Model construction method and estimation method for estimating number of aliasing signal sources

Abstract

The invention provides a model construction method and an estimation method for estimating the number of aliasing signal sources. The model construction method comprises the following steps: constructing a double-path feature complementary fusion network; in the two-way feature complementary fusion network, the aggregation residual network is used as a main network path, and each residual network branch is configured as a dense network structure; constructing an attention coordination mechanism module which comprises a channel attention module and a space attention module; and forming a model based on the two-way feature complementary fusion network and the attention coordinating mechanism, wherein the model is used for estimating the number of aliasing signal sources. The estimation method comprises the following steps: acquiring IQ sequence data of an input signal, and taking the IQ sequence data as model input; and estimating the number of the aliasing signal sources according to a pre-constructed model based on the double-path feature complementary fusion network and the attention coordination mechanism to obtain the number of the aliasing signal sources. According to the invention, the accuracy of single-channel aliasing signal source number estimation under the condition of low signal-to-noise ratio can be effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of signal processing technology, and in particular to a model building method and an estimation method for estimating the number of aliasing signal sources. Background Art

[0002] Source number estimation refers to the technology of determining the number of signal sources that exist simultaneously. This is crucial to many aspects of signal processing, such as blind signal separation, signal source location and channel estimation. It plays an extremely important role in many fields such as electronic reconnaissance, medical diagnosis and image processing.

[0003] Especially in the case of single-channel processing of aliased signals, that is, when only one receiver receives all the aliased signals, accurate estimation of the number of signal sources is a prerequisite for the signal processing algorithm. When the estimation of the number of signal sources is inaccurate, it will seriously affect the accuracy of the subsequent signal processing algorithm.

[0004] Among the related technologies, deep learning technology has been widely used in the estimation of the number of aliasing signal sources due to its good feature extraction ability. However, the convolution operations, encapsulation functions and basic principles in the deep learning algorithm are mostly based on real numbers, without considering the complex data correlation characteristics, and only use a single feature as the basis for estimating the number of aliasing sources, resulting in incomplete feature information. Secondly, because it relies on a single feature and a specific network architecture, when there are high-modulation signals (such as PSK and QAM) in the aliasing source signal, estimation confusion is prone to occur due to the shortened Euclidean distance, resulting in low estimation accuracy. Summary of the invention

[0005] The present invention provides a model building method and an estimation method for estimating the number of aliased signal sources, which solve the problem of low accuracy of single-channel aliased signal source estimation under low signal-to-noise ratio.

[0006] In order to achieve the above objectives, this application adopts the following technical solutions:

[0007] In a first aspect, a model building method for estimating the number of aliasing signal sources is provided, comprising:

[0008] Constructing a dual-path feature complementary fusion network; in the dual-path feature complementary fusion network, the aggregated residual network is used as the main network path, and each residual network branch is configured as a dense network structure;

[0009] Constructing a coordinated attention mechanism module; the coordinated attention mechanism module includes a channel attention module and a spatial attention module, the channel attention module serves as the input part of the coordinated attention mechanism module, and by establishing a jump connection, the output features of the channel attention module serve as the input features of the spatial attention module, and the spatial attention module serves as the output part of the coordinated attention mechanism module; the channel attention module and the spatial attention module are fused in a channel weighted manner to form the coordinated attention mechanism module;

[0010] Based on the two-way feature complementary fusion network and the coordinated attention mechanism module, a model based on the two-way feature complementary fusion network and the coordinated attention mechanism is formed for estimating the number of aliasing signal sources.

[0011] In a first possible implementation manner of the first aspect, the dual-path feature complementary fusion network includes: a convolutional layer, a dual-path module, a fusion transition module, a dense connection module, and a final convolutional layer;

[0012] The initial convolution layer is a 3x3 convolution layer, there are three dual-path modules, the three dual-path modules are arranged in sequence, each of the dual-path modules is composed of a 3x3 and two 1x1 convolution layers, the 3x3 convolution layer is located between two 1x1 convolution layers, any two adjacent dual-path modules are connected through the fusion transition module, there are three final convolution layers, and the densely connected module connects the third dual-path module to the three final convolution layers respectively.

[0013] Based on the first possible implementation of the first aspect, in a second possible implementation of the first aspect, the output y of the dual-path feature complementary fusion network t It is expressed by the following formula:

[0014]

[0015] Among them, x m represents the mth dense network branch, y m represents the mth residual network branch, v t represents the feature learning function, r m Indicates the intermediate variable during iteration, h m Represents the output value of the last transformation function, g m Represents the transformation function used to generate the state of the target.

[0016] In a third possible implementation manner of the first aspect, in the hyperparameter design of the dual-path feature complementary fusion network, the number of residual network channels is greater than the number of dense network channels.

[0017] In a fourth possible implementation manner of the first aspect, the channel attention module is specifically used to extract position perception information, including:

[0018] Use the maximum pooling layer and the average pooling layer to perform pooling operations on the feature input respectively;

[0019] Use two fully connected layers to extract features from the pooled output;

[0020] The feature outputs after the activation function mapping are used as input values ​​and input value weights respectively, and they are reconnected and then inner product calculation is performed as input features of the spatial attention module.

[0021] In a fifth possible implementation manner of the first aspect, the inter-attention module is specifically used to extract spatial features of different dimensions and perform channel weighting, including:

[0022] Based on the feature output of the channel attention module, a maximum pooling layer and an average pooling layer are used to perform a pooling operation in a two-dimensional plane;

[0023] The features after maximum pooling and average pooling are concatenated through the concatenation layer;

[0024] The convolution layer is used to perform a convolution operation on the concatenated features to reduce the dimension of the features, and the activation function is used to normalize the features to obtain the weight dimension;

[0025] The weight dimension is multiplied by the output of the convolution operation to obtain the final feature.

[0026] Based on any possible implementation manner of the first aspect, in a sixth possible implementation manner of the first aspect, IQ sequence data of the input signal is used as a model input.

[0027] In a second aspect, a method for estimating the number of aliased signal sources is provided, which uses a model based on a dual-path feature complementary fusion network and a coordinated attention mechanism as in the first aspect to estimate the number of aliased signal sources, including:

[0028] Acquire IQ sequence data of the input signal, and use the IQ sequence data as model input;

[0029] According to the pre-built model based on the dual-path feature complementary fusion network and the coordinated attention mechanism, the number of aliasing signal sources is estimated to obtain the number of aliasing signal sources.

[0030] In a third aspect, a device for estimating the number of aliased signal sources is provided, which uses the model based on a dual-path feature complementary fusion network and a coordinated attention mechanism as in the first aspect to estimate the number of aliased signal sources, including: a data acquisition module, used to acquire IQ sequence data of an input signal, and use the IQ sequence data as a model input;

[0031] The model module is used to estimate the number of aliasing signal sources according to a pre-built model based on a two-way feature complementary fusion network and a coordinated attention mechanism to obtain the number of aliasing signal sources.

[0032] In a fourth aspect, an electronic device is provided, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program implements the steps of the method described in the first aspect when executed by the processor.

[0033] In a fifth aspect, a readable storage medium is provided, on which a program or instruction is stored, and when the program or instruction is executed by a processor, the steps of the method described in the first aspect are implemented.

[0034] Beneficial effects:

[0035] The model based on the dual-path feature complementary fusion network and coordinated attention mechanism of the present application, by introducing the dual-path feature complementary fusion network, enables the model to more effectively extract and utilize the real and imaginary spatial features of the signal, and improves the estimation accuracy of the number of aliased signal sources; the introduction of the coordinated attention mechanism integrates the attention information of different dimensions of the signal features after convolution, effectively enhancing the anti-noise performance and robustness of the model, while not adding too much computational overhead on the basis of the original network. Compared with the existing deep learning estimation algorithm, the model of the present application deeply mines the aliased signal information, optimizes feature reuse, to enhance the feature extraction ability of the network, and introduces the coordinated attention mechanism dynamic weight allocation, so that the network can continuously and dynamically weight and strengthen cognitive ability through self-learning, autonomously identify effective information and unimportant information, so as to have a higher estimation accuracy. It effectively solves the problem of low accuracy of single-channel aliased electromagnetic signal source number estimation under low signal-to-noise ratio. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 A schematic flow chart of a model building method for estimating the number of aliasing signal sources provided in an embodiment of the present application;

[0037] Figure 2 A schematic diagram of the structure of an aggregated residual network provided in an embodiment of the present application;

[0038] Figure 3 A schematic diagram of a dense network structure provided in an embodiment of the present application;

[0039] Figure 4 A schematic diagram of a recursive structure of a residual unit provided in an embodiment of the present application;

[0040] Figure 5A schematic diagram of a recursive structure of a dense network provided in an embodiment of the present application;

[0041] Figure 6 A schematic diagram of the structure of a dual-path feature complementary fusion network provided in an embodiment of the present application;

[0042] Figure 7 A schematic diagram of the structure of a channel attention module provided in an embodiment of the present application;

[0043] Figure 8 A schematic diagram of the structure of a spatial attention module provided in an embodiment of the present application;

[0044] Fig. 9 A schematic flowchart of a method for estimating the number of aliased signal sources provided in an embodiment of the present application;

[0045] Fig.10 A structural block diagram of a device for estimating the number of aliased signal sources provided in an embodiment of the present application;

[0046] Fig.11 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0047] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined purpose, the technical solutions in the embodiments of the present application are clearly described. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field belong to the scope of protection of the present application.

[0048] The terms "first", "second", etc. in the specification and claims of the present application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the terms used in this way are interchangeable where appropriate, so that the embodiments of the present application can be implemented in an order other than those illustrated or described herein, and the objects distinguished by "first", "second", etc. are generally of one type, and the number of objects is not limited. For example, the first object can be one or more. In addition, "and / or" in the specification and claims represents at least one of the connected objects, and the character " / " generally indicates that the objects associated with each other are in an "or" relationship.

[0049] The description of the method flow in the specification of the present application and the steps of the flowchart in the drawings of the present specification do not have to be strictly executed according to the step numbers, and the method steps can be executed in a different order. Moreover, some steps can be omitted, multiple steps can be combined into one step, and / or one step can be decomposed into multiple steps.

[0050] The following is a detailed description of the model building method and estimation method for estimating the number of aliasing signal sources provided in the embodiments of the present application in conjunction with the accompanying drawings and preferred embodiments.

[0051] First, the model construction method for estimating the number of aliasing signal sources and the application scenarios of the estimation method according to the embodiments of the present application are described in detail.

[0052] According to the related art known to the inventor, there are mainly two designs of network structures for estimating the number of aliased signal sources based on deep learning:

[0053] The first is the convolutional neural network. Shrinivas Mahajan et al. introduced a dropout layer in the network architecture. Without affecting the estimation result of the number of aliasing sources, the training time was shortened by discarding some neurons. In terms of improving the structure of the convolutional neural network, Zhu Yanmin et al. proposed a one-dimensional convolutional residual network structure. When the number of aliasing sources is less than or equal to 7, the estimation accuracy is improved by 8% compared with CNN.

[0054] The second is the recurrent neural network. Bernd Edler et al. used CLDNN in the study of the estimation of the number of aliasing sources. The analysis of the result error and network performance showed that the front CNN layer effectively improved the convergence speed of LSTM.

[0055] In addition, researchers have also proposed many new composite network architectures and unsupervised learning theories, such as hybrid frameworks of classic networks such as AlexNet, VGGNet, GoogleNet, ResNet, SENet, etc., to improve the feature extraction capabilities of deep learning networks.

[0056] The above-mentioned aliasing source number estimation algorithm based on deep learning technology has the following major defects: First, the convolution operation, encapsulation function and basic principles in the traditional deep learning algorithm are mostly based on real numbers, and the complex data correlation characteristics are not considered. Although some technologies use empirical mode decomposition and Hilbert-Huang transform methods to extract phase features to solve the above problems, the algorithm complexity will increase; and only a single feature is used as the basis for estimating the number of aliasing sources, and there is a problem of incomplete feature information. Secondly, when there are high-order PSK and QAM signals in the aliasing source signal, the Euclidean distance becomes shorter, which is prone to estimation confusion, resulting in low estimation accuracy.

[0057] That is to say, when the existing deep learning network estimates the number of sources of aliased signals under low signal-to-noise ratio, the estimation accuracy is easily limited by the complex channel environment, and a single network structure cannot effectively estimate the number of sources of aliased modulated signals. In addition, when there are too many sources in the aliased source signal, adjacent features are seriously confused, which has a great impact on the accuracy of model estimation.

[0058] To address the above problems, related technologies use a method of performing special processing on the aliased modulated signal to obtain the frequency or phase characteristics of the signal and pass it into the network. This method has a large preprocessing and training complexity, the algorithm complexity becomes higher, and data features may be lost.

[0059] Based on this, the present application provides a model construction method and an estimation method for estimating the number of aliasing signal sources, introduces a dual-path convolutional network, integrates the advantages of residual convolution unit feature splicing and dense network generation of new features, utilizes the coordinated attention mechanism to dynamically weighted update the feature channels, and makes full use of the real and imaginary part feature information of the signal to achieve accurate estimation of the number of sources for aliasing signals under low signal-to-noise ratio.

[0060] See also Figure 1 , the embodiment of the present application provides a model construction method for estimating the number of aliasing signal sources, such as Figure 1 As shown, the model building method of the embodiment of the present application includes the following steps:

[0061] Step S1, constructing a two-way feature complementary fusion network; in the two-way feature complementary fusion network, the aggregated residual network is used as the main network path, and each residual network branch is configured as a dense network structure.

[0062] This step is to form a complementary fusion dual-path convolutional network deep learning framework by designing an aggregated residual network and a dense network. The main idea of ​​this framework is a deep learning framework that combines an aggregated residual network and a dense network. The main structure of the complementary dual-path convolutional network is based on the residual network. The aggregated residual network uses jump connections to retain input features, solving the network performance degradation and parameter fitting problems caused by too many network layers; each residual network branch uses a densely connected structure, that is, the current layer is densely connected to all previous layers, and each layer directly receives feature maps from all previous layers as input. By integrating the low redundancy of the aggregated residual network and the ability of the dense network to generate new features, the complementary fusion of dual-path convolutions forms a dual-path convolutional network. The main ideas of this framework are jump connections and feature splicing, and the structure uses dual-path convolutions, integrating the advantages of the low redundancy of the aggregated residual network and the generation of new features and information flow of the dense network.

[0063] Specifically, the dual-path feature complementary fusion network includes: a convolutional layer, a dual-path module, a fusion transition module, a dense connection module, and a final convolutional layer; the initial convolutional layer is a 3x3 convolutional layer, there are three dual-path modules, and the three dual-path modules are arranged in sequence. Each dual-path module is composed of a 3x3 and two 1x1 convolutional layers. The 3x3 convolutional layer is located between two 1x1 convolutional layers. Any two adjacent dual-path modules are connected by a fusion transition module. There are three final convolutional layers, and the dense connection module connects the third dual-path module to the three final convolutional layers respectively.

[0064] Further, the structure of the aggregated residual network can be found in Figure 2 ,In the aggregated residual network, the convolution operation of the intermediate layer is ,configured as grouped convolution, which means that the input features are ,divided into several groups, and each group is ,convolutionally operated independently.

[0065] The structure of dense networks can be found in Figure 3 , the dense network structure is similar to the high-order recurrent neural network, and its hidden output at the mth moment is shown in formula (1-1):

[0066]

[0067] Among them, y 1 =x 1 is the input feature of the dense network structure, is the network feature extraction function, which extracts the hidden features of the previous m-1 moments, w m Represents the feature mapping function, which maps the features of the previous m-1 moments to the hidden state of the current moment. From formula (1-1), it can be seen that for the feature output of each moment m, the dense network has independent mapping parameters. Unlike the high-order recurrent neural network, the high-order recurrent neural network unit samples the weight sharing strategy, and the mapping parameters at different moments are the same. The dense network can be regarded as a higher-order recurrent neural network unit. In addition, the densely connected network uses the skip connection in the residual unit, which is a special structure of the residual unit. The input features of the dense network structure can be expressed by formula (1-2):

[0068]

[0069] y m =g m (r m ) (1-3)

[0070] Substituting formula (1-3) into formula (1-2), formula (1-2) can be simplified to formula (1-4) and formula (1-5):

[0071]

[0072]

[0073] At any time equal When , the dense network structure and residual convolution can be expressed as a recursive form, see Figure, Figure 4-5 The recursive structure of the residual unit and dense network.

[0074] Based on this, the above designed aggregate residual network is combined with the dense network, and the dual-path feature complementary fusion network outputs y t It can be deduced from formula (1-6):

[0075]

[0076] Among them, x m represents the mth dense network branch, y m represents the mth residual network branch, v t represents the feature learning function, r m Indicates the intermediate variable during iteration, h m Represents the output value of the last transformation function, g m Represents the transformation function used to generate the state of the target.

[0077] See also Figure 6 When performing a 1×1 convolution operation, the residual network and the dense network share a parameter. When performing a 3×3 convolution operation, group convolution is used to improve network performance.

[0078] In addition, in the hyperparameter design of the dual-path feature complementary fusion network, in order to prevent the increase in the number of layers from causing excessive consumption of video memory, the number of residual network channels must be greater than the number of dense network channels.

[0079] Based on this, a two-way feature complementary fusion network is constructed in the model, and the low redundancy of the residual network and the dense network feature generation characteristics are utilized to improve the model estimation performance.

[0080] Step S2, constructing a coordinated attention mechanism module; the coordinated attention mechanism module includes a channel attention module and a spatial attention module, the channel attention module serves as the input part of the coordinated attention mechanism module, and by establishing a jump connection, the output features of the channel attention module serve as the input features of the spatial attention module, and the spatial attention module serves as the output part of the coordinated attention mechanism module; the channel attention module and the spatial attention module are fused in a channel-weighted manner to form a coordinated attention mechanism module.

[0081] In this step, the channel attention module and the spatial attention module are used in the input part and the output part respectively to fuse and design the attention mechanism module, and the original features of the input data are retained by establishing short connections. Specifically, first, the channel attention module performs average pooling operations on the input features in the horizontal and vertical directions to extract position perception information; then, the spatial attention module performs encoding conversion on them respectively to extract spatial features of different dimensions; finally, in the spatial attention module, the channels are fused by weighted methods to form a coordinated attention mechanism module. Specifically:

[0082] like Figure 7 As shown, the channel attention module is specifically used to extract position-aware information, including:

[0083] Step S211, using the maximum pooling layer and the average pooling layer to perform pooling operations on the feature input respectively.

[0084] Step S212, using two fully connected layers to perform feature extraction on the pooled output.

[0085] In step S213, the feature output after the activation function mapping is used as the input value and the input value weight respectively, and they are reconnected and then the inner product is calculated as the input feature of the spatial attention module.

[0086] like Figure 8 As shown in the figure, the spatial attention module is specifically used to extract spatial features of different dimensions and perform channel weighting, including:

[0087] Step S221, based on the feature output of the channel attention module, a pooling operation is performed on a two-dimensional plane using a maximum pooling layer and an average pooling layer.

[0088] Step S222, the features after maximum pooling and average pooling are concatenated through a concatenation layer.

[0089] In step S223, a convolution operation is performed on the concatenated features to reduce the dimension of the features using a convolution layer, and the features are normalized using an activation function to obtain a weight dimension.

[0090] Step S224, multiply the weight dimension and the output of the convolution operation to obtain the final feature.

[0091] Step S3, based on the dual-path feature complementary fusion network and coordinated attention mechanism module, a model based on the dual-path feature complementary fusion network and coordinated attention mechanism is formed for estimating the number of aliasing signal sources; wherein the IQ sequence data of the input signal is used as the model input.

[0092] Based on the above scheme, the model based on the dual-path feature complementary fusion network and coordinated attention mechanism of the present application uses IQ data as the estimation feature. Compared with the signal time domain waveform, constellation diagram and spectrum diagram, IQ data retains more effective features of the electromagnetic spectrum signal; by introducing the dual-path feature complementary fusion network, the model can more effectively extract and utilize the real and imaginary spatial features of the signal, and improve the estimation accuracy of the number of aliased signal sources; the introduction of the coordinated attention mechanism integrates the attention information of different dimensions of the IQ signal features after convolution, effectively enhancing the anti-noise performance and robustness of the model, while not adding too much computational overhead to the original network. Compared with the existing deep learning estimation algorithm, the model of the present application deeply mines the aliased signal information, optimizes feature reuse, so as to enhance the feature extraction ability of the network, and introduces the dynamic weight allocation of the coordinated attention mechanism, so that the network can continuously and dynamically weight and strengthen cognitive ability through self-learning, and autonomously identify effective information and unimportant information, so as to have a higher estimation accuracy.

[0093] See also Fig. 9 , the embodiment of the present application provides a method for estimating the number of aliasing signal sources, such as Fig. 9 As shown, the estimation method of the embodiment of the present application adopts the model based on the dual-path feature complementary fusion network and the coordinated attention mechanism of the above embodiment to estimate the number of aliasing signal sources, including the following steps:

[0094] Step A1, obtaining IQ sequence data of an input signal, and using the IQ sequence data as a model input.

[0095] Step A2, according to the pre-built model based on the two-way feature complementary fusion network and the coordinated attention mechanism, the number of aliasing signal sources is estimated to obtain the number of aliasing signal sources.

[0096] Based on the above scheme, the aliasing signal source number estimation method of the present application effectively solves the problem of low accuracy of single-channel aliasing electromagnetic signal source number estimation under low signal-to-noise ratio. Other effects are the same as the steps and processes of the above embodiments, and can achieve the same technical effects. To avoid repetition, they are not repeated here.

[0097] See also Fig.10 Corresponding to the above-mentioned embodiment of the method for estimating the number of aliased signal sources, an embodiment of the present application provides a device for estimating the number of aliased signal sources, which uses the model based on the dual-path feature complementary fusion network and the coordinated attention mechanism of the above-mentioned embodiment to estimate the number of aliased signal sources, including:

[0098] The data acquisition module 1001 is used to acquire IQ sequence data of an input signal and use the IQ sequence data as a model input;

[0099] The model module 1002 is used to estimate the number of aliasing signal sources according to a pre-built model based on a two-way feature complementary fusion network and a coordinated attention mechanism to obtain the number of aliasing signal sources.

[0100] The above-mentioned aliasing signal source number estimation device implements the steps and various processes of the above-mentioned aliasing signal source number estimation method embodiment and can achieve the same technical effect. To avoid repetition, they are not described again here.

[0101] See also Fig.11 Corresponding to the above-mentioned embodiment of the method for estimating the number of aliased signal sources, an embodiment of the present application provides an electronic device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, the steps of the above-mentioned embodiment of the method for estimating the number of aliased signal sources and the various processes of the embodiment are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0102] The memory 1009 can be used to store software programs and various data. The memory 1009 may mainly include a first storage area for storing programs or instructions and a second storage area for storing data, wherein the first storage area may store an operating system, an application program or instructions required for at least one function (such as a sound playback function, an image playback function, etc.), etc. In addition, the memory 1009 may include a volatile memory or a non-volatile memory, or the memory 1009 may include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), a static random access memory (SRAM), a dynamic random access memory (DRAM), a synchronous dynamic random access memory (SDRAM), a double data rate synchronous dynamic random access memory (DDRSDRAM), an enhanced synchronous dynamic random access memory (ESDRAM), a synchronous link dynamic random access memory (SLDRAM) and a direct memory bus random access memory (DRRAM). The memory 1009 in the embodiment of the present application includes but is not limited to these and any other suitable types of memory.

[0103] The processor 1010 may include one or more processing units; optionally, the processor 1010 integrates an application processor and a modem processor, wherein the application processor mainly processes operations related to an operating system, a user interface, and application programs, and the modem processor mainly processes wireless communication signals, such as a baseband processor. It is understandable that the modem processor may not be integrated into the processor 1010.

[0104] Corresponding to the above-mentioned embodiment of the method for estimating the number of aliasing signal sources, the embodiment of the present application also provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the above-mentioned embodiment of the method for estimating the number of aliasing signal sources and the various processes of the embodiment are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0105] The processor is the processor in the electronic device described in the above embodiment of the present application. The readable storage medium includes a computer readable storage medium, such as a computer read-only memory ROM, a random access memory RAM, a magnetic disk or an optical disk.

[0106] It should be noted that, in this article, the terms "comprise", "include" or any other variant thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise one..." do not exclude the presence of other identical elements in the process, method, article or device including the element. In addition, it should be pointed out that the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in reverse order according to the functions involved, for example, the described method may be performed in an order different from that described, and various steps may also be added, omitted, or combined. In addition, the features described with reference to certain examples may be combined in other examples.

[0107] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a computer software product, which is stored in a storage medium (such as ROM / RAM, a disk, or an optical disk), and includes a number of instructions for a terminal (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present application.

[0108] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also make many forms without departing from the purpose of the present application and the scope of protection of the claims, all of which are within the protection of the present application.

Claims

1. A model building method for estimating the number of aliasing signal sources, characterized in that: include: Constructing a dual-path feature complementary fusion network; in the dual-path feature complementary fusion network, the aggregated residual network is used as the main network path, and each residual network branch is configured as a dense network structure; Constructing a coordinated attention mechanism module; the coordinated attention mechanism module includes a channel attention module and a spatial attention module, the channel attention module serves as the input part of the coordinated attention mechanism module, and by establishing a jump connection, the output features of the channel attention module serve as the input features of the spatial attention module, and the spatial attention module serves as the output part of the coordinated attention mechanism module; the channel attention module and the spatial attention module are fused in a channel weighted manner to form the coordinated attention mechanism module; Based on the two-way feature complementary fusion network and the coordinated attention mechanism module, a model based on the two-way feature complementary fusion network and the coordinated attention mechanism is formed for estimating the number of aliasing signal sources.

2. The model building method for estimating the number of aliasing signal sources according to claim 1, characterized in that: The dual-path feature complementary fusion network includes: a convolutional layer, a dual-path module, a fusion transition module, a dense connection module, and a final convolutional layer; The initial convolution layer is a 3x3 convolution layer, there are three dual-path modules, the three dual-path modules are arranged in sequence, each of the dual-path modules is composed of a 3x3 and two 1x1 convolution layers, the 3x3 convolution layer is located between two 1x1 convolution layers, any two adjacent dual-path modules are connected through the fusion transition module, there are three final convolution layers, and the densely connected module connects the third dual-path module to the three final convolution layers respectively.

3. The model building method for estimating the number of aliasing signal sources according to claim 2, characterized in that: The output y of the dual-path feature complementary fusion network t It is expressed by the following formula: Among them, x m represents the mth dense network branch, y m represents the mth residual network branch, v t represents the feature learning function, r m Indicates the intermediate variable during iteration, h m Represents the output value of the last transformation function, g m Represents the transformation function used to generate the state of the target.

4. The model building method for estimating the number of aliasing signal sources according to claim 1, characterized in that: In the hyperparameter design of the dual-path feature complementary fusion network, the number of channels of the residual network is greater than that of the dense network.

5. The model building method for estimating the number of aliasing signal sources according to claim 1, characterized in that: The channel attention module is specifically used to extract position perception information, including: Use the maximum pooling layer and the average pooling layer to perform pooling operations on the feature input respectively; Use two fully connected layers to extract features from the pooled output; The feature outputs after the activation function mapping are used as input values ​​and input value weights respectively, and they are reconnected and then inner product calculation is performed as input features of the spatial attention module.

6. The model building method for estimating the number of aliasing signal sources according to claim 1, characterized in that: The inter-attention module is specifically used to extract spatial features of different dimensions and perform channel weighting, including: Based on the feature output of the channel attention module, a maximum pooling layer and an average pooling layer are used to perform a pooling operation in a two-dimensional plane; The features after maximum pooling and average pooling are concatenated through the concatenation layer; The convolution layer is used to perform a convolution operation on the concatenated features to reduce the dimension of the features, and the activation function is used to normalize the features to obtain the weight dimension; The weight dimension is multiplied by the output of the convolution operation to obtain the final feature.

7. The model building method for estimating the number of aliasing signal sources according to any one of claims 1 to 6, characterized in that: The IQ sequence data of the input signal is used as the model input.

8. A method for estimating the number of aliased signal sources, characterized in that: The method of estimating the number of aliased signal sources by using a model based on a dual-path feature complementary fusion network and a coordinated attention mechanism as claimed in any one of claims 1 to 7 comprises: Acquire IQ sequence data of the input signal, and use the IQ sequence data as model input; According to the pre-built model based on the dual-path feature complementary fusion network and the coordinated attention mechanism, the number of aliasing signal sources is estimated to obtain the number of aliasing signal sources.

9. A device for estimating the number of aliased signal sources, characterized in that: The method of estimating the number of aliased signal sources by using a model based on a dual-path feature complementary fusion network and a coordinated attention mechanism as claimed in any one of claims 1 to 7 comprises: A data acquisition module, used for acquiring IQ sequence data of an input signal, and using the IQ sequence data as a model input; The model module is used to estimate the number of aliasing signal sources according to a pre-built model based on a two-way feature complementary fusion network and a coordinated attention mechanism to obtain the number of aliasing signal sources.

10. An electronic device, characterized in that: The electronic device comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program implements the steps of the method according to any one of claims 1 to 7 when executed by the processor.

Images