Signal recognition model training method, signal recognition method and device

Abstract

The present application relates to the technical field of machine learning, and in particular to a signal recognition model training method, a signal recognition method and a device, wherein the signal recognition model training method comprises: performing information superposition processing on target data to obtain a plurality of processed data; the target data is unlabeled signal data corresponding to a target task; performing signal recognition on the plurality of processed data based on a source signal recognition model to obtain an identification label corresponding to the target data; the source signal recognition model is trained based on source signal data carrying a source label corresponding to the target task; the source label includes a pulsar signal label and a non-pulsar signal label; based on the source signal data carrying the source label and the target data carrying the identification label, the source signal recognition model is trained to obtain a target signal recognition model corresponding to the target task. The present application can improve the accuracy of pulsar signal recognition.

Description

Technical Field

[0001] The present application relates to the technical field of machine learning, and in particular to a signal recognition model training method, a signal recognition method and a device. Background Art

[0002] With the development of astrophysics and radio signal technology, more and more cosmic signals can be observed and recorded by radio telescopes deployed in various regions of the world. At the same time, with the improvement of sensitivity and observation range, the amount of signal data received by astronomical telescopes every day has also exploded. For this reason, auxiliary celestial body identification systems are designed to assist in the search for pulsars. The signal recognition model in the auxiliary celestial body identification system in the prior art is heavily dependent on labeled data, and the labeling cost for pulsar signals is extremely high, so the labeled data used for training is limited, resulting in the signal recognition model trained based on limited labeled data having the problem of data overfitting, which in turn leads to the low recognition accuracy of the signal recognition model. Summary of the invention

[0003] The technical problem to be solved by the present application is to provide a signal recognition model training method, a signal recognition method and a device, which can avoid the problem of data overfitting and improve the recognition accuracy of the signal recognition model.

[0004] In order to solve the above technical problems, on the one hand, the embodiment of the present application provides a signal recognition model training method, including:

[0005] Performing information superposition processing on the target data to obtain multiple processed data; the target data is unlabeled signal data corresponding to the target task;

[0006] Based on a source signal recognition model, signal recognition is performed on the plurality of processed data to obtain an identification label corresponding to the target data; the source signal recognition model is trained based on source signal data carrying source labels corresponding to the target task; the source labels include pulsar signal labels and non-pulsar signal labels;

[0007] Based on the source signal data carrying the source label and the target data carrying the identification label, the source signal identification model is trained to obtain a target signal identification model corresponding to the target task.

[0008] On the other hand, the embodiment of the present application further provides a signal recognition method, including:

[0009] Acquire signal data to be identified;

[0010] Inputting the signal data to be identified into a target signal identification model for signal identification, and obtaining an identification result of the signal data to be identified;

[0011] The target signal recognition model is obtained by training the source signal recognition model based on source signal data carrying source labels and target data carrying identification labels; the identification label is obtained by performing signal recognition on the target data after information superposition processing based on the source signal recognition model; the target data is unlabeled signal data corresponding to the target task; the source signal recognition model is obtained by training the source signal data carrying source labels corresponding to the target task; the source labels include pulsar signal labels and non-pulsar signal labels.

[0012] On the other hand, an embodiment of the present application provides a signal recognition model training device, including:

[0013] An information superposition processing module is used to perform information superposition processing on target data to obtain multiple processed data; the target data is unlabeled signal data corresponding to the target task;

[0014] An identification label determination module is used to perform signal identification on the plurality of processed data based on a source signal identification model to obtain an identification label corresponding to the target data; the source signal identification model is trained based on source signal data carrying source labels corresponding to the target task; the source labels include pulsar signal labels and non-pulsar signal labels;

[0015] The model training module is used to train the source signal recognition model based on the source signal data carrying the source label and the target data carrying the recognition label to obtain a target signal recognition model corresponding to the target task.

[0016] On the other hand, an embodiment of the present application provides a signal recognition device, including:

[0017] A signal data acquisition module to be identified, used to acquire signal data to be identified;

[0018] A signal recognition module, used for inputting the signal data to be recognized into a target signal recognition model for signal recognition, and obtaining a recognition result of the signal data to be recognized;

[0019] The target signal recognition model is obtained by training the source signal recognition model based on source signal data carrying source labels and target data carrying identification labels; the identification label is obtained by performing signal recognition on the target data after information superposition processing based on the source signal recognition model; the target data is unlabeled signal data corresponding to the target task; the source signal recognition model is obtained by training the source signal data carrying source labels corresponding to the target task.

[0020] On the other hand, the present application provides an electronic device, comprising a processor and a memory, wherein the memory stores at least one instruction or at least one program, and the at least one instruction or the at least one program is loaded and executed by the processor to implement the signal recognition model training method or signal recognition method as described above.

[0021] On the other hand, the present application provides a computer storage medium, which stores at least one instruction or at least one program, and the at least one instruction or the at least one program is loaded by a processor and executed by the signal recognition model training method or signal recognition method as described above.

[0022] Implementing the embodiments of the present application has the following beneficial effects:

[0023] The present application obtains multiple processed data by performing information superposition processing on the target data of the unlabeled signal corresponding to the target task; performs signal recognition on the multiple processed data based on the source signal recognition model to obtain the identification label corresponding to the target data; thereby obtaining the identification label of the unlabeled signal data through the label recognition results of the multiple processed data corresponding to the unlabeled signal data, thereby improving the accuracy of the identification label prediction; the source signal recognition model is obtained by training based on the source signal data carrying source labels, and the source labels include pulsar signal labels and non-pulsar signal labels; the source signal recognition model is trained based on the source signal data carrying source labels and the target data carrying identification labels to obtain the target signal recognition model corresponding to the target task; that is, by performing label prediction on the unlabeled signal data to expand the signal recognition model training samples, it is possible to avoid the problem of data overfitting in the signal recognition model used for pulsar signal recognition that is obtained by training based on limited labeled data, thereby improving the recognition ability of the signal recognition model, and thus improving the accuracy of pulsar signal recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present application or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0025] Figure 1 It is a schematic diagram of the implementation environment provided by the embodiment of the present application;

[0026] Figure 2 It is a flow chart of a signal recognition model training method provided in an embodiment of the present application;

[0027] Figure 3This is a flow chart of an information superposition processing method provided in an embodiment of the present application;

[0028] Figure 4 This is a flow chart of a method for determining an identification label corresponding to target data through label fusion provided by an embodiment of the present application;

[0029] Figure 5 This is a flow chart of a method for generating a target signal recognition model provided by an embodiment of the present application;

[0030] Figure 6 is a flow chart of a method for determining an original sequence provided in an embodiment of the present application;

[0031] Figure 7 It is a flow chart of a method for generating a mixed sequence based on weighting provided in an embodiment of the present application;

[0032] Figure 8 is a flow chart of a signal recognition method provided by an embodiment of the present application;

[0033] Fig. 9 It is a schematic diagram of a signal recognition model training device provided in an embodiment of the present application;

[0034] Fig.10 is a schematic diagram of a signal recognition device provided in an embodiment of the present application;

[0035] Fig.11 It is a schematic diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0036] To make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present application.

[0037] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar signals, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0038] See also Figure 1 , which shows a schematic diagram of the implementation environment provided by an embodiment of the present application, the implementation environment may include: at least one signal identification request end 110 and a signal identification processing end 120, and the signal identification request end 110 and the signal identification processing end 120 can communicate data through a network.

[0039] Specifically, the signal identification request end 110 can send a signal identification request to the signal identification processing end 120 according to implementation requirements, and the signal identification request includes the signal data to be identified of the signal to be identified; after receiving the signal identification request, the signal identification processing end 120 calls the corresponding signal identification model to identify the signal data to be identified, obtains the identification result of the signal data to be identified, and returns the identification result to the signal identification request end 110.

[0040] The signal identification request end 110 can communicate with the signal identification processing end 120 based on the browser / server mode (B / S) or the client / server mode (C / S). The signal identification request end 110 may include: physical devices such as smart phones, personal computers, digital assistants, smart wearable devices, and vehicle terminals, and may also include software running in physical devices, such as applications. The operating system running on the signal identification request end 110 in the embodiment of the present application may include but is not limited to Android system, IOS system, Linux, Windows, etc.

[0041] The signal identification processing end 120 can establish a communication connection with the signal identification request end 110 via wired or wireless communication. The signal identification processing end 120 can include an independently operated server, or a distributed server, or a server cluster composed of multiple servers, wherein the server can be a cloud server.

[0042] In order to solve the problem of data overfitting in the signal recognition model obtained by training based on limited labeled data in the prior art, the embodiment of the present application provides a signal recognition model training method, and the execution subject of the method can be the above-mentioned signal recognition processing end; please refer to Figure 2 , the method may include:

[0043] S210. Perform information superposition processing on the target data to obtain multiple processed data; the target data is unlabeled signal data corresponding to the target task.

[0044] In this embodiment, the signal data corresponding to any signal can be used to characterize the signal characteristic information of the signal, and the signal characteristic information may include time domain characteristic information and frequency domain characteristic information. The target data can specifically be unlabeled signal data corresponding to any type of signal, so that the target data can be used to characterize the signal characteristic information of the unlabeled signal.

[0045] Specifically, the time domain characteristic information of the signal may be a phase-time diagram or a phase-frequency diagram of the signal, so that the signal data of the signal may be image data corresponding to the phase-time diagram and / or image data corresponding to the phase-frequency diagram.

[0046] The target data may include one data sub-item or multiple data sub-items, so that each data sub-item may be subjected to multiple information superposition processes to obtain multiple processed data corresponding to each data sub-item. In the process of superposition of information on the same data sub-item, the information superimposed each time is generally different, so that the multiple processed data obtained after the superposition process are generally different. The superimposed information may be information related to the current data sub-item, or information unrelated to the current data sub-item.

[0047] The target task here may specifically be a task of training a signal recognition model, or the target task may be a task of recognizing a signal, so that during the signal recognition model training process or the signal recognition process, the required signal data may be obtained.

[0048] S220. Perform signal recognition on the multiple processed data based on a source signal recognition model to obtain an identification label corresponding to the target data; the source signal recognition model is trained based on the source signal data carrying source labels corresponding to the target task; the source labels include pulsar signal labels and non-pulsar signal labels.

[0049] The source signal recognition model is trained based on source signal data carrying source labels corresponding to the target task. In this embodiment, for the pulsar signal recognition scenario, the corresponding source labels may include pulsar signal labels and non-pulsar signal labels.

[0050] The source signal recognition model already has a certain signal recognition capability, so it can perform signal recognition on the above-mentioned multiple processed data based on the source signal recognition model to obtain the predicted label corresponding to each processed data; then, based on the predicted labels corresponding to the multiple processed data, determine the identification label corresponding to the target data, that is, determine the identification label corresponding to the target data through the label recognition results of multiple processed data.

[0051] For source signal data carrying source labels, the source signal data can be expressed as x s , the corresponding source label can be expressed as y s ,y s Specifically, it can be represented by 0 or 1, 0 represents non-pulsar signal, and 1 represents pulsar signal.

[0052] S230. Based on the source signal data carrying the source label and the target data carrying the identification label, the source signal recognition model is trained to obtain a target signal recognition model corresponding to the target task.

[0053] After determining the identification label corresponding to the target data, the source signal recognition model can be trained based on the source signal data carrying the source label and the target data carrying the identification label. Specifically, the source signal data carrying the source label and the target data carrying the identification label can be fused, and then the source signal recognition model can be further trained based on the fused labeled data to obtain the target signal recognition model.

[0054] In the process of data fusion processing of source signal data carrying source labels and target data carrying identification labels, the source signal data carrying source labels can be fused with the full amount of target data carrying identification labels, thereby increasing the number of training samples.

[0055] In the process of data fusion processing of source signal data carrying source labels and target data carrying identification labels, the source signal data carrying source labels and part of the target data carrying identification labels can be used for data fusion. Since the identification label of the target data is predicted and not necessarily the actual label, the data fusion method can reduce the proportion of target data carrying identification labels in the total training sample data volume, thereby improving the accuracy of the training sample data.

[0056] The present application obtains multiple processed data by performing information superposition processing on the target data of the unlabeled signal corresponding to the target task; performs signal recognition on the multiple processed data based on the source signal recognition model to obtain the identification label corresponding to the target data; thereby obtaining the identification label of the unlabeled signal data through the label recognition results of the multiple processed data corresponding to the unlabeled signal data, thereby improving the accuracy of the identification label prediction; the source signal recognition model is obtained by training based on the source signal data carrying source labels, and the source labels include pulsar signal labels and non-pulsar signal labels; the source signal recognition model is trained based on the source signal data carrying source labels and the target data carrying identification labels to obtain the target signal recognition model corresponding to the target task; that is, by performing label prediction on the unlabeled signal data to expand the signal recognition model training samples, it is possible to avoid the problem of data overfitting in the signal recognition model used for pulsar signal recognition that is obtained by training based on limited labeled data, thereby improving the recognition ability of the signal recognition model, and thus improving the accuracy of pulsar signal recognition.

[0057] For further information, see Figure 3 , which shows an information superposition processing method, which may include:

[0058] S310. When performing information superposition processing on each data sub-item, multiple pieces of random information are generated.

[0059] S320. Superimpose the multiple random information onto each data sub-item respectively to obtain multiple processed data corresponding to each data sub-item; the data features of each data sub-item are the same as the data features of the multiple processed data.

[0060] For each data sub-item, multiple random information may be generated accordingly; the generated multiple random information are respectively superimposed on the data sub-items to obtain multiple processed data.

[0061] Furthermore, for the generation of random information, corresponding superimposed noise can be generated based on the noise consistency principle. The superimposed noise can specifically be a priori probability noise such as Laplace noise, Gaussian noise, Poisson noise, etc.

[0062] For unlabeled target data, it can be expressed as x t , the number of items of processed data can be two or more, and the following explanation is based on the number of items of processed data. t The random information ε1 and ε2 are superimposed on each other to obtain two processed data: x t,1 =x t +ε1,x t,2 =x t +ε2.

[0063] In this embodiment, the superposition of random information does not affect the data characteristics of the data sub-items, that is, the data characteristics of each data sub-item are the same as the data characteristics of the corresponding multiple processed data; that is, the superposition of random information can achieve the diversity of data sub-items on the basis of the original data sub-items, but will not change the data characteristics of the original data sub-items.

[0064] See also Figure 4 , which shows a method for determining an identification label corresponding to target data by label fusion, the method may include:

[0065] S410. Input the plurality of processed data into the source signal recognition model to obtain predicted sub-labels corresponding to the plurality of processed data.

[0066] S420. Perform information fusion on the predicted sub-labels corresponding to the multiple processed data to obtain the identification label.

[0067] The source signal recognition model already has the ability to perform signal recognition, so the source signal recognition model can be used to perform signal recognition on multiple processed data. Specifically, multiple processed data can be input into the source signal recognition model S to obtain corresponding predicted sub-labels. For example, for the above x t,1 , x t,2 The predicted sub-label of can be S(x t,1 ), S(x t,2 ).

[0068] Information fusion is performed on multiple prediction sub-labels, which can be specifically weighted processing of multiple prediction sub-labels, or averaging of multiple prediction sub-labels, etc.; when weighted processing is performed on multiple prediction sub-labels, the weight corresponding to each prediction sub-label can be determined based on a random number, that is, multiple random numbers are generated, and the sum of these multiple random numbers is 1.

[0069] Taking the average of multiple predicted sub-labels to determine the identification label as an example, for the above predicted sub-label S(x t,1 ), S(x t,2 ), the corresponding identification tag can be:

[0070]

[0071] Thus, by fusing multiple predicted sub-labels corresponding to the same data sub-item, an identification label corresponding to the data sub-item is obtained. The identification label is generated based on the fusion of multiple information, which makes the estimation of the identification label more accurate and improves the accuracy of the identification label.

[0072] See also Figure 5 , which shows a method for generating a target signal recognition model, which may include:

[0073] S510. Based on the source signal data carrying the source label and the target data carrying the identification label, determine the original data sequence and the original label sequence; the position of each data sub-item in the original data sequence is the same as the position of the label corresponding to each data sub-item in the original label sequence.

[0074] S520. Rearrange the original data sequence and the original label sequence to obtain a rearranged data sequence and a rearranged label sequence; the position of each data sub-item in the rearranged data sequence is the same as the position of the label corresponding to each data sub-item in the rearranged label sequence.

[0075] S530. Weight the data sub-items at corresponding positions in the original data sequence and the rearranged data sequence to obtain a mixed data sequence, and weight the labels at corresponding positions in the original label sequence and the rearranged label sequence to obtain a mixed label sequence; the weighting method for weighting the data sub-items at the target position is the same as the weighting method for weighting the labels at the target position.

[0076] S540. Train the source signal recognition model based on the mixed data sequence and the mixed label sequence to obtain the target signal recognition model.

[0077] Based on the source signal data and the target data, the corresponding original data sequence can be determined; based on the source label and the identification label, the corresponding original label sequence can be determined. Since the source signal data corresponds to the source label and the target data corresponds to the identification label, in order to facilitate the correspondence between data sub-items and labels, the position of each data sub-item in the original data sequence is the same as the position of the label corresponding to each data sub-item in the original label sequence, thereby realizing the correspondence between data sub-items and labels.

[0078] Each element in the sequence has an arrangement order, so in order to construct a sequence that is inconsistent with the order of the original data sequence and the original label sequence, the original data sequence can be reordered to obtain a reordered data sequence, and the original label sequence can be reordered to obtain a reordered label sequence. In this embodiment, when reordering the original data sequence and the original label sequence, the same reordering method can be used, that is, the original data sequence and the original label sequence are reordered in the same order, so that the position of each data sub-item in the reordered data sequence can be the same as the position of the label corresponding to each data sub-item in the reordered label sequence. For example, the position of data sub-item a1 in the original data sequence is 1, and the position of the corresponding label b1 in the original label sequence is also 1. If after reordering, the position of data sub-item a1 in the reordered data sequence is 5, the position of the corresponding label b1 in the reordered label sequence is also 5.

[0079] When reordering a sequence, a random reordering method or a non-random reordering method may be used. For the random reordering method, a corresponding reordering algorithm may be called, for example:

[0080] M′,N′=Permute(M,N) (2)

[0081] M' is a sequence obtained by randomly reordering M, and N' is a sequence obtained by randomly reordering N, and the order of the random reordering of M and N is the same.

[0082] By using the random reordering method, the data of the reordered sequence can be relatively dispersed compared to the data of the original sequence, so that when the data is subsequently weighted, the diversity of the weighted data can be improved.

[0083] For the non-random rearrangement method, position mapping can be performed based on the position identifier of the element in the sequence, and the element can be mapped to a new position; specifically, based on the preset mapping method f(i)=j, where i is the position identifier of the element in the original sequence, j is obtained by the preset mapping method, and j is the position identifier of the element in the rearranged sequence, thereby realizing the mapping of the position identifier. The original data sequence and the original label sequence are reordered respectively by the non-random rearrangement method, and the position of each data sub-item in the rearranged data sequence is the same as the position of the label corresponding to each data sub-item in the rearranged label sequence.

[0084] Through the non-random rearrangement method, elements at any position can be mapped to a certain position, so as to facilitate the verification of the rearrangement sequence after the random rearrangement to confirm whether the rearrangement is performed according to the preset rearrangement method.

[0085] Therefore, by weighting the original data sequence and the rearranged data sequence, as well as the original label sequence and the rearranged data sequence, it is possible to obtain diversified data sequences and label sequences based on the original data sequence and the original label sequence, thereby diversifying the sample data used to train the source signal recognition model and improving the recognition ability of the signal recognition model.

[0086] For further information, see Figure 6 , which shows a method for determining an original sequence, and the method may also include:

[0087] S610. Determine a source signal data sequence including the source signal data, a source tag sequence including the source tags, a target data sequence including the target data, and an identification tag sequence including the identification tags.

[0088] S620. Merge the source signal data sequence and the target data sequence to obtain the original data sequence.

[0089] S630. Merge the source tag sequence and the identification tag sequence to obtain the original tag sequence.

[0090] The source signal data corresponds one-to-one to the source label, and the target data corresponds one-to-one to the identification label, so that when the corresponding sequence is generated, the data and the label are also one-to-one corresponding; that is, the position of each data sub-item in the original data sequence in the original data sequence is the same as the position of the label corresponding to each data sub-item in the original label sequence.

[0091] Furthermore, the method for merging the source data sequence with the target data sequence may be to directly merge the source data sequence and the target data sequence, specifically, the target data sequence may be spliced ​​after the source data sequence, or the source data sequence may be spliced ​​after the target data sequence. The method for merging the source data sequence with the target data sequence may also be interspersed splicing, that is, inserting each data sub-item in the target data sequence into the source data sequence in a preset insertion order, or inserting the source data sequence into the target data sequence in a preset insertion order. The method for merging the source tag sequence with the identification tag sequence may adopt the corresponding method for merging the source data sequence with the target data sequence.

[0092] By merging the source signal data sequence and the target data sequence, as well as merging the source label sequence and the identification label sequence, the number of samples used for model training can be increased; by mixing the source domain data with the target domain data, the data distribution deviation between different data domains can be weakened, and model training based on the mixed data can improve the generalization ability of the model.

[0093] See also Figure 7, which shows a method for generating a mixed sequence based on weighting, which may include:

[0094] S710. Obtain a random weight base corresponding to the target position.

[0095] S720. Based on the random weight base, determine a first weight and a second weight; the sum of the first weight and the second weight is 1, and the difference between the first weight and the second weight is greater than or equal to a preset value.

[0096] S730. Based on the first weight and the second weight, weight the data sub-item at the target position in the original data sequence and the data sub-item at the target position in the rearranged data sequence respectively to obtain the mixed data sub-item at the target position in the mixed data sequence.

[0097] S740. Based on the first weight and the second weight, weight the label of the target position in the original label sequence and the label of the target position in the rearranged label sequence respectively to obtain a mixed label of the target position in the mixed label sequence.

[0098] S750. Generate the mixed data sequence based on the mixed data sub-items at each target position.

[0099] S760. Generate the mixed label sequence based on the mixed labels of each target position.

[0100] The random weight cardinality can be obtained based on a preset probability distribution; the preset probability distribution includes distribution parameters. In this embodiment, the distribution parameters can make the generated random weight cardinality approach 0 or 1, so that the difference between the first weight and the second weight determined based on the random weight cardinality is greater than or equal to a preset value, specifically, the first weight can be greater than the second weight, or the first weight can be less than the second weight.

[0101] Specifically, if for each target position i, based on the preset probability distribution, the corresponding generated random weight base is μ i , μ i is close to 0 or 1, the corresponding first weight can be λ i =max(μ i ,1-μ i ), the second weight can be 1-λ i ; or the second weight can be λ i =max(μ i ,1-μ i ), the first weight can be 1-λ i .

[0102] The first weight can be λ i =max(μi ,1-μ i ), the second weight can be 1-λ i For example, the mixed data sub-item at the target position in the mixed data sequence obtained after weighting is:

[0103] x d,i =λ i x i +(1-λ i )x′ i (3)

[0104] The mixed label of the target position in the mixed label sequence obtained after weighting is:

[0105] y d,i =λ i y i +(1-λ i )y′ i (4)

[0106] Among them, x i ∈M,x′ i ∈M′,y i ∈N,y′ i ∈N′.

[0107] Based on the above weighting method, mixed data sub-items and mixed labels are generated for each target position, and then corresponding mixed data sequences and mixed label sequences can be generated. In this embodiment, the difference between the first weight and the second weight is greater than or equal to the preset value, which can make the weight more biased towards one of the two weighted sequences, that is, one sequence is the main sequence and the other sequence is the auxiliary sequence, so that on the basis of the sequence weighted with a larger weight, a sequence weighted with a smaller weight is added, so that the disturbance to the sequence weighted with a larger weight is small, avoiding data distortion caused by weighting data with the same or similar weights.

[0108] When the source signal recognition model is trained based on the mixed data sequence and the mixed label sequence, the data sub-items in the mixed data sequence are used as signal data, and the labels at the corresponding positions in the mixed label sequence are used as labels corresponding to the signal data, that is, the mixed training samples can be recorded as (x d,i ,y d,i ), the corresponding loss function can be:

[0109]

[0110] For the trained target signal recognition model T, the output of the target signal recognition model can be judged by taking the threshold method; for example, for the input signal data x, when the corresponding output probability T(x)>τ, it can be judged that the signal corresponding to the input signal data is a pulsar signal; otherwise, it can be judged as a non-pulsar signal.

[0111] This embodiment also provides a signal recognition method. Figure 8 The execution subject of the method may be the above-mentioned signal recognition processing end, and the method may include:

[0112] S810. Obtain signal data to be identified.

[0113] S820. Input the signal data to be identified into a target signal identification model for signal identification, and obtain an identification result of the signal data to be identified.

[0114] The target signal recognition model is obtained by training the source signal recognition model based on source signal data carrying source labels and target data carrying identification labels; the identification label is obtained by performing signal recognition on the target data after information superposition processing based on the source signal recognition model; the target data is unlabeled signal data corresponding to the target task; the source signal recognition model is obtained by training the source signal data carrying source labels corresponding to the target task; the source labels include pulsar signal labels and non-pulsar signal labels.

[0115] The signal data to be identified may be signal data corresponding to the signal to be identified, and the signal data may be used to characterize signal feature information of the signal to be identified, and the signal feature information may include time domain feature information and frequency domain feature information. Specifically, the time domain feature information of the signal may be a phase-time diagram or a phase-frequency diagram of the signal, so that the signal data of the signal may be image data corresponding to the phase-time diagram and / or image data corresponding to the phase-frequency diagram.

[0116] When performing signal recognition, the data input into the target signal recognition model may be a floating-point matrix or vector corresponding to a phase-time diagram or a phase-frequency diagram, and the floating-point matrix or vector may be used to characterize pixel features of an image.

[0117] For the specific method of generating the target signal recognition model, please refer to the above content of this embodiment, which will not be repeated here.

[0118] Since the target signal recognition model trained by the above method has a strong signal recognition capability, the recognition accuracy of the signal data to be recognized can be improved by recognizing the signal data to be recognized through the target signal recognition model.

[0119] The target signal recognition model in the embodiment of the present application can be generated based on a machine learning model, and may include, for example, a convolutional network structure, a linear classifier, a multi-layer perceptron, and other network structures.

[0120] Based on this characteristic of the celestial body exploration scenario, this embodiment proposes a target signal recognition model training method. On the one hand, the pseudo-label algorithm is used to continuously expand the training label samples to improve the model's discrimination ability. On the other hand, by constructing a data mixing domain, the deviation of data distribution between different domains is weakened, thereby improving the generalization ability of the training model.

[0121] See also Fig. 9 , which shows a signal recognition model training device, which may include:

[0122] The information superposition processing module 910 is used to perform information superposition processing on the target data to obtain multiple processed data; the target data is the unlabeled signal data corresponding to the target task;

[0123] An identification label determination module 920 is used to perform signal identification on the plurality of processed data based on a source signal identification model to obtain an identification label corresponding to the target data; the source signal identification model is trained based on source signal data carrying source labels corresponding to the target task; the source labels include pulsar signal labels and non-pulsar signal labels;

[0124] The model training module 930 is used to train the source signal recognition model based on the source signal data carrying the source label and the target data carrying the recognition label to obtain a target signal recognition model corresponding to the target task.

[0125] Further, the target data includes a plurality of data sub-items;

[0126] The information superposition processing module 910 includes:

[0127] A random information generation module is used to generate multiple random information when performing information superposition processing on each data sub-item;

[0128] The data superposition module is used to superimpose the multiple random information onto each data sub-item to obtain multiple processed data corresponding to each data sub-item; the data features of each data sub-item are the same as the data features of the multiple processed data.

[0129] Furthermore, the identification tag determination module 920 includes:

[0130] A prediction sub-label determination module, used for inputting the plurality of processed data into the source signal recognition model to obtain prediction sub-labels corresponding to the plurality of processed data respectively;

[0131] The information fusion module is used to fuse the predicted sub-labels corresponding to the multiple processed data to obtain the identification label.

[0132] Furthermore, the model training module 930 includes:

[0133] an original sequence determination module, configured to determine an original data sequence and an original label sequence based on source signal data carrying the source label and target data carrying the identification label; the position of each data sub-item in the original data sequence is the same as the position of the label corresponding to each data sub-item in the original label sequence;

[0134] a reordering module, configured to reorder the original data sequence and the original label sequence to obtain a reordered data sequence and a reordered label sequence; the position of each data sub-item in the reordered data sequence is the same as the position of the label corresponding to each data sub-item in the reordered label sequence;

[0135] A mixed sequence determination module is used to weight the data sub-items at corresponding positions in the original data sequence and the rearranged data sequence to obtain a mixed data sequence, and to weight the labels at corresponding positions in the original label sequence and the rearranged label sequence to obtain a mixed label sequence; the weighting method for weighting the data sub-items at the target position is the same as the weighting method for weighting the labels at the target position;

[0136] The target signal recognition model determination module is used to train the source signal recognition model based on the mixed data sequence and the mixed label sequence to obtain the target signal recognition model.

[0137] Furthermore, the original sequence determination module includes:

[0138] A first determination module is used to determine a source signal data sequence including the source signal data, a source tag sequence including the source tags, a target data sequence including the target data, and an identification tag sequence including the identification tags;

[0139] An original data sequence determination module, used for merging the source signal data sequence and the target data sequence to obtain the original data sequence;

[0140] The original tag sequence determination module is used to merge the source tag sequence and the identification tag sequence to obtain the original tag sequence.

[0141] Furthermore, the mixed sequence determination module includes:

[0142] A random weight base acquisition module, used to acquire a random weight base corresponding to the target position;

[0143] A weight determination module, configured to determine a first weight and a second weight based on the random weight base; the sum of the first weight and the second weight is 1, and the difference between the first weight and the second weight is greater than or equal to a preset value;

[0144] A first weighting module is used to weight the data sub-item at the target position in the original data sequence and the data sub-item at the target position in the rearranged data sequence based on the first weight and the second weight, respectively, to obtain a mixed data sub-item at the target position in the mixed data sequence;

[0145] a second weighting module, configured to weight the label of the target position in the original label sequence and the label of the target position in the rearranged label sequence based on the first weight and the second weight, respectively, to obtain a mixed label of the target position in the mixed label sequence;

[0146] A mixed data sequence determination module, used to generate the mixed data sequence based on the mixed data sub-items of each target position;

[0147] The mixed label sequence determination module is used to generate the mixed label sequence based on the mixed labels of each target position.

[0148] See also Fig.10 , which shows a signal recognition device, comprising:

[0149] The signal data acquisition module 1010 is used to acquire the signal data to be identified;

[0150] The signal recognition module 1020 is used to input the signal data to be recognized into the target signal recognition model for signal recognition, and obtain the recognition result of the signal data to be recognized;

[0151] The target signal recognition model is obtained by training the source signal recognition model based on source signal data carrying source labels and target data carrying identification labels; the identification label is obtained by performing signal recognition on the target data after information superposition processing based on the source signal recognition model; the target data is unlabeled signal data corresponding to the target task; the source signal recognition model is obtained by training the source signal data carrying source labels corresponding to the target task.

[0152] The device provided in the above embodiment can execute the method provided in any embodiment of the present application, and has the corresponding functional modules and beneficial effects of executing the method. For technical details not described in detail in the above embodiment, please refer to the method provided in any embodiment of the present application.

[0153] This embodiment further provides a computer-readable storage medium, in which at least one instruction or at least one program is stored. The at least one instruction or at least one program is loaded by a processor and executed as any of the above methods of this embodiment.

[0154] According to one aspect of the present application, a computer program product or a computer program is provided, the computer program product or the computer program including computer instructions, the computer instructions being stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes any of the above methods of this embodiment.

[0155] Fig.11 is a block diagram of an electronic device for signal recognition model training or signal recognition according to an exemplary embodiment. The electronic device may be a server, and its internal structure diagram may be as shown in FIG. Fig.11 As shown. The electronic device includes a processor, a memory and a network interface connected via a system bus. The processor of the electronic device is used to provide computing and control capabilities. The memory of the electronic device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the electronic device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a signal recognition model training method or a signal recognition method is implemented.

[0156] Those skilled in the art will understand that Fig.11 The structure shown in the figure is merely a block diagram of a partial structure related to the scheme of the present disclosure, and does not constitute a limitation on the electronic device to which the scheme of the present disclosure is applied. The specific electronic device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0157] This specification provides method operation steps as described in the embodiments or flow charts, but more or fewer operation steps may be included based on conventional or non-creative labor. The steps and sequence listed in the embodiments are only one way of executing the sequence of many steps and do not represent the only execution sequence. When the actual system or interrupt product is executed, it can be executed in sequence or in parallel according to the method shown in the embodiments or the drawings (for example, in a parallel processor or multi-threaded processing environment).

[0158] The structure shown in this embodiment is only a partial structure related to the scheme of the present application, and does not constitute a limitation on the device to which the scheme of the present application is applied. The specific device may include more or fewer components than shown, or combine certain components, or have different arrangements of components. It should be understood that the method, device, etc. disclosed in this embodiment can be implemented in other ways. For example, the device embodiment described above is only schematic. For example, the division of the modules is only a division of a logical function. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or unit modules.

[0159] Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or the whole or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., various media that can store program codes.

[0160] Those skilled in the art may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in this specification can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0161] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A signal recognition model training method, characterized in that: include: Performing information superposition processing on the target data to obtain multiple processed data; the target data is unlabeled signal data corresponding to the target task; Based on a source signal recognition model, signal recognition is performed on the plurality of processed data to obtain an identification label corresponding to the target data; the source signal recognition model is trained based on source signal data carrying source labels corresponding to the target task; the source labels include pulsar signal labels and non-pulsar signal labels; Based on the source signal data carrying the source label and the target data carrying the identification label, an original data sequence and an original label sequence are determined; the position of each data sub-item in the original data sequence is the same as the position of the label corresponding to each data sub-item in the original label sequence; Rearranging the original data sequence and the original label sequence to obtain a rearranged data sequence and a rearranged label sequence; the position of each data sub-item in the rearranged data sequence is the same as the position of the label corresponding to each data sub-item in the rearranged label sequence; The data sub-items at corresponding positions in the original data sequence and the rearranged data sequence are weighted to obtain a mixed data sequence, and the labels at corresponding positions in the original label sequence and the rearranged label sequence are weighted to obtain a mixed label sequence; the weighting method for weighting the data sub-items at the target position is the same as the weighting method for weighting the labels at the target position; The source signal recognition model is trained based on the mixed data sequence and the mixed label sequence to obtain a target signal recognition model corresponding to the target task.

2. The method according to claim 1, characterized in that: The target data includes multiple data sub-items; The target data is subjected to information superposition processing to obtain a plurality of processed data, including: When performing information superposition processing on each data sub-item, multiple random information are generated; The multiple random information are respectively superimposed on each data sub-item to obtain multiple processed data corresponding to each data sub-item; the data features of each data sub-item are the same as the data features of the multiple processed data.

3. The method according to claim 1, characterized in that The performing signal recognition on the plurality of processed data based on the source signal recognition model to obtain an identification label corresponding to the target data includes: Inputting the plurality of processed data into the source signal recognition model to obtain prediction sub-labels corresponding to the plurality of processed data respectively; Information fusion is performed on the predicted sub-labels corresponding to the multiple processed data to obtain the identification label.

4. The method according to claim 1, characterized in that: The determining of the original data sequence and the original label sequence based on the source signal data carrying the source label and the target data carrying the identification label comprises: Determine a source signal data sequence including the source signal data, a source tag sequence including the source tags, a target data sequence including the target data, and an identification tag sequence including the identification tags; Merging the source signal data sequence and the target data sequence to obtain the original data sequence; The source tag sequence and the identification tag sequence are combined to obtain the original tag sequence.

5. The method according to claim 1, characterized in that The step of weighting the data sub-items at corresponding positions in the original data sequence and the rearranged data sequence to obtain a mixed data sequence, and weighting the labels at corresponding positions in the original label sequence and the rearranged label sequence to obtain a mixed label sequence includes: Obtaining a random weight base corresponding to the target position; Based on the random weight base, determining a first weight and a second weight; the sum of the first weight and the second weight is 1, and the difference between the first weight and the second weight is greater than or equal to a preset value; Based on the first weight and the second weight, weighting the data sub-item at the target position in the original data sequence and the data sub-item at the target position in the rearranged data sequence respectively, to obtain the mixed data sub-item at the target position in the mixed data sequence; Based on the first weight and the second weight, weighting the label of the target position in the original label sequence and the label of the target position in the rearranged label sequence respectively to obtain a mixed label of the target position in the mixed label sequence; generating the mixed data sequence based on the mixed data sub-items at each target position; Based on the mixed labels of each target position, the mixed label sequence is generated.

6. A signal recognition method, characterized in that: include: Acquire signal data to be identified; Inputting the signal data to be identified into a target signal identification model for signal identification, and obtaining an identification result of the signal data to be identified; The target signal recognition model is obtained by training the source signal recognition model based on the source signal data carrying the source label and the target data carrying the recognition label; The identification tag is obtained by performing signal identification on the target data after information superposition processing based on the source signal identification model; the target data is unlabeled signal data corresponding to the target task; the source signal identification model is obtained by training based on the source signal data carrying the source label corresponding to the target task; the source label includes a pulsar signal label and a non-pulsar signal label; The training of the source signal recognition model based on the source signal data carrying the source label and the target data carrying the identification label includes: determining an original data sequence and an original label sequence based on the source signal data carrying the source label and the target data carrying the identification label; The position of each data sub-item in the original data sequence is the same as the position of the label corresponding to each data sub-item in the original label sequence; the original data sequence and the original label sequence are reordered to obtain a rearranged data sequence and a rearranged label sequence; the position of each data sub-item in the rearranged data sequence is the same as the position of the label corresponding to each data sub-item in the rearranged label sequence; the data sub-items at corresponding positions in the original data sequence and the rearranged data sequence are weighted to obtain a mixed data sequence, and the labels at corresponding positions in the original label sequence and the rearranged label sequence are weighted to obtain a mixed label sequence; the weighting method for weighting the data sub-items at the target position is the same as the weighting method for weighting the label at the target position; the source signal recognition model is trained based on the mixed data sequence and the mixed label sequence to obtain the target signal recognition model.

7. A signal recognition model training device, characterized in that: include: An information superposition processing module is used to perform information superposition processing on target data to obtain multiple processed data; the target data is unlabeled signal data corresponding to the target task; An identification label determination module is used to perform signal identification on the plurality of processed data based on a source signal identification model to obtain an identification label corresponding to the target data; the source signal identification model is trained based on source signal data carrying source labels corresponding to the target task; the source labels include pulsar signal labels and non-pulsar signal labels; A model training module, used to train the source signal recognition model based on the source signal data carrying the source label and the target data carrying the recognition label to obtain a target signal recognition model corresponding to the target task; The model training module includes: an original sequence determination module, configured to determine an original data sequence and an original label sequence based on source signal data carrying the source label and target data carrying the identification label; the position of each data sub-item in the original data sequence is the same as the position of the label corresponding to each data sub-item in the original label sequence; a reordering module, configured to reorder the original data sequence and the original label sequence to obtain a reordered data sequence and a reordered label sequence; the position of each data sub-item in the reordered data sequence is the same as the position of the label corresponding to each data sub-item in the reordered label sequence; A mixed sequence determination module is used to weight the data sub-items at corresponding positions in the original data sequence and the rearranged data sequence to obtain a mixed data sequence, and to weight the labels at corresponding positions in the original label sequence and the rearranged label sequence to obtain a mixed label sequence; the weighting method for weighting the data sub-items at the target position is the same as the weighting method for weighting the labels at the target position; The target signal recognition model determination module is used to train the source signal recognition model based on the mixed data sequence and the mixed label sequence to obtain the target signal recognition model.

8. The device according to claim 7, characterized in that The target data includes multiple data sub-items; the information superposition processing module includes: A random information generation module is used to generate multiple random information when performing information superposition processing on each data sub-item; The data superposition module is used to superimpose the multiple random information onto each data sub-item to obtain multiple processed data corresponding to each data sub-item; the data features of each data sub-item are the same as the data features of the multiple processed data.

9. The device according to claim 7, characterized in that The identification tag determination module includes: A prediction sub-label determination module, used for inputting the plurality of processed data into the source signal recognition model to obtain prediction sub-labels corresponding to the plurality of processed data respectively; The information fusion module is used to fuse the predicted sub-labels corresponding to the multiple processed data to obtain the identification label.

10. The device according to claim 7, characterized in that The original sequence determination module comprises: A first determination module is used to determine a source signal data sequence including the source signal data, a source tag sequence including the source tags, a target data sequence including the target data, and an identification tag sequence including the identification tags; An original data sequence determination module, used for merging the source signal data sequence and the target data sequence to obtain the original data sequence; The original tag sequence determination module is used to merge the source tag sequence and the identification tag sequence to obtain the original tag sequence.

11. The device according to claim 7, characterized in that The mixed sequence determination module comprises: A random weight base acquisition module, used to acquire a random weight base corresponding to the target position; A weight determination module, configured to determine a first weight and a second weight based on the random weight base; the sum of the first weight and the second weight is 1, and the difference between the first weight and the second weight is greater than or equal to a preset value; A first weighting module is used to weight the data sub-item at the target position in the original data sequence and the data sub-item at the target position in the rearranged data sequence based on the first weight and the second weight, respectively, to obtain a mixed data sub-item at the target position in the mixed data sequence; a second weighting module, configured to weight the label of the target position in the original label sequence and the label of the target position in the rearranged label sequence based on the first weight and the second weight, respectively, to obtain a mixed label of the target position in the mixed label sequence; A mixed data sequence determination module, used to generate the mixed data sequence based on the mixed data sub-items of each target position; The mixed label sequence determination module is used to generate the mixed label sequence based on the mixed labels of each target position.

12. A signal recognition device, characterized in that: include: A signal data acquisition module to be identified, used to acquire signal data to be identified; A signal recognition module, used for inputting the signal data to be recognized into a target signal recognition model for signal recognition, and obtaining a recognition result of the signal data to be recognized; The target signal recognition model is obtained by training the source signal recognition model based on the source signal data carrying the source label and the target data carrying the recognition label; The identification label is obtained by performing signal identification on the target data after information superposition processing based on the source signal identification model; the target data is unlabeled signal data corresponding to the target task; the source signal identification model is trained based on the source signal data carrying the source label corresponding to the target task to obtain the source signal data carrying the source label and the target data carrying the identification label, and the source signal identification model is trained, including: determining the original data sequence and the original label sequence based on the source signal data carrying the source label and the target data carrying the identification label; The position of each data sub-item in the original data sequence is the same as the position of the label corresponding to each data sub-item in the original label sequence; the original data sequence and the original label sequence are reordered to obtain a rearranged data sequence and a rearranged label sequence; the position of each data sub-item in the rearranged data sequence is the same as the position of the label corresponding to each data sub-item in the rearranged label sequence; the data sub-items at corresponding positions in the original data sequence and the rearranged data sequence are weighted to obtain a mixed data sequence, and the labels at corresponding positions in the original label sequence and the rearranged label sequence are weighted to obtain a mixed label sequence; the weighting method for weighting the data sub-items at the target position is the same as the weighting method for weighting the label at the target position; the source signal recognition model is trained based on the mixed data sequence and the mixed label sequence to obtain the target signal recognition model.

13. An electronic device, characterized in that: The device includes a processor and a memory, wherein the memory stores at least one instruction or at least one program, and the at least one instruction or the at least one program is loaded and executed by the processor to implement the signal recognition model training method as described in any one of claims 1 to 5, or the signal recognition method as described in claim 6.

14. A computer storage medium, characterized in that: The storage medium stores at least one instruction or at least one program, and the at least one instruction or the at least one program is loaded by the processor and executed by the signal recognition model training method as described in any one of claims 1 to 5, or the signal recognition method as described in claim 6.

Images