 200 is configured to determine a second identification network for the target domain signal according to the wireless signal data and the first identification network for the source domain signal, and the first identification network is obtained from the first signal data obtained from the source domain.
  • the data processing module 200 can determine the signal that can identify the target domain signal based on the wireless signal data and the first identification network for the source domain signal.
  • the second identification network thereby realizing the field migration of the identification signal, is conducive to improving the accuracy of signal identification, and, because the data processing module 200 can cooperate to realize signal identification through wireless signal data, that is, there is no need to perform signal identification based on a large amount of signal data in the area. Analysis and recognition, therefore, for the data processing module 200, resources can be saved and the difficulty of signal recognition can be reduced.
  • an embodiment of the present application also provides a signal recognition device, which includes: a memory, a processor, and a computer program stored in the memory and operable on the processor.
  • the processor and memory can be connected by a bus or other means.
  • the non-transitory software programs and instructions required to realize the signal recognition methods of the above-mentioned embodiments are stored in the memory, and when executed by the processor, the signal recognition methods of the above-mentioned embodiments are executed, for example, the above-described implementation of the above-described FIG. 1 Method steps S100 to S200, method steps S110 to S130 in FIG. 2 , method steps S131 in FIG. 3 , method steps S210 to S220 in FIG. 4 , method steps S221 to S222 in FIG. 5 , method steps in FIG. 6 S132 or method step S133 in FIG. 7 .
  • the device embodiments described above are only illustrative, and the units described as separate components may or may not be physically separated, that is, they may be located in one place, or may be distributed to multiple network units. Part or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.
  • an embodiment of the present application also provides a computer-readable storage medium, the computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are executed by a processor or a controller, for example, by the above-mentioned Execution by a processor in the device embodiment can cause the processor to execute the signal identification method in the above embodiment, for example, execute the method steps S100 to S200 in FIG. 1 and the method steps S110 to S130 in FIG. 2 described above. , method step S131 in FIG. 3 , method steps S210 to S220 in FIG. 4 , method steps S221 to S222 in FIG. 5 , method step S132 in FIG. 6 or method step S133 in FIG. 7 .
  • the embodiment of the present application includes: obtaining wireless signal data from the target domain; determining a second identification network for the target domain signal according to the wireless signal data and the first identification network for the source domain signal, and the first identification network is obtained from the first identification network obtained from the source domain.
  • a signal data is obtained.
  • the second identification network capable of identifying the target domain signal can be determined by cooperating with the wireless signal data obtained from the target domain, thereby realizing identification
  • the domain migration of signals is conducive to improving the accuracy of signal identification, and since signal identification can be realized through wireless signal data, there is no need to analyze and identify based on a large amount of signal data in the area, which can save resources and reduce the difficulty of signal identification.
  • Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disk (DVD) or other optical disk storage, magnetic cartridges, tape, magnetic disk storage or other magnetic storage devices, or can Any other medium used to store desired information and which can be accessed by a computer.
  • communication media typically embodies computer readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media .

Landscapes

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

Abstract

一种信号识别方法及装置、计算机可读存储介质，其中，信号识别方法包括：从目标域获取无线信号数据(S100)；根据无线信号数据和针对源域信号的第一识别网络确定针对目标域信号的第二识别网络，第一识别网络由从源域获得的第一信号数据得到(S200)。

Description

信号识别方法及装置、计算机可读存储介质
相关申请的交叉引用
本申请基于申请号为202110912337.X、申请日为2021年08月10日的中国专利申请提出，并要求该中国专利申请的优先权，该中国专利申请的全部内容在此引入本申请作为参考。
技术领域
本申请实施例涉及但不限于无线通信技术领域，尤其涉及一种信号识别方法及装置、计算机可读存储介质。
背景技术
无线通信技术，因其移动性、高效、低成本的部署与维护、无需高难度的布线工程与维护措施、高度灵活性的组网操作、扩展性佳以及提供简单的用户添加和删除操作而不会影响通信网络整体性能等优势，近年来得到高速发展。为了满足工业、科学和医学(Industrial Scientific Medical，ISM)的无线通信需求，设置了无需许可证或费用的频段，即ISM频段，具体产生了无线城域网、无线局域网等服务，使得越来越多的应用该频段的产品在智能家居、医疗健康和公共基础设施等各个领域有着更为广泛的应用。然而，随着共享ISM频段的各类无线设备的日益增长，多种样式的无线并发信号对无线局域网产生诸多干扰，制约了无线局域网的部署和使用效率，且各种设备之间的相互干扰还可能会对关键的数据通信造成损失。因此，如何确定干扰来源进而使得关键通信保持稳定，成为当前ISM频段无线通信亟待解决的问题。目前，干扰识别方法通常需要基于区域内的大量信号数据才能够进行分析识别，这会耗费大量资源，也并不简便，并且没有实际考虑到信号数据的采集环境不同所带来的领域偏移影响，因此识别精度不高。
发明内容
以下是对本文详细描述的主题的概述。本概述并非是为了限制权利要求的保护范围。
本申请实施例提供了一种信号识别方法及装置、计算机可读存储介质。
第一方面，本申请实施例提供了一种信号识别方法，包括：从目标域获取无线信号数据；根据所述无线信号数据和针对源域信号的第一识别网络确定针对目标域信号的第二识别网络，所述第一识别网络由从源域获得的第一信号数据得到。
第二方面，本申请实施例还提供了一种信号识别装置，包括：数据采集模块，被设置为从目标域获取无线信号数据；数据处理模块，被设置为根据所述无线信号数据和针对源域信号的第一识别网络确定针对目标域信号的第二识别网络，所述第一识别网络由从源域获得的第一信号数据得到。
第三方面，本申请实施例还提供了一种信号识别装置，包括：存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，所述处理器执行所述计算机程序时实现如上所述第一方面的信号识别方法。
第四方面，本申请实施例还提供了一种计算机可读存储介质，存储有计算机可执行指令，所述计算机可执行指令用于执行如上所述第一方面的信号识别方法。
本申请的其它特征和优点将在随后的说明书中阐述，并且，部分地从说明书中变得显而易见，或者通过实施本申请而了解。本申请的目的和其他优点可通过在说明书、权利要求书以及附图中所特别指出的结构来实现和获得。
附图说明
附图用来提供对本申请技术方案的进一步理解，并且构成说明书的一部分，与本申请的实施例一起用于解释本申请的技术方案，并不构成对本申请技术方案的限制。
图1是本申请一个实施例提供的信号识别方法的流程图；
图2是本申请一个实施例提供的信号识别方法中获取无线信号数据的流程图；
图3是本申请一个实施例提供的信号识别方法中确定无线信号数据的流程图；
图4是本申请一个实施例提供的信号识别方法中确定第二识别网络的流程图；
图5是本申请一个实施例提供的信号识别方法中更新第一识别网络的流程图；
图6是本申请一个实施例提供的信号识别方法中确定无线信号数据之后的流程图；
图7是本申请一个实施例提供的信号识别方法中确定待识别信号数据之后的流程图；
图8是本申请一个实施例提供的信号识别方法的执行示意图；
图9是本申请一个实施例提供的WADA算法的执行示意图；
图10是本申请一个实施例提供的信号识别装置的示意图；
图11是本申请另一个实施例提供的信号识别装置的示意图。
具体实施方式
为了使本申请的目的、技术方案及优点更加清楚明白，以下结合附图及实施例，对本申请进行进一步详细说明。应当理解，此处所描述的具体实施例仅用以解释本申请，并不用于限定本申请。
需要注意的是，虽然在装置示意图中进行了功能模块划分，在流程图中示出了逻辑顺序，但是在某些情况下，可以以不同于装置中的模块划分，或流程图中的顺序执行所示出或描述的步骤。说明书和权利要求书及上述附图中的术语“第一”、“第二”等是用于区别类似的对象，而不必用于描述特定的顺序或先后次序。
本申请提供了一种信号识别方法及装置、计算机可读存储介质，在能够识别源域信号的第一识别网络的基础上，配合从目标域获得的无线信号数据即可确定能够识别目标域信号的第二识别网络，从而实现识别信号的领域迁移，有利于提高信号识别精度，并且，由于通过无线信号数据即可配合实现信号识别，因此无需基于区域内的大量信号数据以进行分析识别，从而能够节省资源，降低信号识别难度。
下面结合附图，对本申请实施例作进一步阐述。
如图1所示，图1是本申请一个实施例提供的信号识别方法的流程图，该信号识别方法包括但不限于步骤S100和步骤S200。
步骤S100，从目标域获取无线信号数据。
在一实施例中，目标域与源域相区分，其中，源域表征已知环境或原始环境，通常来讲， 源域内的无线信号已经能够很好地被识别；目标域表征未知环境或新环境，相比于源域，目标域与源域之间存在领域之间的差异，相应地，这可能会带来信号识别之间的差异，因此需要对目标域的信号识别进行进一步地扩展。
在一实施例中，无线信号数据的种类可以有多种，针对不同类型的无线信号数据，可以后续选择适应的方式对其进行处理，例如，在以下各实施例中，主要以无线信号数据为WIFI信号进行说明，相应的数据处理方式也基于无线信号数据为WIFI信号而进行说明，但本领域技术人员可以根据实际应用场景进行相应设置，这在本实施例中并未限定。
在图2的示例中，步骤S100包括但不限于步骤S110至S130。
步骤S110，从目标域获取原始信号数据；
步骤S120，筛选原始信号数据，得到携带有效信号的第二信号数据；
步骤S130，从第二信号数据中确定无线信号数据。
在一实施例中，通过筛选原始信号数据可以剔除无效的第二信号数据，以便于进一步根据第二信号数据得到有效的无线信号数据，从而确保无线信号数据具有良好的实用效果，以便于提升后续处理无线信号数据的可靠性，进而提高信号识别精度。
在一实施例中，可以设置一无线接入点(Access Point,AP)用于获取原始信号数据，同时设置与AP所匹配连接的至少一个终端用于解析，即，通过AP将所获取的原始信号数据传送给终端，进而利用该终端对原始信号数据进行解析筛选，以便于提取出有效的第二信号数据作为信号样本。
需要说明的是，在通信系统中，为了消除噪声信号带来的影响，可以通过剔除原始信号数据中携带有噪声信号的相应数据，从而筛选得到不包含噪声信号的所有数据，在这种情况下，除噪声信号之外的信号即为有效信号，此时的有效信号可以有多个种类，包括传递用户所需信息的信号，或者，用于让接收设备收到信号后产生一个预先设定的动作的信号等，相应地，筛选得到的所有数据均为第二信号数据，但此处的“有效信号”并非属于限制性概念，可以根据实际的不同应用场景进行区分，不仅限于噪声信号的剔除，本领域技术人员可以自行设定区分形式，这在本实施例中并未限定。
在一实施例中，与AP匹配的终端可以设置为多个，各个终端均可以称为接入终端、用户设备(User Equipment，UE)、用户单元、用户站、移动站、移动台、远方站、远程终端、移动设备、用户终端、无线通信设备、用户代理或用户装置。例如，各个终端均可以是蜂窝电话、无绳电话、会话启动协议(Session Initiation Protocol，SIP)电话、无线本地环路(Wireless Local Loop，WLL)站、个人数字处理(Personal Digital Assistant，PDA)、具有无线通信功能的手持设备、计算设备或连接到无线调制解调器的其它处理设备、车载设备、可穿戴设备、5G网络或者未来5G以上网络中的终端设备等，本实施例对此并不作具体限定。
在图3的示例中，步骤S130包括但不限于步骤S131。
步骤S131，从第二信号数据中确定具有无线信号协议栈指示位的第二信号数据为无线信号数据。
在一实施例中，由于无线信号协议栈定义了通信硬件和软件针对于无线信号而在不同层次进行协调工作的特性，因此，当确定存在无线信号协议栈指示位，即可以确定无线信号协议栈已经被设定，说明无线信号相应存在，从而能够确定具有无线信号协议栈指示位的第二 信号数据为无线信号数据，因此，通过该方式可以简便可靠地筛选得到无线信号数据，能够提高信号分类处理数据，可以理解地是，当无线信号为WIFI信号，无线信号协议栈指示位为WIFI信号协议栈指示位，以此类推，不再赘述。
步骤S200，根据无线信号数据和针对源域信号的第一识别网络确定针对目标域信号的第二识别网络，第一识别网络由从源域获得的第一信号数据得到。
在一实施例中，源域作为已知环境，可以依据第一识别网络对相关信号进行准确地识别，这是由于第一识别网络正是由源域中的第一信号数据而得到的，以深度卷积神经网络为例，通过训练第一信号数据而得到训练合格的第一识别网络，而目标域作为新环境，由于受到信号的领域迁移的影响，因而仅凭借第一识别网络无法准确地识别相关信息，为了解决上述疑难，在能够识别源域信号的第一识别网络的基础上，配合从目标域获得的无线信号数据即可确定能够识别目标域信号的第二识别网络，从而实现识别信号的领域迁移，即，将信号识别域从原先的目标域扩展到当前的目标域，不仅能够识别原先的源域内的无线信号，还能够识别当前的目标域内的无线信号，因此大大提升了信号识别的区域范围，相比于已有技术，能够提高针对无线信号的识别精度，并且，由于通过无线信号数据即可配合实现信号识别，因此无需基于区域内的大量信号数据以进行分析识别，也无需增加新的网络架构或者参数来进行辅助识别，从而能够节省相关资源，进一步地降低信号识别难度。
在一实施例中，第一信号数据的类型可以有多种，例如，可以涉及目前常应用于ISM频段下的诸如蓝牙(Bluetooth)信号、无绳电话(Cordless Phone)信号、游戏手柄(Game Controller)信号、微波炉(Microwave Oven)信号、无限保真((Wireless Fidelity,WIFI)信号、无线模拟视频监控器(Video Monitor)信号以及ZigBee信号等，换言之，本实施例中的第一信号数据可以是以上各示例中的一种或多种，这在本实施例中并未限定。
需要说明的是，第一信号数据、第一识别网络以及第二识别网络可以根据实际应用场景自行选取或设置，为了简要清楚地说明本申请的工作原理，以下主要基于深度卷积神经网络的背景以进行说明，但并未唯一限制。
在图4的示例中，步骤S200包括但不限于步骤S210和步骤S220。
步骤S210，将无线信号数据和从源域获得的第一信号数据输入到特征提取网络，以使特征提取网络提取并输出对应于源域信号的第一样本特征和对应于目标域信号的第二样本特征；
步骤S220，根据第一样本特征和第二样本特征更新针对于源域信号的第一识别网络，确定更新后的第一识别网络为针对于目标域信号的第二识别网络。
在一实施例中，由于特征提取网络能够解析无线信号数据和从源域获得的第一信号数据，从而能够输出对应于源域信号的第一样本特征和对应于目标域信号的第二样本特征，即，能够分别获知源域和目标域的信号状态，从而根据源域和目标域的信号状态以对已经处理好的第一识别网络进行更新，相当于在第一样本特征的基础上进一步考虑第二样本特征的影响，从而得到更新后的第一识别网络，即确定第二识别网络，使得第二识别网络能够针对目标域的信号进行识别。
在图5的示例中，步骤S220中的“根据第一样本特征和第二样本特征更新针对于源域信号的第一识别网络”包括但不限于步骤S221和步骤S222。
步骤S221，在第一样本特征和第二样本特征互相映射于源域与目标域之间的情况下，根据映射条件确定第一损失函数，映射条件用于表征互相映射前后的第一样本特征和第二样本 特征对应的数据类别均不变；
步骤S222，根据第一损失函数和预设的第二损失函数更新针对于源域信号的第一识别网络。
在一实施例中，第一样本特征和第二样本特征可以为多个且数量可以保持一致，并且各个第一样本特征与各个第二样本特征可以实现彼此对应，有利于第一样本特征与第二样本特征彼此之间实现互相映射。
在一实施例中，映射条件表征循环一致性(Cycle Consistency,CC)，表明在第一样本特征和第二样本特征互相映射于源域与目标域之间的情况下，无论第一样本特征向第二样本特征进行映射几次，或者，第二样本特征向第一样本特征进行映射几次，映射前后的第一样本特征或者第二样本特征的类别不会发生改变，例如，第一样本特征的类别为蓝牙信号对应的类别，那么经过映射之后的第一样本特征的类别仍为蓝牙信号对应的类别，或者，第一样本特征的类别服从均匀分布，则经过映射变换后的数据类别同样服从均匀分布，通过映射条件以确定第一损失函数，并且基于第一损失函数和预设的第二损失函数更新第一识别网络，使得第一样本特征和第二样本特征在经过数据处理之后的变化差异不会过大，可以减小数据处理的误差，提高源域与目标域之间的数据关联性。
在一实施例中，由于仅依赖循环一致性控制两个域间的关联存在不足，且不同特征间的关联方式并不唯一，因此上述第一损失函数的最小化可以通过仅关联目标域的简单样本数据实现，即，可能仅挖掘到两个域间的一小部分共性特征，而不能够对更一般性的相似性特征进行表征，因此引入预设的第二损失函数，可以更好地表征领域迁移的特征，从而达到减小误差的效果，有利于进一步地提高信号识别精度。
需要说明的是，“互相映射”指的是第一样本特征和第二样本特征在源域与目标域之间的来回映射，或者叫做循环映射，例如，一个第一样本特征可以连续地从源域映射到目标域中的某个第二样本特征，然后该第二样本特征再从目标域映射回到源域的另一个第一样本特征，即相当于第一样本特征和第二样本特征在源域与目标域之间完成了一次基本的互相映射流程，上述互相映射流程也可以执行多次，但无论哪种情况均可以叫做互相映射，类似地，一个第二样本特征可以连续地从目标域映射到源域中的某个第一样本特征，然后该第一样本特征再从源域映射回到目标域的另一个第二样本特征，其原理与上述说明的互相映射流程的原理类似，在此不做赘述。
第二损失函数可以包括但不限于如下类型中的至少一种：
与源域信号和目标域信号关联的完整性损失函数；或
与第一信号数据关联的第一标签损失函数；或
与无线信号数据关联的第二标签损失函数。
在一实施例中，完整性损失函数能够弥补第一损失函数可能存在的关联性不足的问题，完整性损失函数由目标域上的样本数据的均匀分布与从任何源域上的样本数据访问目标域上的样本数据的概率之间的交叉熵定义，当源域和目标域的数据类别分布存在差别时，针对完整性损失函数的权重可以适当降低。
在一实施例中，第一标签损失函数和第二标签损失函数表征域内关联，作为域间关联函数的补充，可以进一步提升针对领域迁移的损失调整。
需要说明的是，本申请实施例针对信号识别所利用的框架简便可靠，易于实现和处理， 适用于不同环境下的数据采集场景，利用有限的无线信号数据和第一信号数据通过相同的特征提取网络，达到构造两个域样本的特征向量的目的，同时通过域间转换构造源域样本特征和目标域样本特征的预测标签循环一致性，结合完整性正则要求和传统的有监督损失作为总的损失函数进行第一识别网络的更新，从而得到第二识别网络，该第二识别网络既能够继承源域对领域偏移的不变性，也能够实现对于目标域内的无线信号的高辨识性，提高源域的领域泛化性和在目标域下的信号测试性能。
在图6的示例中，步骤S131之后，还包括但不限于步骤S132。
步骤S132，从第二信号数据中确定除无线信号数据之外的第二信号数据为目标域中的待识别信号数据。
在一实施例中，由于无线信号数据主要用于进行数据处理，以使第二识别网络能够可靠地对信号进行识别，因此，针对于已经确定的无线信号数据，可以不用对其进行再次地识别，同时由于第二信号数据是从目标域中获得的未知数据，因此，可以将除无线信号数据之外的第二信号数据作为待识别信号数据，以便于对该待识别信号数据进行识别，从而验证从源域到目标域的领域迁移效果。
需要说明的是，也可以针对无线信号数据进行多次地识别，这在本实施例中并未限制。
在图7的示例中，步骤S132之后，还包括但不限于步骤S133。
步骤S133，输入待识别信号数据至第二识别网络，以使第二识别网络输出针对待识别信号数据的信号识别结果。
在一实施例中，以除无线信号数据之外的第二信号数据作为待识别信号数据，防止无线信号数据影响识别，并且通过第二识别网络对该待识别信号数据进行识别，可以良好地检测第二识别网络的构建效果，进一步验证从源域到目标域的领域迁移效果。
下面给出相关实施例进行说明。
在一些实施例中，参照图8，给出了实现信号识别的执行流程示意图。
首先，基于第一AP获取原始信号数据；然后，由第一终端提取原始信号数据中的有效部分，从而得到第二信号数据；然后，对第二信号数据进行检测，即，针对第二信号数据提取WIFI信号协议栈指示位,从而区分得到WIFI信号样本和目标域中其余干扰信号样本；然后，将WIFI信号样本和源域中的第一信号数据共同作为输入，传递给WIFI信号激励关联领域自适应校正算法(以下记为“WADA算法”)，实现对第一识别网络进行目标域环境下的适应性校正，得到第二识别网络，其中，WADA算法为迁移学习中的领域自适应(Domain Adaptation)算法的一种，旨在利用源域数据和目标域数据的相关性得到对域偏移(domain shift)不敏感的域不变表征，为目标任务获取丰富的信息，在得到源域数据和目标域数据尽可能相似的特征的同时，降低两个域中数据类别的预测误差。
最后，将目标域中其余干扰信号样本输入至第二识别网络中，输出目标域内的信号识别结果。
根据实验，在各种目标域环境下，通过校正后的识别网络对目标域的各类干扰信号的平均准确率均超过90％，同时第二识别网络能够保持对源域的各种干扰信号的平均准确率同样保持在90％以上，能够满足信号识别的精度要求。
在一些实施例中，参照图9，给出了WADA算法的原理示意图。
如图9所示，该WADA算法对应的网络结构，包括特征提取网络卷积神经架构和类别预测 网络全连接架构，其基本原理如下：
首先，输入目标域的WIFI信号样本和源域的第一信号数据至特征提取网络卷积神经架构，使其输出源域和目标域的样本特征；
然后，针对源域和目标域的样本特征进行关联映射：
已知源域数据样本
Figure PCTCN2022104966-appb-000001
和目标域数据样本
Figure PCTCN2022104966-appb-000002
对其中的单个样本
Figure PCTCN2022104966-appb-000003
通过源域数据训练的特征提取网络
Figure PCTCN2022104966-appb-000004
分别得到其对应的低维空间特征
Figure PCTCN2022104966-appb-000005
和
Figure PCTCN2022104966-appb-000006
此时，来自不同域的两个样本的相似度可以用A i和B i的内积M i,j＝<A i,B i>来表示，其中，N s表示源域中包含
Figure PCTCN2022104966-appb-000007
类别在内的样本集合，N t表示目标域中包含
Figure PCTCN2022104966-appb-000008
类别在内的样本集合。在一示例中，源域样本
Figure PCTCN2022104966-appb-000009
到目标域样本
Figure PCTCN2022104966-appb-000010
的关联概率
Figure PCTCN2022104966-appb-000011
可以由以下的类softmax模型公式定义：
Figure PCTCN2022104966-appb-000012
相似地，将上式中的M ij替换为
Figure PCTCN2022104966-appb-000013
可以得到从目标域样本关联回源域样本的概率，即
Figure PCTCN2022104966-appb-000014
最终通过一轮循环，特征从源域映射到目标域，再映射回源域的概率可以写为：
Figure PCTCN2022104966-appb-000015
为了保证该循环有效，其需要满足一定程度的循环一致性，即起始数据和循环后的数据的类别要相同，换句话说，通过循环映射后源域的数据类别不发生改变。假设源域数据类别服从均匀分布，则经过循环变换后的数据类别同样应该服从均匀分布。因此，该循环一致性可以通过下述损失函数定义完成：
Figure PCTCN2022104966-appb-000016
其中，H(T,P aba)表示交叉熵；
Figure PCTCN2022104966-appb-000017
表示源域中和A i类别相同的样本集合；
class()表示样本的类别。
在一示例中，仅仅依赖循环一致性控制两个域间的关联有所不足，由于样本间的关联方式并不唯一，上述损失函数的最小化可以通过仅关联目标域的简单样本实现，即仅挖掘到两个域间的一小部分共性特征，而不能够对更一般性的相似性进行描述。此时，额外引入完整性损失函数，即：
Figure PCTCN2022104966-appb-000018
其由目标样本上的均匀分布与从任何源域样本开始访问目标域样本的概率之间的交叉熵而定义，当源域数据和目标域数据的类别分布存在差别时，完整性损失函数L v在整体损失函数中的权重可以适当降低。
除了上述的域间关联损失函数外，同样需要在有标签的源域数据上达到优秀的分类性能。因此，整体损失函数中还包括源域的有标签交叉熵损失L cla1与目标域的WIFI数据有标签损失L cla2。
综上所述，涉及领域迁移的整体损失函数L可以写为：
L＝L cla1+L cla2+α 1L w+α 2L v
其中，α 1,α 2∈[0,1]为权重参数。
经过以上损失函数训练后，针对源域的第一识别网络便通过WIFI信号样本的校正变为能够适应当前目标域环境数据的第二识别网络，即，如图9所示的类别预测网络全连接架构，其中，类别预测网络全连接架构为卷积网络模型，包括数个依次连接的卷积层、批归一化层和池化层，之后通过全局平均池化后，通过数个全连接层，最终输出针对当前目标域环境数据的类别预测结果。
另外，如图10所示，本申请的一个实施例还提供了一种信号识别装置，该装置包括：
数据采集模块100，被设置为从目标域获取无线信号数据；
数据处理模块200，被设置为根据无线信号数据和针对源域信号的第一识别网络确定针对目标域信号的第二识别网络，第一识别网络由从源域获得的第一信号数据得到。
在一实施例中，在数据采集模块100从目标域获取无线信号数据的前提下，数据处理模块200基于无线信号数据配合针对源域信号的第一识别网络，即可确定能够识别目标域信号的第二识别网络，从而实现识别信号的领域迁移，有利于提高信号识别精度，并且，由于数据处理模块200通过无线信号数据即可配合实现信号识别，即，无需基于区域内的大量信号数据以进行分析识别，因此对于数据处理模块200而言，能够节省资源，降低信号识别难度。
此外，参照图11，本申请的一个实施例还提供了一种信号识别装置，该信号识别装置包括：存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序。
处理器和存储器可以通过总线或者其他方式连接。
实现上述实施例的信号识别方法所需的非暂态软件程序以及指令存储在存储器中，当被处理器执行时，执行上述各实施例的信号识别方法，例如，执行以上描述的图1中的方法步骤S100至S200、图2中的方法步骤S110至S130、图3中的方法步骤S131、图4中的方法步骤S210至S220、图5中的方法步骤S221至S222、图6中的方法步骤S132或图7中的方法步骤S133。
以上所描述的装置实施例仅仅是示意性的，其中作为分离部件说明的单元可以是或者也可以不是物理上分开的，即可以位于一个地方，或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部模块来实现本实施例方案的目的。
此外，本申请的一个实施例还提供了一种计算机可读存储介质，该计算机可读存储介质存储有计算机可执行指令，该计算机可执行指令被一个处理器或控制器执行，例如，被上述设备实施例中的一个处理器执行，可使得上述处理器执行上述实施例中的信号识别方法，例如，执行以上描述的图1中的方法步骤S100至S200、图2中的方法步骤S110至S130、图3中的方法步骤S131、图4中的方法步骤S210至S220、图5中的方法步骤S221至S222、图6中的方法步骤S132或图7中的方法步骤S133。
本申请实施例包括：从目标域获取无线信号数据；根据无线信号数据和针对源域信号的第一识别网络确定针对目标域信号的第二识别网络，第一识别网络由从源域获得的第一信号数据得到。根据本申请实施例提供的方案，在能够识别源域信号的第一识别网络的基础上，配合从目标域获得的无线信号数据即可确定能够识别目标域信号的第二识别网络，从而实现识别信号的领域迁移，有利于提高信号识别精度，并且，由于通过无线信号数据即可配合实现信号识别，因此无需基于区域内的大量信号数据以进行分析识别，从而能够节省资源，降 低信号识别难度。
本领域普通技术人员可以理解，上文中所公开方法中的全部或某些步骤、系统可以被实施为软件、固件、硬件及其适当的组合。某些物理组件或所有物理组件可以被实施为由处理器，如中央处理器、数字信号处理器或微处理器执行的软件，或者被实施为硬件，或者被实施为集成电路，如专用集成电路。这样的软件可以分布在计算机可读介质上，计算机可读介质可以包括计算机存储介质(或非暂时性介质)和通信介质(或暂时性介质)。如本领域普通技术人员公知的，术语计算机存储介质包括在用于存储信息(诸如计算机可读指令、数据结构、程序模块或其他数据)的任何方法或技术中实施的易失性和非易失性、可移除和不可移除介质。计算机存储介质包括但不限于RAM、ROM、EEPROM、闪存或其他存储器技术、CD-ROM、数字多功能盘(DVD)或其他光盘存储、磁盒、磁带、磁盘存储或其他磁存储装置、或者可以用于存储期望的信息并且可以被计算机访问的任何其他的介质。此外，本领域普通技术人员公知的是，通信介质通常包含计算机可读指令、数据结构、程序模块或者诸如载波或其他传输机制之类的调制数据信号中的其他数据，并且可包括任何信息递送介质。
以上是对本申请的若干实施方式进行的具体说明，但本申请并不局限于上述实施方式，熟悉本领域的技术人员在不违背本申请精神的前提下还可作出种种的等同变形或替换，这些等同的变形或替换均包含在本申请权利要求所限定的范围内。

Claims (11)

  1. 一种信号识别方法，包括：
    从目标域获取无线信号数据；
    根据所述无线信号数据和针对源域信号的第一识别网络确定针对目标域信号的第二识别网络，所述第一识别网络由从源域获得的第一信号数据得到。
  2. 根据权利要求1所述的信号识别方法，其中，所述根据所述无线信号数据和针对源域信号的第一识别网络确定针对目标域信号的第二识别网络，包括：
    将所述无线信号数据和从源域获得的第一信号数据输入到特征提取网络，以使所述特征提取网络提取并输出对应于源域信号的第一样本特征和对应于目标域信号的第二样本特征；
    根据所述第一样本特征和所述第二样本特征更新针对于所述源域信号的第一识别网络，确定更新后的所述第一识别网络为针对于所述目标域信号的第二识别网络。
  3. 根据权利要求2所述的信号识别方法，其中，所述根据所述第一样本特征和所述第二样本特征更新针对于所述源域信号的第一识别网络，包括：
    在所述第一样本特征和所述第二样本特征互相映射于所述源域与所述目标域之间的情况下，根据映射条件确定第一损失函数，所述映射条件用于表征互相映射前后的所述第一样本特征和所述第二样本特征对应的数据类别均不变；
    根据所述第一损失函数和预设的第二损失函数更新针对于所述源域信号的第一识别网络。
  4. 根据权利要求3所述的信号识别方法，其中，所述第二损失函数包括如下类型中的至少一种：
    与所述源域信号和所述目标域信号关联的完整性损失函数；或
    与所述第一信号数据关联的第一标签损失函数；或
    与所述无线信号数据关联的第二标签损失函数。
  5. 根据权利要求1所述的信号识别方法，其中，所述从目标域获取无线信号数据，包括：
    从目标域获取原始信号数据；
    筛选所述原始信号数据，得到携带有效信号的第二信号数据；
    从所述第二信号数据中确定无线信号数据。
  6. 根据权利要求5所述的信号识别方法，其中，所述从所述第二信号数据中确定无线信号数据，包括:
    从所述第二信号数据中确定具有无线信号协议栈指示位的所述第二信号数据为无线信号数据。
  7. 根据权利要求5或6所述的信号识别方法，其中，所述从所述第二信号数据中确定无线信号数据之后，还包括：
    从所述第二信号数据中确定除所述无线信号数据之外的所述第二信号数据为所述目标域中的待识别信号数据。
  8. 根据权利要求7所述的信号识别方法，其中，所述从所述第二信号数据中确定除所述无线信号数据之外的所述第二信号数据为所述目标域中的待识别信号数据之后，还包括：
    输入所述待识别信号数据至所述第二识别网络，以使所述第二识别网络输出针对所述待识别信号数据的信号识别结果。
  9. 一种信号识别装置，包括：
    数据采集模块，被设置为从目标域获取无线信号数据；
    数据处理模块，被设置为根据所述无线信号数据和针对源域信号的第一识别网络确定针对目标域信号的第二识别网络，所述第一识别网络由从源域获得的第一信号数据得到。
  10. 一种信号识别装置，包括：存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，其中，所述处理器执行所述计算机程序时实现如权利要求1至8中任意一项所述的信号识别方法。
  11. 一种计算机可读存储介质，存储有计算机可执行指令，所述计算机可执行指令用于执行权利要求1至8中任意一项所述的信号识别方法。
PCT/CN2022/104966 2021-08-10 2022-07-11 信号识别方法及装置、计算机可读存储介质 Ceased WO2023016168A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202110912337.XA CN115705395A (zh) 2021-08-10 2021-08-10 信号识别方法及装置、计算机可读存储介质
CN202110912337.X 2021-08-10

Publications (1)

Publication Number Publication Date
WO2023016168A1 true WO2023016168A1 (zh) 2023-02-16

Family

ID=85179473

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/104966 Ceased WO2023016168A1 (zh) 2021-08-10 2022-07-11 信号识别方法及装置、计算机可读存储介质

Country Status (2)

Country Link
CN (1) CN115705395A (zh)
WO (1) WO2023016168A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117409560A (zh) * 2023-09-12 2024-01-16 西安电子科技大学 基于无监督的无线电信号调制方式的识别方法及装置

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2018109505A1 (en) * 2016-12-15 2018-06-21 Google Llc Transforming source domain images into target domain images
CN108508411A (zh) * 2018-03-22 2018-09-07 天津大学 基于迁移学习的被动雷达外辐射源信号识别方法
WO2019204547A1 (en) * 2018-04-18 2019-10-24 Maneesh Kumar Singh Systems and methods for automatic speech recognition using domain adaptation techniques
CN111582449A (zh) * 2020-05-07 2020-08-25 广州视源电子科技股份有限公司 一种目标域检测网络的训练方法、装置、设备及存储介质

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2018109505A1 (en) * 2016-12-15 2018-06-21 Google Llc Transforming source domain images into target domain images
CN108508411A (zh) * 2018-03-22 2018-09-07 天津大学 基于迁移学习的被动雷达外辐射源信号识别方法
WO2019204547A1 (en) * 2018-04-18 2019-10-24 Maneesh Kumar Singh Systems and methods for automatic speech recognition using domain adaptation techniques
CN111582449A (zh) * 2020-05-07 2020-08-25 广州视源电子科技股份有限公司 一种目标域检测网络的训练方法、装置、设备及存储介质

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117409560A (zh) * 2023-09-12 2024-01-16 西安电子科技大学 基于无监督的无线电信号调制方式的识别方法及装置

Also Published As

Publication number Publication date
CN115705395A (zh) 2023-02-17

Similar Documents

Publication Publication Date Title
US9226262B2 (en) Location model updating apparatus and position estimating method
CN112651467B (zh) 卷积神经网络的训练方法和系统以及预测方法和系统
CN113902944B (zh) 模型的训练及场景识别方法、装置、设备及介质
CN118312108B (zh) 预测模型的跨域迁移方法、装置、设备和存储介质
CN114626335B (zh) 文字生成方法、网络训练方法、装置、设备及存储介质
WO2023016168A1 (zh) 信号识别方法及装置、计算机可读存储介质
CN113140012B (zh) 图像处理方法、装置、介质及电子设备
CN117351995B (zh) 基于深度学习复合模型的环境声音识别方法及装置
CN117373008A (zh) 一种车牌识别方法、装置、电子设备及存储介质
WO2022007527A1 (zh) 样本数据标注系统、方法以及相关设备
US12190237B2 (en) Pattern recognition device and learned model
CN114723074A (zh) 聚类联邦学习框架下的主动学习客户选择方法和装置
CN111539477B (zh) 水质监测管理方法、装置、服务器及可读存储介质
CN118521951B (zh) 公共安全信息识别模型的联邦学习方法、识别方法及装置
CN116438546A (zh) 模型蒸馏训练的方法及相关装置和设备、可读存储介质
CN114332529B (zh) 图像分类模型的训练方法、装置、电子设备及存储介质
CN115187809B (zh) 分类及分类模型训练方法、装置及存储介质
CN118097293A (zh) 基于残差图卷积网络和自注意力的小样本数据分类方法及系统
CN116390227A (zh) 基于无线信号的定位方法、装置、存储介质及计算机设备
CN116310966A (zh) 视频动作定位模型训练方法、视频动作定位方法和系统
CN116631431A (zh) 混叠信号识别方法、计算机设备和存储介质
CN116128073A (zh) 基于域增量学习的模型调优方法、装置、设备及存储介质
CN116932316A (zh) 系统参数信息确定方法、数据处理方法、装置及设备
CN113343093A (zh) 基于人工智能的智慧在线服务信息推荐方法及云计算系统
CN115829136A (zh) 一种水质预测模型的优化方法、装置及网络设备

Legal Events

Date Code Title Description
NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 22855151

Country of ref document: EP

Kind code of ref document: A1